1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "logging/logStream.hpp"
37 #include "memory/allocation.inline.hpp"
38 #include "memory/filemap.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_posix.inline.hpp"
42 #include "os_share_linux.hpp"
43 #include "osContainer_linux.hpp"
44 #include "prims/jniFastGetField.hpp"
45 #include "prims/jvm_misc.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/atomic.hpp"
48 #include "runtime/extendedPC.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/interfaceSupport.inline.hpp"
51 #include "runtime/init.hpp"
52 #include "runtime/java.hpp"
53 #include "runtime/javaCalls.hpp"
54 #include "runtime/mutexLocker.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/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
available_memory()181 julong os::available_memory() {
182 return Linux::available_memory();
183 }
184
available_memory()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
physical_memory()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
next_line(FILE * f)233 static void next_line(FILE *f) {
234 int c;
235 do {
236 c = fgetc(f);
237 } while (c != '\n' && c != EOF);
238 }
239
get_tick_information(CPUPerfTicks * pticks,int which_logical_cpu)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
have_special_privileges()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.
gettid()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
initialize_system_info()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
init_system_properties_values()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
breakpoint()500 void os::breakpoint() {
501 BREAKPOINT;
502 }
503
breakpoint()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
signal_sets_init()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.)
unblocked_signals()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.
vm_signals()569 sigset_t* os::Linux::vm_signals() {
570 assert(signal_sets_initialized, "Not initialized");
571 return &vm_sigs;
572 }
573
hotspot_sigmask(Thread * thread)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
libpthread_init()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
_expand_stack_to(address bottom)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
expand_stack_to(address bottom)713 void os::Linux::expand_stack_to(address bottom) {
714 _expand_stack_to(bottom);
715 }
716
manually_expand_stack(JavaThread * t,address addr)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 assert(t->stack_base() != NULL, "stack_base was not initialized");
721
722 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
723 sigset_t mask_all, old_sigset;
724 sigfillset(&mask_all);
725 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
726 _expand_stack_to(addr);
727 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
728 return true;
729 }
730 return false;
731 }
732
733 //////////////////////////////////////////////////////////////////////////////
734 // create new thread
735
736 // Thread start routine for all newly created threads
thread_native_entry(Thread * thread)737 static void *thread_native_entry(Thread *thread) {
738
739 thread->record_stack_base_and_size();
740
741 // Try to randomize the cache line index of hot stack frames.
742 // This helps when threads of the same stack traces evict each other's
743 // cache lines. The threads can be either from the same JVM instance, or
744 // from different JVM instances. The benefit is especially true for
745 // processors with hyperthreading technology.
746 static int counter = 0;
747 int pid = os::current_process_id();
748 alloca(((pid ^ counter++) & 7) * 128);
749
750 thread->initialize_thread_current();
751
752 OSThread* osthread = thread->osthread();
753 Monitor* sync = osthread->startThread_lock();
754
755 osthread->set_thread_id(os::current_thread_id());
756
757 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
758 os::current_thread_id(), (uintx) pthread_self());
759
760 if (UseNUMA) {
761 int lgrp_id = os::numa_get_group_id();
762 if (lgrp_id != -1) {
763 thread->set_lgrp_id(lgrp_id);
764 }
765 }
766 // initialize signal mask for this thread
767 os::Linux::hotspot_sigmask(thread);
768
769 // initialize floating point control register
770 os::Linux::init_thread_fpu_state();
771
772 // handshaking with parent thread
773 {
774 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
775
776 // notify parent thread
777 osthread->set_state(INITIALIZED);
778 sync->notify_all();
779
780 // wait until os::start_thread()
781 while (osthread->get_state() == INITIALIZED) {
782 sync->wait_without_safepoint_check();
783 }
784 }
785
786 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
787
788 // call one more level start routine
789 thread->call_run();
790
791 // Note: at this point the thread object may already have deleted itself.
792 // Prevent dereferencing it from here on out.
793 thread = NULL;
794
795 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
796 os::current_thread_id(), (uintx) pthread_self());
797
798 return 0;
799 }
800
801 // On Linux, glibc places static TLS blocks (for __thread variables) on
802 // the thread stack. This decreases the stack size actually available
803 // to threads.
804 //
805 // For large static TLS sizes, this may cause threads to malfunction due
806 // to insufficient stack space. This is a well-known issue in glibc:
807 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
808 //
809 // As a workaround, we call a private but assumed-stable glibc function,
810 // __pthread_get_minstack() to obtain the minstack size and derive the
811 // static TLS size from it. We then increase the user requested stack
812 // size by this TLS size.
813 //
814 // Due to compatibility concerns, this size adjustment is opt-in and
815 // controlled via AdjustStackSizeForTLS.
816 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
817
818 GetMinStack _get_minstack_func = NULL;
819
get_minstack_init()820 static void get_minstack_init() {
821 _get_minstack_func =
822 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
823 log_info(os, thread)("Lookup of __pthread_get_minstack %s",
824 _get_minstack_func == NULL ? "failed" : "succeeded");
825 }
826
827 // Returns the size of the static TLS area glibc puts on thread stacks.
828 // The value is cached on first use, which occurs when the first thread
829 // is created during VM initialization.
get_static_tls_area_size(const pthread_attr_t * attr)830 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
831 size_t tls_size = 0;
832 if (_get_minstack_func != NULL) {
833 // Obtain the pthread minstack size by calling __pthread_get_minstack.
834 size_t minstack_size = _get_minstack_func(attr);
835
836 // Remove non-TLS area size included in minstack size returned
837 // by __pthread_get_minstack() to get the static TLS size.
838 // In glibc before 2.27, minstack size includes guard_size.
839 // In glibc 2.27 and later, guard_size is automatically added
840 // to the stack size by pthread_create and is no longer included
841 // in minstack size. In both cases, the guard_size is taken into
842 // account, so there is no need to adjust the result for that.
843 //
844 // Although __pthread_get_minstack() is a private glibc function,
845 // it is expected to have a stable behavior across future glibc
846 // versions while glibc still allocates the static TLS blocks off
847 // the stack. Following is glibc 2.28 __pthread_get_minstack():
848 //
849 // size_t
850 // __pthread_get_minstack (const pthread_attr_t *attr)
851 // {
852 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
853 // }
854 //
855 //
856 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
857 // if check is done for precaution.
858 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
859 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
860 }
861 }
862
863 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
864 tls_size);
865 return tls_size;
866 }
867
create_thread(Thread * thread,ThreadType thr_type,size_t req_stack_size)868 bool os::create_thread(Thread* thread, ThreadType thr_type,
869 size_t req_stack_size) {
870 assert(thread->osthread() == NULL, "caller responsible");
871
872 // Allocate the OSThread object
873 OSThread* osthread = new OSThread(NULL, NULL);
874 if (osthread == NULL) {
875 return false;
876 }
877
878 // set the correct thread state
879 osthread->set_thread_type(thr_type);
880
881 // Initial state is ALLOCATED but not INITIALIZED
882 osthread->set_state(ALLOCATED);
883
884 thread->set_osthread(osthread);
885
886 // init thread attributes
887 pthread_attr_t attr;
888 pthread_attr_init(&attr);
889 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
890
891 // Calculate stack size if it's not specified by caller.
892 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
893 // In glibc versions prior to 2.7 the guard size mechanism
894 // is not implemented properly. The posix standard requires adding
895 // the size of the guard pages to the stack size, instead Linux
896 // takes the space out of 'stacksize'. Thus we adapt the requested
897 // stack_size by the size of the guard pages to mimick proper
898 // behaviour. However, be careful not to end up with a size
899 // of zero due to overflow. Don't add the guard page in that case.
900 size_t guard_size = os::Linux::default_guard_size(thr_type);
901 // Configure glibc guard page. Must happen before calling
902 // get_static_tls_area_size(), which uses the guard_size.
903 pthread_attr_setguardsize(&attr, guard_size);
904
905 size_t stack_adjust_size = 0;
906 if (AdjustStackSizeForTLS) {
907 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
908 stack_adjust_size += get_static_tls_area_size(&attr);
909 } else {
910 stack_adjust_size += guard_size;
911 }
912
913 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
914 if (stack_size <= SIZE_MAX - stack_adjust_size) {
915 stack_size += stack_adjust_size;
916 }
917 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
918
919 int status = pthread_attr_setstacksize(&attr, stack_size);
920 assert_status(status == 0, status, "pthread_attr_setstacksize");
921
922 ThreadState state;
923
924 {
925 pthread_t tid;
926 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
927
928 char buf[64];
929 if (ret == 0) {
930 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
931 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
932 } else {
933 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
934 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
935 // Log some OS information which might explain why creating the thread failed.
936 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
937 LogStream st(Log(os, thread)::info());
938 os::Posix::print_rlimit_info(&st);
939 os::print_memory_info(&st);
940 os::Linux::print_proc_sys_info(&st);
941 os::Linux::print_container_info(&st);
942 }
943
944 pthread_attr_destroy(&attr);
945
946 if (ret != 0) {
947 // Need to clean up stuff we've allocated so far
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
952
953 // Store pthread info into the OSThread
954 osthread->set_pthread_id(tid);
955
956 // Wait until child thread is either initialized or aborted
957 {
958 Monitor* sync_with_child = osthread->startThread_lock();
959 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
960 while ((state = osthread->get_state()) == ALLOCATED) {
961 sync_with_child->wait_without_safepoint_check();
962 }
963 }
964 }
965
966 // Aborted due to thread limit being reached
967 if (state == ZOMBIE) {
968 thread->set_osthread(NULL);
969 delete osthread;
970 return false;
971 }
972
973 // The thread is returned suspended (in state INITIALIZED),
974 // and is started higher up in the call chain
975 assert(state == INITIALIZED, "race condition");
976 return true;
977 }
978
979 /////////////////////////////////////////////////////////////////////////////
980 // attach existing thread
981
982 // bootstrap the main thread
create_main_thread(JavaThread * thread)983 bool os::create_main_thread(JavaThread* thread) {
984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
985 return create_attached_thread(thread);
986 }
987
create_attached_thread(JavaThread * thread)988 bool os::create_attached_thread(JavaThread* thread) {
989 #ifdef ASSERT
990 thread->verify_not_published();
991 #endif
992
993 // Allocate the OSThread object
994 OSThread* osthread = new OSThread(NULL, NULL);
995
996 if (osthread == NULL) {
997 return false;
998 }
999
1000 // Store pthread info into the OSThread
1001 osthread->set_thread_id(os::Linux::gettid());
1002 osthread->set_pthread_id(::pthread_self());
1003
1004 // initialize floating point control register
1005 os::Linux::init_thread_fpu_state();
1006
1007 // Initial thread state is RUNNABLE
1008 osthread->set_state(RUNNABLE);
1009
1010 thread->set_osthread(osthread);
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018
1019 if (os::is_primordial_thread()) {
1020 // If current thread is primordial thread, its stack is mapped on demand,
1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022 // the entire stack region to avoid SEGV in stack banging.
1023 // It is also useful to get around the heap-stack-gap problem on SuSE
1024 // kernel (see 4821821 for details). We first expand stack to the top
1025 // of yellow zone, then enable stack yellow zone (order is significant,
1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027 // is no gap between the last two virtual memory regions.
1028
1029 JavaThread *jt = (JavaThread *)thread;
1030 address addr = jt->stack_reserved_zone_base();
1031 assert(addr != NULL, "initialization problem?");
1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033
1034 osthread->set_expanding_stack();
1035 os::Linux::manually_expand_stack(jt, addr);
1036 osthread->clear_expanding_stack();
1037 }
1038
1039 // initialize signal mask for this thread
1040 // and save the caller's signal mask
1041 os::Linux::hotspot_sigmask(thread);
1042
1043 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1044 os::current_thread_id(), (uintx) pthread_self());
1045
1046 return true;
1047 }
1048
pd_start_thread(Thread * thread)1049 void os::pd_start_thread(Thread* thread) {
1050 OSThread * osthread = thread->osthread();
1051 assert(osthread->get_state() != INITIALIZED, "just checking");
1052 Monitor* sync_with_child = osthread->startThread_lock();
1053 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1054 sync_with_child->notify();
1055 }
1056
1057 // Free Linux resources related to the OSThread
free_thread(OSThread * osthread)1058 void os::free_thread(OSThread* osthread) {
1059 assert(osthread != NULL, "osthread not set");
1060
1061 // We are told to free resources of the argument thread,
1062 // but we can only really operate on the current thread.
1063 assert(Thread::current()->osthread() == osthread,
1064 "os::free_thread but not current thread");
1065
1066 #ifdef ASSERT
1067 sigset_t current;
1068 sigemptyset(¤t);
1069 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
1070 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
1071 #endif
1072
1073 // Restore caller's signal mask
1074 sigset_t sigmask = osthread->caller_sigmask();
1075 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1076
1077 delete osthread;
1078 }
1079
1080 //////////////////////////////////////////////////////////////////////////////
1081 // primordial thread
1082
1083 // Check if current thread is the primordial thread, similar to Solaris thr_main.
is_primordial_thread(void)1084 bool os::is_primordial_thread(void) {
1085 if (suppress_primordial_thread_resolution) {
1086 return false;
1087 }
1088 char dummy;
1089 // If called before init complete, thread stack bottom will be null.
1090 // Can be called if fatal error occurs before initialization.
1091 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1092 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1093 os::Linux::initial_thread_stack_size() != 0,
1094 "os::init did not locate primordial thread's stack region");
1095 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1096 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1097 os::Linux::initial_thread_stack_size()) {
1098 return true;
1099 } else {
1100 return false;
1101 }
1102 }
1103
1104 // Find the virtual memory area that contains addr
find_vma(address addr,address * vma_low,address * vma_high)1105 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1106 FILE *fp = fopen("/proc/self/maps", "r");
1107 if (fp) {
1108 address low, high;
1109 while (!feof(fp)) {
1110 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1111 if (low <= addr && addr < high) {
1112 if (vma_low) *vma_low = low;
1113 if (vma_high) *vma_high = high;
1114 fclose(fp);
1115 return true;
1116 }
1117 }
1118 for (;;) {
1119 int ch = fgetc(fp);
1120 if (ch == EOF || ch == (int)'\n') break;
1121 }
1122 }
1123 fclose(fp);
1124 }
1125 return false;
1126 }
1127
1128 // Locate primordial thread stack. This special handling of primordial thread stack
1129 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1130 // bogus value for the primordial process thread. While the launcher has created
1131 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1132 // JNI invocation API from a primordial thread.
capture_initial_stack(size_t max_size)1133 void os::Linux::capture_initial_stack(size_t max_size) {
1134
1135 // max_size is either 0 (which means accept OS default for thread stacks) or
1136 // a user-specified value known to be at least the minimum needed. If we
1137 // are actually on the primordial thread we can make it appear that we have a
1138 // smaller max_size stack by inserting the guard pages at that location. But we
1139 // cannot do anything to emulate a larger stack than what has been provided by
1140 // the OS or threading library. In fact if we try to use a stack greater than
1141 // what is set by rlimit then we will crash the hosting process.
1142
1143 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1144 // If this is "unlimited" then it will be a huge value.
1145 struct rlimit rlim;
1146 getrlimit(RLIMIT_STACK, &rlim);
1147 size_t stack_size = rlim.rlim_cur;
1148
1149 // 6308388: a bug in ld.so will relocate its own .data section to the
1150 // lower end of primordial stack; reduce ulimit -s value a little bit
1151 // so we won't install guard page on ld.so's data section.
1152 // But ensure we don't underflow the stack size - allow 1 page spare
1153 if (stack_size >= (size_t)(3 * page_size())) {
1154 stack_size -= 2 * page_size();
1155 }
1156
1157 // Try to figure out where the stack base (top) is. This is harder.
1158 //
1159 // When an application is started, glibc saves the initial stack pointer in
1160 // a global variable "__libc_stack_end", which is then used by system
1161 // libraries. __libc_stack_end should be pretty close to stack top. The
1162 // variable is available since the very early days. However, because it is
1163 // a private interface, it could disappear in the future.
1164 //
1165 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1166 // to __libc_stack_end, it is very close to stack top, but isn't the real
1167 // stack top. Note that /proc may not exist if VM is running as a chroot
1168 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1169 // /proc/<pid>/stat could change in the future (though unlikely).
1170 //
1171 // We try __libc_stack_end first. If that doesn't work, look for
1172 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1173 // as a hint, which should work well in most cases.
1174
1175 uintptr_t stack_start;
1176
1177 // try __libc_stack_end first
1178 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1179 if (p && *p) {
1180 stack_start = *p;
1181 } else {
1182 // see if we can get the start_stack field from /proc/self/stat
1183 FILE *fp;
1184 int pid;
1185 char state;
1186 int ppid;
1187 int pgrp;
1188 int session;
1189 int nr;
1190 int tpgrp;
1191 unsigned long flags;
1192 unsigned long minflt;
1193 unsigned long cminflt;
1194 unsigned long majflt;
1195 unsigned long cmajflt;
1196 unsigned long utime;
1197 unsigned long stime;
1198 long cutime;
1199 long cstime;
1200 long prio;
1201 long nice;
1202 long junk;
1203 long it_real;
1204 uintptr_t start;
1205 uintptr_t vsize;
1206 intptr_t rss;
1207 uintptr_t rsslim;
1208 uintptr_t scodes;
1209 uintptr_t ecode;
1210 int i;
1211
1212 // Figure what the primordial thread stack base is. Code is inspired
1213 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1214 // followed by command name surrounded by parentheses, state, etc.
1215 char stat[2048];
1216 int statlen;
1217
1218 fp = fopen("/proc/self/stat", "r");
1219 if (fp) {
1220 statlen = fread(stat, 1, 2047, fp);
1221 stat[statlen] = '\0';
1222 fclose(fp);
1223
1224 // Skip pid and the command string. Note that we could be dealing with
1225 // weird command names, e.g. user could decide to rename java launcher
1226 // to "java 1.4.2 :)", then the stat file would look like
1227 // 1234 (java 1.4.2 :)) R ... ...
1228 // We don't really need to know the command string, just find the last
1229 // occurrence of ")" and then start parsing from there. See bug 4726580.
1230 char * s = strrchr(stat, ')');
1231
1232 i = 0;
1233 if (s) {
1234 // Skip blank chars
1235 do { s++; } while (s && isspace(*s));
1236
1237 #define _UFM UINTX_FORMAT
1238 #define _DFM INTX_FORMAT
1239
1240 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1241 // 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
1242 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,
1243 &state, // 3 %c
1244 &ppid, // 4 %d
1245 &pgrp, // 5 %d
1246 &session, // 6 %d
1247 &nr, // 7 %d
1248 &tpgrp, // 8 %d
1249 &flags, // 9 %lu
1250 &minflt, // 10 %lu
1251 &cminflt, // 11 %lu
1252 &majflt, // 12 %lu
1253 &cmajflt, // 13 %lu
1254 &utime, // 14 %lu
1255 &stime, // 15 %lu
1256 &cutime, // 16 %ld
1257 &cstime, // 17 %ld
1258 &prio, // 18 %ld
1259 &nice, // 19 %ld
1260 &junk, // 20 %ld
1261 &it_real, // 21 %ld
1262 &start, // 22 UINTX_FORMAT
1263 &vsize, // 23 UINTX_FORMAT
1264 &rss, // 24 INTX_FORMAT
1265 &rsslim, // 25 UINTX_FORMAT
1266 &scodes, // 26 UINTX_FORMAT
1267 &ecode, // 27 UINTX_FORMAT
1268 &stack_start); // 28 UINTX_FORMAT
1269 }
1270
1271 #undef _UFM
1272 #undef _DFM
1273
1274 if (i != 28 - 2) {
1275 assert(false, "Bad conversion from /proc/self/stat");
1276 // product mode - assume we are the primordial thread, good luck in the
1277 // embedded case.
1278 warning("Can't detect primordial thread stack location - bad conversion");
1279 stack_start = (uintptr_t) &rlim;
1280 }
1281 } else {
1282 // For some reason we can't open /proc/self/stat (for example, running on
1283 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1284 // most cases, so don't abort:
1285 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1286 stack_start = (uintptr_t) &rlim;
1287 }
1288 }
1289
1290 // Now we have a pointer (stack_start) very close to the stack top, the
1291 // next thing to do is to figure out the exact location of stack top. We
1292 // can find out the virtual memory area that contains stack_start by
1293 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1294 // and its upper limit is the real stack top. (again, this would fail if
1295 // running inside chroot, because /proc may not exist.)
1296
1297 uintptr_t stack_top;
1298 address low, high;
1299 if (find_vma((address)stack_start, &low, &high)) {
1300 // success, "high" is the true stack top. (ignore "low", because initial
1301 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1302 stack_top = (uintptr_t)high;
1303 } else {
1304 // failed, likely because /proc/self/maps does not exist
1305 warning("Can't detect primordial thread stack location - find_vma failed");
1306 // best effort: stack_start is normally within a few pages below the real
1307 // stack top, use it as stack top, and reduce stack size so we won't put
1308 // guard page outside stack.
1309 stack_top = stack_start;
1310 stack_size -= 16 * page_size();
1311 }
1312
1313 // stack_top could be partially down the page so align it
1314 stack_top = align_up(stack_top, page_size());
1315
1316 // Allowed stack value is minimum of max_size and what we derived from rlimit
1317 if (max_size > 0) {
1318 _initial_thread_stack_size = MIN2(max_size, stack_size);
1319 } else {
1320 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1321 // clamp it at 8MB as we do on Solaris
1322 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1323 }
1324 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1325 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1326
1327 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1328
1329 if (log_is_enabled(Info, os, thread)) {
1330 // See if we seem to be on primordial process thread
1331 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1332 uintptr_t(&rlim) < stack_top;
1333
1334 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1335 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1336 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1337 stack_top, intptr_t(_initial_thread_stack_bottom));
1338 }
1339 }
1340
1341 ////////////////////////////////////////////////////////////////////////////////
1342 // time support
1343
1344 #ifndef SUPPORTS_CLOCK_MONOTONIC
1345 #error "Build platform doesn't support clock_gettime and related functionality"
1346 #endif
1347
1348 // Time since start-up in seconds to a fine granularity.
1349 // Used by VMSelfDestructTimer and the MemProfiler.
elapsedTime()1350 double os::elapsedTime() {
1351
1352 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1353 }
1354
elapsed_counter()1355 jlong os::elapsed_counter() {
1356 return javaTimeNanos() - initial_time_count;
1357 }
1358
elapsed_frequency()1359 jlong os::elapsed_frequency() {
1360 return NANOSECS_PER_SEC; // nanosecond resolution
1361 }
1362
supports_vtime()1363 bool os::supports_vtime() { return true; }
1364
elapsedVTime()1365 double os::elapsedVTime() {
1366 struct rusage usage;
1367 int retval = getrusage(RUSAGE_THREAD, &usage);
1368 if (retval == 0) {
1369 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);
1370 } else {
1371 // better than nothing, but not much
1372 return elapsedTime();
1373 }
1374 }
1375
javaTimeMillis()1376 jlong os::javaTimeMillis() {
1377 timeval time;
1378 int status = gettimeofday(&time, NULL);
1379 assert(status != -1, "linux error");
1380 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1381 }
1382
javaTimeSystemUTC(jlong & seconds,jlong & nanos)1383 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1384 timeval time;
1385 int status = gettimeofday(&time, NULL);
1386 assert(status != -1, "linux error");
1387 seconds = jlong(time.tv_sec);
1388 nanos = jlong(time.tv_usec) * 1000;
1389 }
1390
fast_thread_clock_init()1391 void os::Linux::fast_thread_clock_init() {
1392 if (!UseLinuxPosixThreadCPUClocks) {
1393 return;
1394 }
1395 clockid_t clockid;
1396 struct timespec tp;
1397 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1398 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1399
1400 // Switch to using fast clocks for thread cpu time if
1401 // the clock_getres() returns 0 error code.
1402 // Note, that some kernels may support the current thread
1403 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1404 // returned by the pthread_getcpuclockid().
1405 // If the fast Posix clocks are supported then the clock_getres()
1406 // must return at least tp.tv_sec == 0 which means a resolution
1407 // better than 1 sec. This is extra check for reliability.
1408
1409 if (pthread_getcpuclockid_func &&
1410 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1411 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1412 _supports_fast_thread_cpu_time = true;
1413 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1414 }
1415 }
1416
javaTimeNanos()1417 jlong os::javaTimeNanos() {
1418 if (os::supports_monotonic_clock()) {
1419 struct timespec tp;
1420 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1421 assert(status == 0, "gettime error");
1422 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1423 return result;
1424 } else {
1425 timeval time;
1426 int status = gettimeofday(&time, NULL);
1427 assert(status != -1, "linux error");
1428 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1429 return 1000 * usecs;
1430 }
1431 }
1432
javaTimeNanos_info(jvmtiTimerInfo * info_ptr)1433 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1434 if (os::supports_monotonic_clock()) {
1435 info_ptr->max_value = ALL_64_BITS;
1436
1437 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1438 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1439 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1440 } else {
1441 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1442 info_ptr->max_value = ALL_64_BITS;
1443
1444 // gettimeofday is a real time clock so it skips
1445 info_ptr->may_skip_backward = true;
1446 info_ptr->may_skip_forward = true;
1447 }
1448
1449 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1450 }
1451
1452 // Return the real, user, and system times in seconds from an
1453 // arbitrary fixed point in the past.
getTimesSecs(double * process_real_time,double * process_user_time,double * process_system_time)1454 bool os::getTimesSecs(double* process_real_time,
1455 double* process_user_time,
1456 double* process_system_time) {
1457 struct tms ticks;
1458 clock_t real_ticks = times(&ticks);
1459
1460 if (real_ticks == (clock_t) (-1)) {
1461 return false;
1462 } else {
1463 double ticks_per_second = (double) clock_tics_per_sec;
1464 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1465 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1466 *process_real_time = ((double) real_ticks) / ticks_per_second;
1467
1468 return true;
1469 }
1470 }
1471
1472
local_time_string(char * buf,size_t buflen)1473 char * os::local_time_string(char *buf, size_t buflen) {
1474 struct tm t;
1475 time_t long_time;
1476 time(&long_time);
1477 localtime_r(&long_time, &t);
1478 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1479 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1480 t.tm_hour, t.tm_min, t.tm_sec);
1481 return buf;
1482 }
1483
localtime_pd(const time_t * clock,struct tm * res)1484 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1485 return localtime_r(clock, res);
1486 }
1487
1488 ////////////////////////////////////////////////////////////////////////////////
1489 // runtime exit support
1490
1491 // Note: os::shutdown() might be called very early during initialization, or
1492 // called from signal handler. Before adding something to os::shutdown(), make
1493 // sure it is async-safe and can handle partially initialized VM.
shutdown()1494 void os::shutdown() {
1495
1496 // allow PerfMemory to attempt cleanup of any persistent resources
1497 perfMemory_exit();
1498
1499 // needs to remove object in file system
1500 AttachListener::abort();
1501
1502 // flush buffered output, finish log files
1503 ostream_abort();
1504
1505 // Check for abort hook
1506 abort_hook_t abort_hook = Arguments::abort_hook();
1507 if (abort_hook != NULL) {
1508 abort_hook();
1509 }
1510
1511 }
1512
1513 // Note: os::abort() might be called very early during initialization, or
1514 // called from signal handler. Before adding something to os::abort(), make
1515 // sure it is async-safe and can handle partially initialized VM.
abort(bool dump_core,void * siginfo,const void * context)1516 void os::abort(bool dump_core, void* siginfo, const void* context) {
1517 os::shutdown();
1518 if (dump_core) {
1519 if (DumpPrivateMappingsInCore) {
1520 ClassLoader::close_jrt_image();
1521 }
1522 #ifndef PRODUCT
1523 fdStream out(defaultStream::output_fd());
1524 out.print_raw("Current thread is ");
1525 char buf[16];
1526 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1527 out.print_raw_cr(buf);
1528 out.print_raw_cr("Dumping core ...");
1529 #endif
1530 ::abort(); // dump core
1531 }
1532
1533 ::exit(1);
1534 }
1535
1536 // Die immediately, no exit hook, no abort hook, no cleanup.
1537 // Dump a core file, if possible, for debugging.
die()1538 void os::die() {
1539 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1540 // For TimeoutInErrorHandlingTest.java, we just kill the VM
1541 // and don't take the time to generate a core file.
1542 os::signal_raise(SIGKILL);
1543 } else {
1544 ::abort();
1545 }
1546 }
1547
1548 // thread_id is kernel thread id (similar to Solaris LWP id)
current_thread_id()1549 intx os::current_thread_id() { return os::Linux::gettid(); }
current_process_id()1550 int os::current_process_id() {
1551 return ::getpid();
1552 }
1553
1554 // DLL functions
1555
dll_file_extension()1556 const char* os::dll_file_extension() { return ".so"; }
1557
1558 // This must be hard coded because it's the system's temporary
1559 // directory not the java application's temp directory, ala java.io.tmpdir.
get_temp_directory()1560 const char* os::get_temp_directory() { return "/tmp"; }
1561
file_exists(const char * filename)1562 static bool file_exists(const char* filename) {
1563 struct stat statbuf;
1564 if (filename == NULL || strlen(filename) == 0) {
1565 return false;
1566 }
1567 return os::stat(filename, &statbuf) == 0;
1568 }
1569
1570 // check if addr is inside libjvm.so
address_is_in_vm(address addr)1571 bool os::address_is_in_vm(address addr) {
1572 static address libjvm_base_addr;
1573 Dl_info dlinfo;
1574
1575 if (libjvm_base_addr == NULL) {
1576 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1577 libjvm_base_addr = (address)dlinfo.dli_fbase;
1578 }
1579 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1580 }
1581
1582 if (dladdr((void *)addr, &dlinfo) != 0) {
1583 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1584 }
1585
1586 return false;
1587 }
1588
dll_address_to_function_name(address addr,char * buf,int buflen,int * offset,bool demangle)1589 bool os::dll_address_to_function_name(address addr, char *buf,
1590 int buflen, int *offset,
1591 bool demangle) {
1592 // buf is not optional, but offset is optional
1593 assert(buf != NULL, "sanity check");
1594
1595 Dl_info dlinfo;
1596
1597 if (dladdr((void*)addr, &dlinfo) != 0) {
1598 // see if we have a matching symbol
1599 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1600 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1601 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1602 }
1603 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1604 return true;
1605 }
1606 // no matching symbol so try for just file info
1607 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1608 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1609 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1610 return true;
1611 }
1612 }
1613 }
1614
1615 buf[0] = '\0';
1616 if (offset != NULL) *offset = -1;
1617 return false;
1618 }
1619
1620 struct _address_to_library_name {
1621 address addr; // input : memory address
1622 size_t buflen; // size of fname
1623 char* fname; // output: library name
1624 address base; // library base addr
1625 };
1626
address_to_library_name_callback(struct dl_phdr_info * info,size_t size,void * data)1627 static int address_to_library_name_callback(struct dl_phdr_info *info,
1628 size_t size, void *data) {
1629 int i;
1630 bool found = false;
1631 address libbase = NULL;
1632 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1633
1634 // iterate through all loadable segments
1635 for (i = 0; i < info->dlpi_phnum; i++) {
1636 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1637 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1638 // base address of a library is the lowest address of its loaded
1639 // segments.
1640 if (libbase == NULL || libbase > segbase) {
1641 libbase = segbase;
1642 }
1643 // see if 'addr' is within current segment
1644 if (segbase <= d->addr &&
1645 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1646 found = true;
1647 }
1648 }
1649 }
1650
1651 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1652 // so dll_address_to_library_name() can fall through to use dladdr() which
1653 // can figure out executable name from argv[0].
1654 if (found && info->dlpi_name && info->dlpi_name[0]) {
1655 d->base = libbase;
1656 if (d->fname) {
1657 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1658 }
1659 return 1;
1660 }
1661 return 0;
1662 }
1663
dll_address_to_library_name(address addr,char * buf,int buflen,int * offset)1664 bool os::dll_address_to_library_name(address addr, char* buf,
1665 int buflen, int* offset) {
1666 // buf is not optional, but offset is optional
1667 assert(buf != NULL, "sanity check");
1668
1669 Dl_info dlinfo;
1670 struct _address_to_library_name data;
1671
1672 // There is a bug in old glibc dladdr() implementation that it could resolve
1673 // to wrong library name if the .so file has a base address != NULL. Here
1674 // we iterate through the program headers of all loaded libraries to find
1675 // out which library 'addr' really belongs to. This workaround can be
1676 // removed once the minimum requirement for glibc is moved to 2.3.x.
1677 data.addr = addr;
1678 data.fname = buf;
1679 data.buflen = buflen;
1680 data.base = NULL;
1681 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1682
1683 if (rslt) {
1684 // buf already contains library name
1685 if (offset) *offset = addr - data.base;
1686 return true;
1687 }
1688 if (dladdr((void*)addr, &dlinfo) != 0) {
1689 if (dlinfo.dli_fname != NULL) {
1690 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1691 }
1692 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1693 *offset = addr - (address)dlinfo.dli_fbase;
1694 }
1695 return true;
1696 }
1697
1698 buf[0] = '\0';
1699 if (offset) *offset = -1;
1700 return false;
1701 }
1702
1703 // Loads .dll/.so and
1704 // in case of error it checks if .dll/.so was built for the
1705 // same architecture as Hotspot is running on
1706
1707
1708 // Remember the stack's state. The Linux dynamic linker will change
1709 // the stack to 'executable' at most once, so we must safepoint only once.
1710 bool os::Linux::_stack_is_executable = false;
1711
1712 // VM operation that loads a library. This is necessary if stack protection
1713 // of the Java stacks can be lost during loading the library. If we
1714 // do not stop the Java threads, they can stack overflow before the stacks
1715 // are protected again.
1716 class VM_LinuxDllLoad: public VM_Operation {
1717 private:
1718 const char *_filename;
1719 char *_ebuf;
1720 int _ebuflen;
1721 void *_lib;
1722 public:
VM_LinuxDllLoad(const char * fn,char * ebuf,int ebuflen)1723 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1724 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
type() const1725 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
doit()1726 void doit() {
1727 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1728 os::Linux::_stack_is_executable = true;
1729 }
loaded_library()1730 void* loaded_library() { return _lib; }
1731 };
1732
dll_load(const char * filename,char * ebuf,int ebuflen)1733 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1734 void * result = NULL;
1735 bool load_attempted = false;
1736
1737 log_info(os)("attempting shared library load of %s", filename);
1738
1739 // Check whether the library to load might change execution rights
1740 // of the stack. If they are changed, the protection of the stack
1741 // guard pages will be lost. We need a safepoint to fix this.
1742 //
1743 // See Linux man page execstack(8) for more info.
1744 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1745 if (!ElfFile::specifies_noexecstack(filename)) {
1746 if (!is_init_completed()) {
1747 os::Linux::_stack_is_executable = true;
1748 // This is OK - No Java threads have been created yet, and hence no
1749 // stack guard pages to fix.
1750 //
1751 // Dynamic loader will make all stacks executable after
1752 // this function returns, and will not do that again.
1753 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1754 } else {
1755 warning("You have loaded library %s which might have disabled stack guard. "
1756 "The VM will try to fix the stack guard now.\n"
1757 "It's highly recommended that you fix the library with "
1758 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1759 filename);
1760
1761 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1762 JavaThread *jt = JavaThread::current();
1763 if (jt->thread_state() != _thread_in_native) {
1764 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1765 // that requires ExecStack. Cannot enter safe point. Let's give up.
1766 warning("Unable to fix stack guard. Giving up.");
1767 } else {
1768 if (!LoadExecStackDllInVMThread) {
1769 // This is for the case where the DLL has an static
1770 // constructor function that executes JNI code. We cannot
1771 // load such DLLs in the VMThread.
1772 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1773 }
1774
1775 ThreadInVMfromNative tiv(jt);
1776 debug_only(VMNativeEntryWrapper vew;)
1777
1778 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1779 VMThread::execute(&op);
1780 if (LoadExecStackDllInVMThread) {
1781 result = op.loaded_library();
1782 }
1783 load_attempted = true;
1784 }
1785 }
1786 }
1787 }
1788
1789 if (!load_attempted) {
1790 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1791 }
1792
1793 if (result != NULL) {
1794 // Successful loading
1795 return result;
1796 }
1797
1798 Elf32_Ehdr elf_head;
1799 int diag_msg_max_length=ebuflen-strlen(ebuf);
1800 char* diag_msg_buf=ebuf+strlen(ebuf);
1801
1802 if (diag_msg_max_length==0) {
1803 // No more space in ebuf for additional diagnostics message
1804 return NULL;
1805 }
1806
1807
1808 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1809
1810 if (file_descriptor < 0) {
1811 // Can't open library, report dlerror() message
1812 return NULL;
1813 }
1814
1815 bool failed_to_read_elf_head=
1816 (sizeof(elf_head)!=
1817 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1818
1819 ::close(file_descriptor);
1820 if (failed_to_read_elf_head) {
1821 // file i/o error - report dlerror() msg
1822 return NULL;
1823 }
1824
1825 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1826 // handle invalid/out of range endianness values
1827 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1828 return NULL;
1829 }
1830
1831 #if defined(VM_LITTLE_ENDIAN)
1832 // VM is LE, shared object BE
1833 elf_head.e_machine = be16toh(elf_head.e_machine);
1834 #else
1835 // VM is BE, shared object LE
1836 elf_head.e_machine = le16toh(elf_head.e_machine);
1837 #endif
1838 }
1839
1840 typedef struct {
1841 Elf32_Half code; // Actual value as defined in elf.h
1842 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1843 unsigned char elf_class; // 32 or 64 bit
1844 unsigned char endianness; // MSB or LSB
1845 char* name; // String representation
1846 } arch_t;
1847
1848 #ifndef EM_486
1849 #define EM_486 6 /* Intel 80486 */
1850 #endif
1851 #ifndef EM_AARCH64
1852 #define EM_AARCH64 183 /* ARM AARCH64 */
1853 #endif
1854
1855 static const arch_t arch_array[]={
1856 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1857 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1858 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1859 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1860 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1861 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1862 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1863 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1864 #if defined(VM_LITTLE_ENDIAN)
1865 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1866 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1867 #else
1868 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1869 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1870 #endif
1871 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1872 // we only support 64 bit z architecture
1873 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1874 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1875 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1876 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1877 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1878 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1879 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1880 };
1881
1882 #if (defined IA32)
1883 static Elf32_Half running_arch_code=EM_386;
1884 #elif (defined AMD64) || (defined X32)
1885 static Elf32_Half running_arch_code=EM_X86_64;
1886 #elif (defined IA64)
1887 static Elf32_Half running_arch_code=EM_IA_64;
1888 #elif (defined __sparc) && (defined _LP64)
1889 static Elf32_Half running_arch_code=EM_SPARCV9;
1890 #elif (defined __sparc) && (!defined _LP64)
1891 static Elf32_Half running_arch_code=EM_SPARC;
1892 #elif (defined __powerpc64__)
1893 static Elf32_Half running_arch_code=EM_PPC64;
1894 #elif (defined __powerpc__)
1895 static Elf32_Half running_arch_code=EM_PPC;
1896 #elif (defined AARCH64)
1897 static Elf32_Half running_arch_code=EM_AARCH64;
1898 #elif (defined ARM)
1899 static Elf32_Half running_arch_code=EM_ARM;
1900 #elif (defined S390)
1901 static Elf32_Half running_arch_code=EM_S390;
1902 #elif (defined ALPHA)
1903 static Elf32_Half running_arch_code=EM_ALPHA;
1904 #elif (defined MIPSEL)
1905 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1906 #elif (defined PARISC)
1907 static Elf32_Half running_arch_code=EM_PARISC;
1908 #elif (defined MIPS)
1909 static Elf32_Half running_arch_code=EM_MIPS;
1910 #elif (defined M68K)
1911 static Elf32_Half running_arch_code=EM_68K;
1912 #elif (defined SH)
1913 static Elf32_Half running_arch_code=EM_SH;
1914 #else
1915 #error Method os::dll_load requires that one of following is defined:\
1916 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1917 #endif
1918
1919 // Identify compatibility class for VM's architecture and library's architecture
1920 // Obtain string descriptions for architectures
1921
1922 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1923 int running_arch_index=-1;
1924
1925 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1926 if (running_arch_code == arch_array[i].code) {
1927 running_arch_index = i;
1928 }
1929 if (lib_arch.code == arch_array[i].code) {
1930 lib_arch.compat_class = arch_array[i].compat_class;
1931 lib_arch.name = arch_array[i].name;
1932 }
1933 }
1934
1935 assert(running_arch_index != -1,
1936 "Didn't find running architecture code (running_arch_code) in arch_array");
1937 if (running_arch_index == -1) {
1938 // Even though running architecture detection failed
1939 // we may still continue with reporting dlerror() message
1940 return NULL;
1941 }
1942
1943 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1944 if (lib_arch.name != NULL) {
1945 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1946 " (Possible cause: can't load %s .so on a %s platform)",
1947 lib_arch.name, arch_array[running_arch_index].name);
1948 } else {
1949 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1950 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1951 lib_arch.code, arch_array[running_arch_index].name);
1952 }
1953 return NULL;
1954 }
1955
1956 if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1957 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1958 return NULL;
1959 }
1960
1961 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1962 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1963 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1964 return NULL;
1965 }
1966
1967 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1968 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1969 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1970 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1971 return NULL;
1972 }
1973
1974 return NULL;
1975 }
1976
dlopen_helper(const char * filename,char * ebuf,int ebuflen)1977 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1978 int ebuflen) {
1979 void * result = ::dlopen(filename, RTLD_LAZY);
1980 if (result == NULL) {
1981 const char* error_report = ::dlerror();
1982 if (error_report == NULL) {
1983 error_report = "dlerror returned no error description";
1984 }
1985 if (ebuf != NULL && ebuflen > 0) {
1986 ::strncpy(ebuf, error_report, ebuflen-1);
1987 ebuf[ebuflen-1]='\0';
1988 }
1989 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1990 log_info(os)("shared library load of %s failed, %s", filename, error_report);
1991 } else {
1992 Events::log(NULL, "Loaded shared library %s", filename);
1993 log_info(os)("shared library load of %s was successful", filename);
1994 }
1995 return result;
1996 }
1997
dll_load_in_vmthread(const char * filename,char * ebuf,int ebuflen)1998 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1999 int ebuflen) {
2000 void * result = NULL;
2001 if (LoadExecStackDllInVMThread) {
2002 result = dlopen_helper(filename, ebuf, ebuflen);
2003 }
2004
2005 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2006 // library that requires an executable stack, or which does not have this
2007 // stack attribute set, dlopen changes the stack attribute to executable. The
2008 // read protection of the guard pages gets lost.
2009 //
2010 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2011 // may have been queued at the same time.
2012
2013 if (!_stack_is_executable) {
2014 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2015 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2016 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
2017 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2018 warning("Attempt to reguard stack yellow zone failed.");
2019 }
2020 }
2021 }
2022 }
2023
2024 return result;
2025 }
2026
dll_lookup(void * handle,const char * name)2027 void* os::dll_lookup(void* handle, const char* name) {
2028 void* res = dlsym(handle, name);
2029 return res;
2030 }
2031
get_default_process_handle()2032 void* os::get_default_process_handle() {
2033 return (void*)::dlopen(NULL, RTLD_LAZY);
2034 }
2035
_print_ascii_file(const char * filename,outputStream * st,const char * hdr=NULL)2036 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2037 int fd = ::open(filename, O_RDONLY);
2038 if (fd == -1) {
2039 return false;
2040 }
2041
2042 if (hdr != NULL) {
2043 st->print_cr("%s", hdr);
2044 }
2045
2046 char buf[33];
2047 int bytes;
2048 buf[32] = '\0';
2049 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2050 st->print_raw(buf, bytes);
2051 }
2052
2053 ::close(fd);
2054
2055 return true;
2056 }
2057
print_dll_info(outputStream * st)2058 void os::print_dll_info(outputStream *st) {
2059 st->print_cr("Dynamic libraries:");
2060
2061 char fname[32];
2062 pid_t pid = os::Linux::gettid();
2063
2064 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2065
2066 if (!_print_ascii_file(fname, st)) {
2067 st->print("Can not get library information for pid = %d\n", pid);
2068 }
2069 }
2070
get_loaded_modules_info(os::LoadedModulesCallbackFunc callback,void * param)2071 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2072 FILE *procmapsFile = NULL;
2073
2074 // Open the procfs maps file for the current process
2075 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2076 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2077 char line[PATH_MAX + 100];
2078
2079 // Read line by line from 'file'
2080 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2081 u8 base, top, offset, inode;
2082 char permissions[5];
2083 char device[6];
2084 char name[sizeof(line)];
2085
2086 // Parse fields from line
2087 int matches = sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
2088 &base, &top, permissions, &offset, device, &inode, name);
2089 // the last entry 'name' is empty for some entries, so we might have 6 matches instead of 7 for some lines
2090 if (matches < 6) continue;
2091 if (matches == 6) name[0] = '\0';
2092
2093 // Filter by device id '00:00' so that we only get file system mapped files.
2094 if (strcmp(device, "00:00") != 0) {
2095
2096 // Call callback with the fields of interest
2097 if(callback(name, (address)base, (address)top, param)) {
2098 // Oops abort, callback aborted
2099 fclose(procmapsFile);
2100 return 1;
2101 }
2102 }
2103 }
2104 fclose(procmapsFile);
2105 }
2106 return 0;
2107 }
2108
print_os_info_brief(outputStream * st)2109 void os::print_os_info_brief(outputStream* st) {
2110 os::Linux::print_distro_info(st);
2111
2112 os::Posix::print_uname_info(st);
2113
2114 os::Linux::print_libversion_info(st);
2115
2116 }
2117
print_os_info(outputStream * st)2118 void os::print_os_info(outputStream* st) {
2119 st->print("OS:");
2120
2121 os::Linux::print_distro_info(st);
2122
2123 os::Posix::print_uname_info(st);
2124
2125 os::Linux::print_uptime_info(st);
2126
2127 // Print warning if unsafe chroot environment detected
2128 if (unsafe_chroot_detected) {
2129 st->print("WARNING!! ");
2130 st->print_cr("%s", unstable_chroot_error);
2131 }
2132
2133 os::Linux::print_libversion_info(st);
2134
2135 os::Posix::print_rlimit_info(st);
2136
2137 os::Posix::print_load_average(st);
2138
2139 os::Linux::print_full_memory_info(st);
2140
2141 os::Linux::print_proc_sys_info(st);
2142
2143 os::Linux::print_ld_preload_file(st);
2144
2145 os::Linux::print_container_info(st);
2146
2147 VM_Version::print_platform_virtualization_info(st);
2148
2149 os::Linux::print_steal_info(st);
2150 }
2151
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2163 // search for redhat-release / SuSE-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort. The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort. It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173
2174 const char* distro_files[] = {
2175 "/etc/oracle-release",
2176 "/etc/mandriva-release",
2177 "/etc/mandrake-release",
2178 "/etc/sun-release",
2179 "/etc/redhat-release",
2180 "/etc/SuSE-release",
2181 "/etc/lsb-release",
2182 "/etc/turbolinux-release",
2183 "/etc/gentoo-release",
2184 "/etc/ltib-release",
2185 "/etc/angstrom-version",
2186 "/etc/system-release",
2187 "/etc/os-release",
2188 NULL };
2189
print_distro_info(outputStream * st)2190 void os::Linux::print_distro_info(outputStream* st) {
2191 for (int i = 0;; i++) {
2192 const char* file = distro_files[i];
2193 if (file == NULL) {
2194 break; // done
2195 }
2196 // If file prints, we found it.
2197 if (_print_ascii_file(file, st)) {
2198 return;
2199 }
2200 }
2201
2202 if (file_exists("/etc/debian_version")) {
2203 st->print("Debian ");
2204 _print_ascii_file("/etc/debian_version", st);
2205 } else {
2206 st->print("Linux");
2207 }
2208 st->cr();
2209 }
2210
parse_os_info_helper(FILE * fp,char * distro,size_t length,bool get_first_line)2211 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2212 char buf[256];
2213 while (fgets(buf, sizeof(buf), fp)) {
2214 // Edit out extra stuff in expected format
2215 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2216 char* ptr = strstr(buf, "\""); // the name is in quotes
2217 if (ptr != NULL) {
2218 ptr++; // go beyond first quote
2219 char* nl = strchr(ptr, '\"');
2220 if (nl != NULL) *nl = '\0';
2221 strncpy(distro, ptr, length);
2222 } else {
2223 ptr = strstr(buf, "=");
2224 ptr++; // go beyond equals then
2225 char* nl = strchr(ptr, '\n');
2226 if (nl != NULL) *nl = '\0';
2227 strncpy(distro, ptr, length);
2228 }
2229 return;
2230 } else if (get_first_line) {
2231 char* nl = strchr(buf, '\n');
2232 if (nl != NULL) *nl = '\0';
2233 strncpy(distro, buf, length);
2234 return;
2235 }
2236 }
2237 // print last line and close
2238 char* nl = strchr(buf, '\n');
2239 if (nl != NULL) *nl = '\0';
2240 strncpy(distro, buf, length);
2241 }
2242
parse_os_info(char * distro,size_t length,const char * file)2243 static void parse_os_info(char* distro, size_t length, const char* file) {
2244 FILE* fp = fopen(file, "r");
2245 if (fp != NULL) {
2246 // if suse format, print out first line
2247 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2248 parse_os_info_helper(fp, distro, length, get_first_line);
2249 fclose(fp);
2250 }
2251 }
2252
get_summary_os_info(char * buf,size_t buflen)2253 void os::get_summary_os_info(char* buf, size_t buflen) {
2254 for (int i = 0;; i++) {
2255 const char* file = distro_files[i];
2256 if (file == NULL) {
2257 break; // ran out of distro_files
2258 }
2259 if (file_exists(file)) {
2260 parse_os_info(buf, buflen, file);
2261 return;
2262 }
2263 }
2264 // special case for debian
2265 if (file_exists("/etc/debian_version")) {
2266 strncpy(buf, "Debian ", buflen);
2267 if (buflen > 7) {
2268 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2269 }
2270 } else {
2271 strncpy(buf, "Linux", buflen);
2272 }
2273 }
2274
print_libversion_info(outputStream * st)2275 void os::Linux::print_libversion_info(outputStream* st) {
2276 // libc, pthread
2277 st->print("libc:");
2278 st->print("%s ", os::Linux::glibc_version());
2279 st->print("%s ", os::Linux::libpthread_version());
2280 st->cr();
2281 }
2282
print_proc_sys_info(outputStream * st)2283 void os::Linux::print_proc_sys_info(outputStream* st) {
2284 st->cr();
2285 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2286 _print_ascii_file("/proc/sys/kernel/threads-max", st);
2287 st->cr();
2288 st->cr();
2289
2290 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2291 _print_ascii_file("/proc/sys/vm/max_map_count", st);
2292 st->cr();
2293 st->cr();
2294
2295 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2296 _print_ascii_file("/proc/sys/kernel/pid_max", st);
2297 st->cr();
2298 st->cr();
2299 }
2300
print_full_memory_info(outputStream * st)2301 void os::Linux::print_full_memory_info(outputStream* st) {
2302 st->print("\n/proc/meminfo:\n");
2303 _print_ascii_file("/proc/meminfo", st);
2304 st->cr();
2305 }
2306
print_ld_preload_file(outputStream * st)2307 void os::Linux::print_ld_preload_file(outputStream* st) {
2308 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2309 st->cr();
2310 }
2311
print_uptime_info(outputStream * st)2312 void os::Linux::print_uptime_info(outputStream* st) {
2313 struct sysinfo sinfo;
2314 int ret = sysinfo(&sinfo);
2315 if (ret == 0) {
2316 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime);
2317 }
2318 }
2319
2320
print_container_info(outputStream * st)2321 void os::Linux::print_container_info(outputStream* st) {
2322 if (!OSContainer::is_containerized()) {
2323 return;
2324 }
2325
2326 st->print("container (cgroup) information:\n");
2327
2328 const char *p_ct = OSContainer::container_type();
2329 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2330
2331 char *p = OSContainer::cpu_cpuset_cpus();
2332 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2333 free(p);
2334
2335 p = OSContainer::cpu_cpuset_memory_nodes();
2336 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2337 free(p);
2338
2339 int i = OSContainer::active_processor_count();
2340 st->print("active_processor_count: ");
2341 if (i > 0) {
2342 st->print("%d\n", i);
2343 } else {
2344 st->print("not supported\n");
2345 }
2346
2347 i = OSContainer::cpu_quota();
2348 st->print("cpu_quota: ");
2349 if (i > 0) {
2350 st->print("%d\n", i);
2351 } else {
2352 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2353 }
2354
2355 i = OSContainer::cpu_period();
2356 st->print("cpu_period: ");
2357 if (i > 0) {
2358 st->print("%d\n", i);
2359 } else {
2360 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2361 }
2362
2363 i = OSContainer::cpu_shares();
2364 st->print("cpu_shares: ");
2365 if (i > 0) {
2366 st->print("%d\n", i);
2367 } else {
2368 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2369 }
2370
2371 jlong j = OSContainer::memory_limit_in_bytes();
2372 st->print("memory_limit_in_bytes: ");
2373 if (j > 0) {
2374 st->print(JLONG_FORMAT "\n", j);
2375 } else {
2376 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2377 }
2378
2379 j = OSContainer::memory_and_swap_limit_in_bytes();
2380 st->print("memory_and_swap_limit_in_bytes: ");
2381 if (j > 0) {
2382 st->print(JLONG_FORMAT "\n", j);
2383 } else {
2384 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2385 }
2386
2387 j = OSContainer::memory_soft_limit_in_bytes();
2388 st->print("memory_soft_limit_in_bytes: ");
2389 if (j > 0) {
2390 st->print(JLONG_FORMAT "\n", j);
2391 } else {
2392 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2393 }
2394
2395 j = OSContainer::OSContainer::memory_usage_in_bytes();
2396 st->print("memory_usage_in_bytes: ");
2397 if (j > 0) {
2398 st->print(JLONG_FORMAT "\n", j);
2399 } else {
2400 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2401 }
2402
2403 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2404 st->print("memory_max_usage_in_bytes: ");
2405 if (j > 0) {
2406 st->print(JLONG_FORMAT "\n", j);
2407 } else {
2408 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2409 }
2410 st->cr();
2411 }
2412
print_steal_info(outputStream * st)2413 void os::Linux::print_steal_info(outputStream* st) {
2414 if (has_initial_tick_info) {
2415 CPUPerfTicks pticks;
2416 bool res = os::Linux::get_tick_information(&pticks, -1);
2417
2418 if (res && pticks.has_steal_ticks) {
2419 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2420 uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2421 double steal_ticks_perc = 0.0;
2422 if (total_ticks_difference != 0) {
2423 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2424 }
2425 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2426 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2427 }
2428 }
2429 }
2430
print_memory_info(outputStream * st)2431 void os::print_memory_info(outputStream* st) {
2432
2433 st->print("Memory:");
2434 st->print(" %dk page", os::vm_page_size()>>10);
2435
2436 // values in struct sysinfo are "unsigned long"
2437 struct sysinfo si;
2438 sysinfo(&si);
2439
2440 st->print(", physical " UINT64_FORMAT "k",
2441 os::physical_memory() >> 10);
2442 st->print("(" UINT64_FORMAT "k free)",
2443 os::available_memory() >> 10);
2444 st->print(", swap " UINT64_FORMAT "k",
2445 ((jlong)si.totalswap * si.mem_unit) >> 10);
2446 st->print("(" UINT64_FORMAT "k free)",
2447 ((jlong)si.freeswap * si.mem_unit) >> 10);
2448 st->cr();
2449 }
2450
2451 // Print the first "model name" line and the first "flags" line
2452 // that we find and nothing more. We assume "model name" comes
2453 // before "flags" so if we find a second "model name", then the
2454 // "flags" field is considered missing.
print_model_name_and_flags(outputStream * st,char * buf,size_t buflen)2455 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2456 #if defined(IA32) || defined(AMD64)
2457 // Other platforms have less repetitive cpuinfo files
2458 FILE *fp = fopen("/proc/cpuinfo", "r");
2459 if (fp) {
2460 while (!feof(fp)) {
2461 if (fgets(buf, buflen, fp)) {
2462 // Assume model name comes before flags
2463 bool model_name_printed = false;
2464 if (strstr(buf, "model name") != NULL) {
2465 if (!model_name_printed) {
2466 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2467 st->print_raw(buf);
2468 model_name_printed = true;
2469 } else {
2470 // model name printed but not flags? Odd, just return
2471 fclose(fp);
2472 return true;
2473 }
2474 }
2475 // print the flags line too
2476 if (strstr(buf, "flags") != NULL) {
2477 st->print_raw(buf);
2478 fclose(fp);
2479 return true;
2480 }
2481 }
2482 }
2483 fclose(fp);
2484 }
2485 #endif // x86 platforms
2486 return false;
2487 }
2488
pd_print_cpu_info(outputStream * st,char * buf,size_t buflen)2489 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2490 // Only print the model name if the platform provides this as a summary
2491 if (!print_model_name_and_flags(st, buf, buflen)) {
2492 st->print("\n/proc/cpuinfo:\n");
2493 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2494 st->print_cr(" <Not Available>");
2495 }
2496 }
2497 }
2498
2499 #if defined(AMD64) || defined(IA32) || defined(X32)
2500 const char* search_string = "model name";
2501 #elif defined(M68K)
2502 const char* search_string = "CPU";
2503 #elif defined(PPC64)
2504 const char* search_string = "cpu";
2505 #elif defined(S390)
2506 const char* search_string = "machine =";
2507 #elif defined(SPARC)
2508 const char* search_string = "cpu";
2509 #else
2510 const char* search_string = "Processor";
2511 #endif
2512
2513 // Parses the cpuinfo file for string representing the model name.
get_summary_cpu_info(char * cpuinfo,size_t length)2514 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2515 FILE* fp = fopen("/proc/cpuinfo", "r");
2516 if (fp != NULL) {
2517 while (!feof(fp)) {
2518 char buf[256];
2519 if (fgets(buf, sizeof(buf), fp)) {
2520 char* start = strstr(buf, search_string);
2521 if (start != NULL) {
2522 char *ptr = start + strlen(search_string);
2523 char *end = buf + strlen(buf);
2524 while (ptr != end) {
2525 // skip whitespace and colon for the rest of the name.
2526 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2527 break;
2528 }
2529 ptr++;
2530 }
2531 if (ptr != end) {
2532 // reasonable string, get rid of newline and keep the rest
2533 char* nl = strchr(buf, '\n');
2534 if (nl != NULL) *nl = '\0';
2535 strncpy(cpuinfo, ptr, length);
2536 fclose(fp);
2537 return;
2538 }
2539 }
2540 }
2541 }
2542 fclose(fp);
2543 }
2544 // cpuinfo not found or parsing failed, just print generic string. The entire
2545 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2546 #if defined(AARCH64)
2547 strncpy(cpuinfo, "AArch64", length);
2548 #elif defined(AMD64)
2549 strncpy(cpuinfo, "x86_64", length);
2550 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2551 strncpy(cpuinfo, "ARM", length);
2552 #elif defined(IA32)
2553 strncpy(cpuinfo, "x86_32", length);
2554 #elif defined(IA64)
2555 strncpy(cpuinfo, "IA64", length);
2556 #elif defined(PPC)
2557 strncpy(cpuinfo, "PPC64", length);
2558 #elif defined(S390)
2559 strncpy(cpuinfo, "S390", length);
2560 #elif defined(SPARC)
2561 strncpy(cpuinfo, "sparcv9", length);
2562 #elif defined(ZERO_LIBARCH)
2563 strncpy(cpuinfo, ZERO_LIBARCH, length);
2564 #else
2565 strncpy(cpuinfo, "unknown", length);
2566 #endif
2567 }
2568
2569 static void print_signal_handler(outputStream* st, int sig,
2570 char* buf, size_t buflen);
2571
print_signal_handlers(outputStream * st,char * buf,size_t buflen)2572 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2573 st->print_cr("Signal Handlers:");
2574 print_signal_handler(st, SIGSEGV, buf, buflen);
2575 print_signal_handler(st, SIGBUS , buf, buflen);
2576 print_signal_handler(st, SIGFPE , buf, buflen);
2577 print_signal_handler(st, SIGPIPE, buf, buflen);
2578 print_signal_handler(st, SIGXFSZ, buf, buflen);
2579 print_signal_handler(st, SIGILL , buf, buflen);
2580 print_signal_handler(st, SR_signum, buf, buflen);
2581 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2582 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2583 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2584 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2585 #if defined(PPC64)
2586 print_signal_handler(st, SIGTRAP, buf, buflen);
2587 #endif
2588 }
2589
2590 static char saved_jvm_path[MAXPATHLEN] = {0};
2591
2592 // Find the full path to the current module, libjvm.so
jvm_path(char * buf,jint buflen)2593 void os::jvm_path(char *buf, jint buflen) {
2594 // Error checking.
2595 if (buflen < MAXPATHLEN) {
2596 assert(false, "must use a large-enough buffer");
2597 buf[0] = '\0';
2598 return;
2599 }
2600 // Lazy resolve the path to current module.
2601 if (saved_jvm_path[0] != 0) {
2602 strcpy(buf, saved_jvm_path);
2603 return;
2604 }
2605
2606 char dli_fname[MAXPATHLEN];
2607 bool ret = dll_address_to_library_name(
2608 CAST_FROM_FN_PTR(address, os::jvm_path),
2609 dli_fname, sizeof(dli_fname), NULL);
2610 assert(ret, "cannot locate libjvm");
2611 char *rp = NULL;
2612 if (ret && dli_fname[0] != '\0') {
2613 rp = os::Posix::realpath(dli_fname, buf, buflen);
2614 }
2615 if (rp == NULL) {
2616 return;
2617 }
2618
2619 if (Arguments::sun_java_launcher_is_altjvm()) {
2620 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2621 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2622 // If "/jre/lib/" appears at the right place in the string, then
2623 // assume we are installed in a JDK and we're done. Otherwise, check
2624 // for a JAVA_HOME environment variable and fix up the path so it
2625 // looks like libjvm.so is installed there (append a fake suffix
2626 // hotspot/libjvm.so).
2627 const char *p = buf + strlen(buf) - 1;
2628 for (int count = 0; p > buf && count < 5; ++count) {
2629 for (--p; p > buf && *p != '/'; --p)
2630 /* empty */ ;
2631 }
2632
2633 if (strncmp(p, "/jre/lib/", 9) != 0) {
2634 // Look for JAVA_HOME in the environment.
2635 char* java_home_var = ::getenv("JAVA_HOME");
2636 if (java_home_var != NULL && java_home_var[0] != 0) {
2637 char* jrelib_p;
2638 int len;
2639
2640 // Check the current module name "libjvm.so".
2641 p = strrchr(buf, '/');
2642 if (p == NULL) {
2643 return;
2644 }
2645 assert(strstr(p, "/libjvm") == p, "invalid library name");
2646
2647 rp = os::Posix::realpath(java_home_var, buf, buflen);
2648 if (rp == NULL) {
2649 return;
2650 }
2651
2652 // determine if this is a legacy image or modules image
2653 // modules image doesn't have "jre" subdirectory
2654 len = strlen(buf);
2655 assert(len < buflen, "Ran out of buffer room");
2656 jrelib_p = buf + len;
2657 snprintf(jrelib_p, buflen-len, "/jre/lib");
2658 if (0 != access(buf, F_OK)) {
2659 snprintf(jrelib_p, buflen-len, "/lib");
2660 }
2661
2662 if (0 == access(buf, F_OK)) {
2663 // Use current module name "libjvm.so"
2664 len = strlen(buf);
2665 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2666 } else {
2667 // Go back to path of .so
2668 rp = os::Posix::realpath(dli_fname, buf, buflen);
2669 if (rp == NULL) {
2670 return;
2671 }
2672 }
2673 }
2674 }
2675 }
2676
2677 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2678 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2679 }
2680
print_jni_name_prefix_on(outputStream * st,int args_size)2681 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2682 // no prefix required, not even "_"
2683 }
2684
print_jni_name_suffix_on(outputStream * st,int args_size)2685 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2686 // no suffix required
2687 }
2688
2689 ////////////////////////////////////////////////////////////////////////////////
2690 // sun.misc.Signal support
2691
UserHandler(int sig,void * siginfo,void * context)2692 static void UserHandler(int sig, void *siginfo, void *context) {
2693 // Ctrl-C is pressed during error reporting, likely because the error
2694 // handler fails to abort. Let VM die immediately.
2695 if (sig == SIGINT && VMError::is_error_reported()) {
2696 os::die();
2697 }
2698
2699 os::signal_notify(sig);
2700 }
2701
user_handler()2702 void* os::user_handler() {
2703 return CAST_FROM_FN_PTR(void*, UserHandler);
2704 }
2705
2706 extern "C" {
2707 typedef void (*sa_handler_t)(int);
2708 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2709 }
2710
signal(int signal_number,void * handler)2711 void* os::signal(int signal_number, void* handler) {
2712 struct sigaction sigAct, oldSigAct;
2713
2714 sigfillset(&(sigAct.sa_mask));
2715 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2716 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2717
2718 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2719 // -1 means registration failed
2720 return (void *)-1;
2721 }
2722
2723 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2724 }
2725
signal_raise(int signal_number)2726 void os::signal_raise(int signal_number) {
2727 ::raise(signal_number);
2728 }
2729
2730 // The following code is moved from os.cpp for making this
2731 // code platform specific, which it is by its very nature.
2732
2733 // Will be modified when max signal is changed to be dynamic
sigexitnum_pd()2734 int os::sigexitnum_pd() {
2735 return NSIG;
2736 }
2737
2738 // a counter for each possible signal value
2739 static volatile jint pending_signals[NSIG+1] = { 0 };
2740
2741 // Linux(POSIX) specific hand shaking semaphore.
2742 static Semaphore* sig_sem = NULL;
2743 static PosixSemaphore sr_semaphore;
2744
jdk_misc_signal_init()2745 static void jdk_misc_signal_init() {
2746 // Initialize signal structures
2747 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2748
2749 // Initialize signal semaphore
2750 sig_sem = new Semaphore();
2751 }
2752
signal_notify(int sig)2753 void os::signal_notify(int sig) {
2754 if (sig_sem != NULL) {
2755 Atomic::inc(&pending_signals[sig]);
2756 sig_sem->signal();
2757 } else {
2758 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2759 // initialization isn't called.
2760 assert(ReduceSignalUsage, "signal semaphore should be created");
2761 }
2762 }
2763
check_pending_signals()2764 static int check_pending_signals() {
2765 for (;;) {
2766 for (int i = 0; i < NSIG + 1; i++) {
2767 jint n = pending_signals[i];
2768 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) {
2769 return i;
2770 }
2771 }
2772 JavaThread *thread = JavaThread::current();
2773 ThreadBlockInVM tbivm(thread);
2774
2775 bool threadIsSuspended;
2776 do {
2777 thread->set_suspend_equivalent();
2778 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2779 sig_sem->wait();
2780
2781 // were we externally suspended while we were waiting?
2782 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2783 if (threadIsSuspended) {
2784 // The semaphore has been incremented, but while we were waiting
2785 // another thread suspended us. We don't want to continue running
2786 // while suspended because that would surprise the thread that
2787 // suspended us.
2788 sig_sem->signal();
2789
2790 thread->java_suspend_self();
2791 }
2792 } while (threadIsSuspended);
2793 }
2794 }
2795
signal_wait()2796 int os::signal_wait() {
2797 return check_pending_signals();
2798 }
2799
2800 ////////////////////////////////////////////////////////////////////////////////
2801 // Virtual Memory
2802
vm_page_size()2803 int os::vm_page_size() {
2804 // Seems redundant as all get out
2805 assert(os::Linux::page_size() != -1, "must call os::init");
2806 return os::Linux::page_size();
2807 }
2808
2809 // Solaris allocates memory by pages.
vm_allocation_granularity()2810 int os::vm_allocation_granularity() {
2811 assert(os::Linux::page_size() != -1, "must call os::init");
2812 return os::Linux::page_size();
2813 }
2814
2815 // Rationale behind this function:
2816 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2817 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2818 // samples for JITted code. Here we create private executable mapping over the code cache
2819 // and then we can use standard (well, almost, as mapping can change) way to provide
2820 // info for the reporting script by storing timestamp and location of symbol
linux_wrap_code(char * base,size_t size)2821 void linux_wrap_code(char* base, size_t size) {
2822 static volatile jint cnt = 0;
2823
2824 if (!UseOprofile) {
2825 return;
2826 }
2827
2828 char buf[PATH_MAX+1];
2829 int num = Atomic::add(&cnt, 1);
2830
2831 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2832 os::get_temp_directory(), os::current_process_id(), num);
2833 unlink(buf);
2834
2835 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2836
2837 if (fd != -1) {
2838 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2839 if (rv != (off_t)-1) {
2840 if (::write(fd, "", 1) == 1) {
2841 mmap(base, size,
2842 PROT_READ|PROT_WRITE|PROT_EXEC,
2843 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2844 }
2845 }
2846 ::close(fd);
2847 unlink(buf);
2848 }
2849 }
2850
recoverable_mmap_error(int err)2851 static bool recoverable_mmap_error(int err) {
2852 // See if the error is one we can let the caller handle. This
2853 // list of errno values comes from JBS-6843484. I can't find a
2854 // Linux man page that documents this specific set of errno
2855 // values so while this list currently matches Solaris, it may
2856 // change as we gain experience with this failure mode.
2857 switch (err) {
2858 case EBADF:
2859 case EINVAL:
2860 case ENOTSUP:
2861 // let the caller deal with these errors
2862 return true;
2863
2864 default:
2865 // Any remaining errors on this OS can cause our reserved mapping
2866 // to be lost. That can cause confusion where different data
2867 // structures think they have the same memory mapped. The worst
2868 // scenario is if both the VM and a library think they have the
2869 // same memory mapped.
2870 return false;
2871 }
2872 }
2873
warn_fail_commit_memory(char * addr,size_t size,bool exec,int err)2874 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2875 int err) {
2876 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2877 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2878 os::strerror(err), err);
2879 }
2880
warn_fail_commit_memory(char * addr,size_t size,size_t alignment_hint,bool exec,int err)2881 static void warn_fail_commit_memory(char* addr, size_t size,
2882 size_t alignment_hint, bool exec,
2883 int err) {
2884 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2885 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2886 alignment_hint, exec, os::strerror(err), err);
2887 }
2888
2889 // NOTE: Linux kernel does not really reserve the pages for us.
2890 // All it does is to check if there are enough free pages
2891 // left at the time of mmap(). This could be a potential
2892 // problem.
commit_memory_impl(char * addr,size_t size,bool exec)2893 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2894 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2895 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2896 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2897 if (res != (uintptr_t) MAP_FAILED) {
2898 if (UseNUMAInterleaving) {
2899 numa_make_global(addr, size);
2900 }
2901 return 0;
2902 }
2903
2904 int err = errno; // save errno from mmap() call above
2905
2906 if (!recoverable_mmap_error(err)) {
2907 warn_fail_commit_memory(addr, size, exec, err);
2908 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2909 }
2910
2911 return err;
2912 }
2913
pd_commit_memory(char * addr,size_t size,bool exec)2914 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2915 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2916 }
2917
pd_commit_memory_or_exit(char * addr,size_t size,bool exec,const char * mesg)2918 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2919 const char* mesg) {
2920 assert(mesg != NULL, "mesg must be specified");
2921 int err = os::Linux::commit_memory_impl(addr, size, exec);
2922 if (err != 0) {
2923 // the caller wants all commit errors to exit with the specified mesg:
2924 warn_fail_commit_memory(addr, size, exec, err);
2925 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2926 }
2927 }
2928
2929 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2930 #ifndef MAP_HUGETLB
2931 #define MAP_HUGETLB 0x40000
2932 #endif
2933
2934 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2935 #ifndef MADV_HUGEPAGE
2936 #define MADV_HUGEPAGE 14
2937 #endif
2938
commit_memory_impl(char * addr,size_t size,size_t alignment_hint,bool exec)2939 int os::Linux::commit_memory_impl(char* addr, size_t size,
2940 size_t alignment_hint, bool exec) {
2941 int err = os::Linux::commit_memory_impl(addr, size, exec);
2942 if (err == 0) {
2943 realign_memory(addr, size, alignment_hint);
2944 }
2945 return err;
2946 }
2947
pd_commit_memory(char * addr,size_t size,size_t alignment_hint,bool exec)2948 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2949 bool exec) {
2950 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2951 }
2952
pd_commit_memory_or_exit(char * addr,size_t size,size_t alignment_hint,bool exec,const char * mesg)2953 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2954 size_t alignment_hint, bool exec,
2955 const char* mesg) {
2956 assert(mesg != NULL, "mesg must be specified");
2957 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2958 if (err != 0) {
2959 // the caller wants all commit errors to exit with the specified mesg:
2960 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2961 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2962 }
2963 }
2964
pd_realign_memory(char * addr,size_t bytes,size_t alignment_hint)2965 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2966 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2967 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2968 // be supported or the memory may already be backed by huge pages.
2969 ::madvise(addr, bytes, MADV_HUGEPAGE);
2970 }
2971 }
2972
pd_free_memory(char * addr,size_t bytes,size_t alignment_hint)2973 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2974 // This method works by doing an mmap over an existing mmaping and effectively discarding
2975 // the existing pages. However it won't work for SHM-based large pages that cannot be
2976 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2977 // small pages on top of the SHM segment. This method always works for small pages, so we
2978 // allow that in any case.
2979 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2980 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2981 }
2982 }
2983
numa_make_global(char * addr,size_t bytes)2984 void os::numa_make_global(char *addr, size_t bytes) {
2985 Linux::numa_interleave_memory(addr, bytes);
2986 }
2987
2988 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2989 // bind policy to MPOL_PREFERRED for the current thread.
2990 #define USE_MPOL_PREFERRED 0
2991
numa_make_local(char * addr,size_t bytes,int lgrp_hint)2992 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2993 // To make NUMA and large pages more robust when both enabled, we need to ease
2994 // the requirements on where the memory should be allocated. MPOL_BIND is the
2995 // default policy and it will force memory to be allocated on the specified
2996 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2997 // the specified node, but will not force it. Using this policy will prevent
2998 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2999 // free large pages.
3000 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
3001 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
3002 }
3003
numa_topology_changed()3004 bool os::numa_topology_changed() { return false; }
3005
numa_get_groups_num()3006 size_t os::numa_get_groups_num() {
3007 // Return just the number of nodes in which it's possible to allocate memory
3008 // (in numa terminology, configured nodes).
3009 return Linux::numa_num_configured_nodes();
3010 }
3011
numa_get_group_id()3012 int os::numa_get_group_id() {
3013 int cpu_id = Linux::sched_getcpu();
3014 if (cpu_id != -1) {
3015 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3016 if (lgrp_id != -1) {
3017 return lgrp_id;
3018 }
3019 }
3020 return 0;
3021 }
3022
numa_get_group_id_for_address(const void * address)3023 int os::numa_get_group_id_for_address(const void* address) {
3024 void** pages = const_cast<void**>(&address);
3025 int id = -1;
3026
3027 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) {
3028 return -1;
3029 }
3030 if (id < 0) {
3031 return -1;
3032 }
3033 return id;
3034 }
3035
get_existing_num_nodes()3036 int os::Linux::get_existing_num_nodes() {
3037 int node;
3038 int highest_node_number = Linux::numa_max_node();
3039 int num_nodes = 0;
3040
3041 // Get the total number of nodes in the system including nodes without memory.
3042 for (node = 0; node <= highest_node_number; node++) {
3043 if (is_node_in_existing_nodes(node)) {
3044 num_nodes++;
3045 }
3046 }
3047 return num_nodes;
3048 }
3049
numa_get_leaf_groups(int * ids,size_t size)3050 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3051 int highest_node_number = Linux::numa_max_node();
3052 size_t i = 0;
3053
3054 // Map all node ids in which it is possible to allocate memory. Also nodes are
3055 // not always consecutively available, i.e. available from 0 to the highest
3056 // node number. If the nodes have been bound explicitly using numactl membind,
3057 // then allocate memory from those nodes only.
3058 for (int node = 0; node <= highest_node_number; node++) {
3059 if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3060 ids[i++] = node;
3061 }
3062 }
3063 return i;
3064 }
3065
get_page_info(char * start,page_info * info)3066 bool os::get_page_info(char *start, page_info* info) {
3067 return false;
3068 }
3069
scan_pages(char * start,char * end,page_info * page_expected,page_info * page_found)3070 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3071 page_info* page_found) {
3072 return end;
3073 }
3074
3075
sched_getcpu_syscall(void)3076 int os::Linux::sched_getcpu_syscall(void) {
3077 unsigned int cpu = 0;
3078 int retval = -1;
3079
3080 #if defined(IA32)
3081 #ifndef SYS_getcpu
3082 #define SYS_getcpu 318
3083 #endif
3084 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3085 #elif defined(AMD64)
3086 // Unfortunately we have to bring all these macros here from vsyscall.h
3087 // to be able to compile on old linuxes.
3088 #define __NR_vgetcpu 2
3089 #define VSYSCALL_START (-10UL << 20)
3090 #define VSYSCALL_SIZE 1024
3091 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3092 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3093 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3094 retval = vgetcpu(&cpu, NULL, NULL);
3095 #endif
3096
3097 return (retval == -1) ? retval : cpu;
3098 }
3099
sched_getcpu_init()3100 void os::Linux::sched_getcpu_init() {
3101 // sched_getcpu() should be in libc.
3102 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3103 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3104
3105 // If it's not, try a direct syscall.
3106 if (sched_getcpu() == -1) {
3107 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3108 (void*)&sched_getcpu_syscall));
3109 }
3110
3111 if (sched_getcpu() == -1) {
3112 vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3113 }
3114 }
3115
3116 // Something to do with the numa-aware allocator needs these symbols
numa_warn(int number,char * where,...)3117 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
numa_error(char * where)3118 extern "C" JNIEXPORT void numa_error(char *where) { }
3119
3120 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3121 // load symbol from base version instead.
libnuma_dlsym(void * handle,const char * name)3122 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3123 void *f = dlvsym(handle, name, "libnuma_1.1");
3124 if (f == NULL) {
3125 f = dlsym(handle, name);
3126 }
3127 return f;
3128 }
3129
3130 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3131 // Return NULL if the symbol is not defined in this particular version.
libnuma_v2_dlsym(void * handle,const char * name)3132 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3133 return dlvsym(handle, name, "libnuma_1.2");
3134 }
3135
libnuma_init()3136 bool os::Linux::libnuma_init() {
3137 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3138 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3139 if (handle != NULL) {
3140 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3141 libnuma_dlsym(handle, "numa_node_to_cpus")));
3142 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3143 libnuma_dlsym(handle, "numa_max_node")));
3144 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3145 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3146 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3147 libnuma_dlsym(handle, "numa_available")));
3148 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3149 libnuma_dlsym(handle, "numa_tonode_memory")));
3150 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3151 libnuma_dlsym(handle, "numa_interleave_memory")));
3152 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3153 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3154 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3155 libnuma_dlsym(handle, "numa_set_bind_policy")));
3156 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3157 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3158 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3159 libnuma_dlsym(handle, "numa_distance")));
3160 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3161 libnuma_v2_dlsym(handle, "numa_get_membind")));
3162 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3163 libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3164 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t,
3165 libnuma_dlsym(handle, "numa_move_pages")));
3166
3167 if (numa_available() != -1) {
3168 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3169 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3170 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3171 set_numa_interleave_bitmask(_numa_get_interleave_mask());
3172 set_numa_membind_bitmask(_numa_get_membind());
3173 // Create an index -> node mapping, since nodes are not always consecutive
3174 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3175 rebuild_nindex_to_node_map();
3176 // Create a cpu -> node mapping
3177 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3178 rebuild_cpu_to_node_map();
3179 return true;
3180 }
3181 }
3182 }
3183 return false;
3184 }
3185
default_guard_size(os::ThreadType thr_type)3186 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3187 // Creating guard page is very expensive. Java thread has HotSpot
3188 // guard pages, only enable glibc guard page for non-Java threads.
3189 // (Remember: compiler thread is a Java thread, too!)
3190 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3191 }
3192
rebuild_nindex_to_node_map()3193 void os::Linux::rebuild_nindex_to_node_map() {
3194 int highest_node_number = Linux::numa_max_node();
3195
3196 nindex_to_node()->clear();
3197 for (int node = 0; node <= highest_node_number; node++) {
3198 if (Linux::is_node_in_existing_nodes(node)) {
3199 nindex_to_node()->append(node);
3200 }
3201 }
3202 }
3203
3204 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3205 // The table is later used in get_node_by_cpu().
rebuild_cpu_to_node_map()3206 void os::Linux::rebuild_cpu_to_node_map() {
3207 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3208 // in libnuma (possible values are starting from 16,
3209 // and continuing up with every other power of 2, but less
3210 // than the maximum number of CPUs supported by kernel), and
3211 // is a subject to change (in libnuma version 2 the requirements
3212 // are more reasonable) we'll just hardcode the number they use
3213 // in the library.
3214 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3215
3216 size_t cpu_num = processor_count();
3217 size_t cpu_map_size = NCPUS / BitsPerCLong;
3218 size_t cpu_map_valid_size =
3219 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3220
3221 cpu_to_node()->clear();
3222 cpu_to_node()->at_grow(cpu_num - 1);
3223
3224 size_t node_num = get_existing_num_nodes();
3225
3226 int distance = 0;
3227 int closest_distance = INT_MAX;
3228 int closest_node = 0;
3229 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3230 for (size_t i = 0; i < node_num; i++) {
3231 // Check if node is configured (not a memory-less node). If it is not, find
3232 // the closest configured node. Check also if node is bound, i.e. it's allowed
3233 // to allocate memory from the node. If it's not allowed, map cpus in that node
3234 // to the closest node from which memory allocation is allowed.
3235 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3236 !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3237 closest_distance = INT_MAX;
3238 // Check distance from all remaining nodes in the system. Ignore distance
3239 // from itself, from another non-configured node, and from another non-bound
3240 // node.
3241 for (size_t m = 0; m < node_num; m++) {
3242 if (m != i &&
3243 is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3244 is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3245 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3246 // If a closest node is found, update. There is always at least one
3247 // configured and bound node in the system so there is always at least
3248 // one node close.
3249 if (distance != 0 && distance < closest_distance) {
3250 closest_distance = distance;
3251 closest_node = nindex_to_node()->at(m);
3252 }
3253 }
3254 }
3255 } else {
3256 // Current node is already a configured node.
3257 closest_node = nindex_to_node()->at(i);
3258 }
3259
3260 // Get cpus from the original node and map them to the closest node. If node
3261 // is a configured node (not a memory-less node), then original node and
3262 // closest node are the same.
3263 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3264 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3265 if (cpu_map[j] != 0) {
3266 for (size_t k = 0; k < BitsPerCLong; k++) {
3267 if (cpu_map[j] & (1UL << k)) {
3268 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3269 }
3270 }
3271 }
3272 }
3273 }
3274 }
3275 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3276 }
3277
get_node_by_cpu(int cpu_id)3278 int os::Linux::get_node_by_cpu(int cpu_id) {
3279 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3280 return cpu_to_node()->at(cpu_id);
3281 }
3282 return -1;
3283 }
3284
3285 GrowableArray<int>* os::Linux::_cpu_to_node;
3286 GrowableArray<int>* os::Linux::_nindex_to_node;
3287 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3288 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3289 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3290 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3291 os::Linux::numa_available_func_t os::Linux::_numa_available;
3292 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3293 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3294 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3295 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3296 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3297 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3298 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3299 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3300 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages;
3301 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3302 unsigned long* os::Linux::_numa_all_nodes;
3303 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3304 struct bitmask* os::Linux::_numa_nodes_ptr;
3305 struct bitmask* os::Linux::_numa_interleave_bitmask;
3306 struct bitmask* os::Linux::_numa_membind_bitmask;
3307
pd_uncommit_memory(char * addr,size_t size)3308 bool os::pd_uncommit_memory(char* addr, size_t size) {
3309 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3310 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3311 return res != (uintptr_t) MAP_FAILED;
3312 }
3313
get_stack_commited_bottom(address bottom,size_t size)3314 static address get_stack_commited_bottom(address bottom, size_t size) {
3315 address nbot = bottom;
3316 address ntop = bottom + size;
3317
3318 size_t page_sz = os::vm_page_size();
3319 unsigned pages = size / page_sz;
3320
3321 unsigned char vec[1];
3322 unsigned imin = 1, imax = pages + 1, imid;
3323 int mincore_return_value = 0;
3324
3325 assert(imin <= imax, "Unexpected page size");
3326
3327 while (imin < imax) {
3328 imid = (imax + imin) / 2;
3329 nbot = ntop - (imid * page_sz);
3330
3331 // Use a trick with mincore to check whether the page is mapped or not.
3332 // mincore sets vec to 1 if page resides in memory and to 0 if page
3333 // is swapped output but if page we are asking for is unmapped
3334 // it returns -1,ENOMEM
3335 mincore_return_value = mincore(nbot, page_sz, vec);
3336
3337 if (mincore_return_value == -1) {
3338 // Page is not mapped go up
3339 // to find first mapped page
3340 if (errno != EAGAIN) {
3341 assert(errno == ENOMEM, "Unexpected mincore errno");
3342 imax = imid;
3343 }
3344 } else {
3345 // Page is mapped go down
3346 // to find first not mapped page
3347 imin = imid + 1;
3348 }
3349 }
3350
3351 nbot = nbot + page_sz;
3352
3353 // Adjust stack bottom one page up if last checked page is not mapped
3354 if (mincore_return_value == -1) {
3355 nbot = nbot + page_sz;
3356 }
3357
3358 return nbot;
3359 }
3360
committed_in_range(address start,size_t size,address & committed_start,size_t & committed_size)3361 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3362 int mincore_return_value;
3363 const size_t stripe = 1024; // query this many pages each time
3364 unsigned char vec[stripe + 1];
3365 // set a guard
3366 vec[stripe] = 'X';
3367
3368 const size_t page_sz = os::vm_page_size();
3369 size_t pages = size / page_sz;
3370
3371 assert(is_aligned(start, page_sz), "Start address must be page aligned");
3372 assert(is_aligned(size, page_sz), "Size must be page aligned");
3373
3374 committed_start = NULL;
3375
3376 int loops = (pages + stripe - 1) / stripe;
3377 int committed_pages = 0;
3378 address loop_base = start;
3379 bool found_range = false;
3380
3381 for (int index = 0; index < loops && !found_range; index ++) {
3382 assert(pages > 0, "Nothing to do");
3383 int pages_to_query = (pages >= stripe) ? stripe : pages;
3384 pages -= pages_to_query;
3385
3386 // Get stable read
3387 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3388
3389 // During shutdown, some memory goes away without properly notifying NMT,
3390 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3391 // Bailout and return as not committed for now.
3392 if (mincore_return_value == -1 && errno == ENOMEM) {
3393 return false;
3394 }
3395
3396 assert(vec[stripe] == 'X', "overflow guard");
3397 assert(mincore_return_value == 0, "Range must be valid");
3398 // Process this stripe
3399 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3400 if ((vec[vecIdx] & 0x01) == 0) { // not committed
3401 // End of current contiguous region
3402 if (committed_start != NULL) {
3403 found_range = true;
3404 break;
3405 }
3406 } else { // committed
3407 // Start of region
3408 if (committed_start == NULL) {
3409 committed_start = loop_base + page_sz * vecIdx;
3410 }
3411 committed_pages ++;
3412 }
3413 }
3414
3415 loop_base += pages_to_query * page_sz;
3416 }
3417
3418 if (committed_start != NULL) {
3419 assert(committed_pages > 0, "Must have committed region");
3420 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3421 assert(committed_start >= start && committed_start < start + size, "Out of range");
3422 committed_size = page_sz * committed_pages;
3423 return true;
3424 } else {
3425 assert(committed_pages == 0, "Should not have committed region");
3426 return false;
3427 }
3428 }
3429
3430
3431 // Linux uses a growable mapping for the stack, and if the mapping for
3432 // the stack guard pages is not removed when we detach a thread the
3433 // stack cannot grow beyond the pages where the stack guard was
3434 // mapped. If at some point later in the process the stack expands to
3435 // that point, the Linux kernel cannot expand the stack any further
3436 // because the guard pages are in the way, and a segfault occurs.
3437 //
3438 // However, it's essential not to split the stack region by unmapping
3439 // a region (leaving a hole) that's already part of the stack mapping,
3440 // so if the stack mapping has already grown beyond the guard pages at
3441 // the time we create them, we have to truncate the stack mapping.
3442 // So, we need to know the extent of the stack mapping when
3443 // create_stack_guard_pages() is called.
3444
3445 // We only need this for stacks that are growable: at the time of
3446 // writing thread stacks don't use growable mappings (i.e. those
3447 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3448 // only applies to the main thread.
3449
3450 // If the (growable) stack mapping already extends beyond the point
3451 // where we're going to put our guard pages, truncate the mapping at
3452 // that point by munmap()ping it. This ensures that when we later
3453 // munmap() the guard pages we don't leave a hole in the stack
3454 // mapping. This only affects the main/primordial thread
3455
pd_create_stack_guard_pages(char * addr,size_t size)3456 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3457 if (os::is_primordial_thread()) {
3458 // As we manually grow stack up to bottom inside create_attached_thread(),
3459 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3460 // we don't need to do anything special.
3461 // Check it first, before calling heavy function.
3462 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3463 unsigned char vec[1];
3464
3465 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3466 // Fallback to slow path on all errors, including EAGAIN
3467 stack_extent = (uintptr_t) get_stack_commited_bottom(
3468 os::Linux::initial_thread_stack_bottom(),
3469 (size_t)addr - stack_extent);
3470 }
3471
3472 if (stack_extent < (uintptr_t)addr) {
3473 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3474 }
3475 }
3476
3477 return os::commit_memory(addr, size, !ExecMem);
3478 }
3479
3480 // If this is a growable mapping, remove the guard pages entirely by
3481 // munmap()ping them. If not, just call uncommit_memory(). This only
3482 // affects the main/primordial thread, but guard against future OS changes.
3483 // It's safe to always unmap guard pages for primordial thread because we
3484 // always place it right after end of the mapped region.
3485
remove_stack_guard_pages(char * addr,size_t size)3486 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3487 uintptr_t stack_extent, stack_base;
3488
3489 if (os::is_primordial_thread()) {
3490 return ::munmap(addr, size) == 0;
3491 }
3492
3493 return os::uncommit_memory(addr, size);
3494 }
3495
3496 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3497 // at 'requested_addr'. If there are existing memory mappings at the same
3498 // location, however, they will be overwritten. If 'fixed' is false,
3499 // 'requested_addr' is only treated as a hint, the return value may or
3500 // may not start from the requested address. Unlike Linux mmap(), this
3501 // function returns NULL to indicate failure.
anon_mmap(char * requested_addr,size_t bytes,bool fixed)3502 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3503 char * addr;
3504 int flags;
3505
3506 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3507 if (fixed) {
3508 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3509 flags |= MAP_FIXED;
3510 }
3511
3512 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3513 // touch an uncommitted page. Otherwise, the read/write might
3514 // succeed if we have enough swap space to back the physical page.
3515 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3516 flags, -1, 0);
3517
3518 return addr == MAP_FAILED ? NULL : addr;
3519 }
3520
3521 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3522 // (req_addr != NULL) or with a given alignment.
3523 // - bytes shall be a multiple of alignment.
3524 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3525 // - alignment sets the alignment at which memory shall be allocated.
3526 // It must be a multiple of allocation granularity.
3527 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3528 // req_addr or NULL.
anon_mmap_aligned(size_t bytes,size_t alignment,char * req_addr)3529 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3530
3531 size_t extra_size = bytes;
3532 if (req_addr == NULL && alignment > 0) {
3533 extra_size += alignment;
3534 }
3535
3536 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3538 -1, 0);
3539 if (start == MAP_FAILED) {
3540 start = NULL;
3541 } else {
3542 if (req_addr != NULL) {
3543 if (start != req_addr) {
3544 ::munmap(start, extra_size);
3545 start = NULL;
3546 }
3547 } else {
3548 char* const start_aligned = align_up(start, alignment);
3549 char* const end_aligned = start_aligned + bytes;
3550 char* const end = start + extra_size;
3551 if (start_aligned > start) {
3552 ::munmap(start, start_aligned - start);
3553 }
3554 if (end_aligned < end) {
3555 ::munmap(end_aligned, end - end_aligned);
3556 }
3557 start = start_aligned;
3558 }
3559 }
3560 return start;
3561 }
3562
anon_munmap(char * addr,size_t size)3563 static int anon_munmap(char * addr, size_t size) {
3564 return ::munmap(addr, size) == 0;
3565 }
3566
pd_reserve_memory(size_t bytes,char * requested_addr,size_t alignment_hint)3567 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3568 size_t alignment_hint) {
3569 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3570 }
3571
pd_release_memory(char * addr,size_t size)3572 bool os::pd_release_memory(char* addr, size_t size) {
3573 return anon_munmap(addr, size);
3574 }
3575
linux_mprotect(char * addr,size_t size,int prot)3576 static bool linux_mprotect(char* addr, size_t size, int prot) {
3577 // Linux wants the mprotect address argument to be page aligned.
3578 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3579
3580 // According to SUSv3, mprotect() should only be used with mappings
3581 // established by mmap(), and mmap() always maps whole pages. Unaligned
3582 // 'addr' likely indicates problem in the VM (e.g. trying to change
3583 // protection of malloc'ed or statically allocated memory). Check the
3584 // caller if you hit this assert.
3585 assert(addr == bottom, "sanity check");
3586
3587 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3588 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3589 return ::mprotect(bottom, size, prot) == 0;
3590 }
3591
3592 // Set protections specified
protect_memory(char * addr,size_t bytes,ProtType prot,bool is_committed)3593 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3594 bool is_committed) {
3595 unsigned int p = 0;
3596 switch (prot) {
3597 case MEM_PROT_NONE: p = PROT_NONE; break;
3598 case MEM_PROT_READ: p = PROT_READ; break;
3599 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3600 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3601 default:
3602 ShouldNotReachHere();
3603 }
3604 // is_committed is unused.
3605 return linux_mprotect(addr, bytes, p);
3606 }
3607
guard_memory(char * addr,size_t size)3608 bool os::guard_memory(char* addr, size_t size) {
3609 return linux_mprotect(addr, size, PROT_NONE);
3610 }
3611
unguard_memory(char * addr,size_t size)3612 bool os::unguard_memory(char* addr, size_t size) {
3613 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3614 }
3615
transparent_huge_pages_sanity_check(bool warn,size_t page_size)3616 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3617 size_t page_size) {
3618 bool result = false;
3619 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3620 MAP_ANONYMOUS|MAP_PRIVATE,
3621 -1, 0);
3622 if (p != MAP_FAILED) {
3623 void *aligned_p = align_up(p, page_size);
3624
3625 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3626
3627 munmap(p, page_size * 2);
3628 }
3629
3630 if (warn && !result) {
3631 warning("TransparentHugePages is not supported by the operating system.");
3632 }
3633
3634 return result;
3635 }
3636
hugetlbfs_sanity_check(bool warn,size_t page_size)3637 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3638 bool result = false;
3639 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3640 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3641 -1, 0);
3642
3643 if (p != MAP_FAILED) {
3644 // We don't know if this really is a huge page or not.
3645 FILE *fp = fopen("/proc/self/maps", "r");
3646 if (fp) {
3647 while (!feof(fp)) {
3648 char chars[257];
3649 long x = 0;
3650 if (fgets(chars, sizeof(chars), fp)) {
3651 if (sscanf(chars, "%lx-%*x", &x) == 1
3652 && x == (long)p) {
3653 if (strstr (chars, "hugepage")) {
3654 result = true;
3655 break;
3656 }
3657 }
3658 }
3659 }
3660 fclose(fp);
3661 }
3662 munmap(p, page_size);
3663 }
3664
3665 if (warn && !result) {
3666 warning("HugeTLBFS is not supported by the operating system.");
3667 }
3668
3669 return result;
3670 }
3671
3672 // From the coredump_filter documentation:
3673 //
3674 // - (bit 0) anonymous private memory
3675 // - (bit 1) anonymous shared memory
3676 // - (bit 2) file-backed private memory
3677 // - (bit 3) file-backed shared memory
3678 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3679 // effective only if the bit 2 is cleared)
3680 // - (bit 5) hugetlb private memory
3681 // - (bit 6) hugetlb shared memory
3682 // - (bit 7) dax private memory
3683 // - (bit 8) dax shared memory
3684 //
set_coredump_filter(CoredumpFilterBit bit)3685 static void set_coredump_filter(CoredumpFilterBit bit) {
3686 FILE *f;
3687 long cdm;
3688
3689 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3690 return;
3691 }
3692
3693 if (fscanf(f, "%lx", &cdm) != 1) {
3694 fclose(f);
3695 return;
3696 }
3697
3698 long saved_cdm = cdm;
3699 rewind(f);
3700 cdm |= bit;
3701
3702 if (cdm != saved_cdm) {
3703 fprintf(f, "%#lx", cdm);
3704 }
3705
3706 fclose(f);
3707 }
3708
3709 // Large page support
3710
3711 static size_t _large_page_size = 0;
3712
find_large_page_size()3713 size_t os::Linux::find_large_page_size() {
3714 size_t large_page_size = 0;
3715
3716 // large_page_size on Linux is used to round up heap size. x86 uses either
3717 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3718 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3719 // page as large as 256M.
3720 //
3721 // Here we try to figure out page size by parsing /proc/meminfo and looking
3722 // for a line with the following format:
3723 // Hugepagesize: 2048 kB
3724 //
3725 // If we can't determine the value (e.g. /proc is not mounted, or the text
3726 // format has been changed), we'll use the largest page size supported by
3727 // the processor.
3728
3729 #ifndef ZERO
3730 large_page_size =
3731 AARCH64_ONLY(2 * M)
3732 AMD64_ONLY(2 * M)
3733 ARM32_ONLY(2 * M)
3734 IA32_ONLY(4 * M)
3735 IA64_ONLY(256 * M)
3736 PPC_ONLY(4 * M)
3737 S390_ONLY(1 * M)
3738 SPARC_ONLY(4 * M);
3739 #endif // ZERO
3740
3741 FILE *fp = fopen("/proc/meminfo", "r");
3742 if (fp) {
3743 while (!feof(fp)) {
3744 int x = 0;
3745 char buf[16];
3746 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3747 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3748 large_page_size = x * K;
3749 break;
3750 }
3751 } else {
3752 // skip to next line
3753 for (;;) {
3754 int ch = fgetc(fp);
3755 if (ch == EOF || ch == (int)'\n') break;
3756 }
3757 }
3758 }
3759 fclose(fp);
3760 }
3761
3762 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3763 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3764 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3765 proper_unit_for_byte_size(large_page_size));
3766 }
3767
3768 return large_page_size;
3769 }
3770
setup_large_page_size()3771 size_t os::Linux::setup_large_page_size() {
3772 _large_page_size = Linux::find_large_page_size();
3773 const size_t default_page_size = (size_t)Linux::page_size();
3774 if (_large_page_size > default_page_size) {
3775 _page_sizes[0] = _large_page_size;
3776 _page_sizes[1] = default_page_size;
3777 _page_sizes[2] = 0;
3778 }
3779
3780 return _large_page_size;
3781 }
3782
setup_large_page_type(size_t page_size)3783 bool os::Linux::setup_large_page_type(size_t page_size) {
3784 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3785 FLAG_IS_DEFAULT(UseSHM) &&
3786 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3787
3788 // The type of large pages has not been specified by the user.
3789
3790 // Try UseHugeTLBFS and then UseSHM.
3791 UseHugeTLBFS = UseSHM = true;
3792
3793 // Don't try UseTransparentHugePages since there are known
3794 // performance issues with it turned on. This might change in the future.
3795 UseTransparentHugePages = false;
3796 }
3797
3798 if (UseTransparentHugePages) {
3799 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3800 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3801 UseHugeTLBFS = false;
3802 UseSHM = false;
3803 return true;
3804 }
3805 UseTransparentHugePages = false;
3806 }
3807
3808 if (UseHugeTLBFS) {
3809 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3810 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3811 UseSHM = false;
3812 return true;
3813 }
3814 UseHugeTLBFS = false;
3815 }
3816
3817 return UseSHM;
3818 }
3819
large_page_init()3820 void os::large_page_init() {
3821 if (!UseLargePages &&
3822 !UseTransparentHugePages &&
3823 !UseHugeTLBFS &&
3824 !UseSHM) {
3825 // Not using large pages.
3826 return;
3827 }
3828
3829 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3830 // The user explicitly turned off large pages.
3831 // Ignore the rest of the large pages flags.
3832 UseTransparentHugePages = false;
3833 UseHugeTLBFS = false;
3834 UseSHM = false;
3835 return;
3836 }
3837
3838 size_t large_page_size = Linux::setup_large_page_size();
3839 UseLargePages = Linux::setup_large_page_type(large_page_size);
3840
3841 set_coredump_filter(LARGEPAGES_BIT);
3842 }
3843
3844 #ifndef SHM_HUGETLB
3845 #define SHM_HUGETLB 04000
3846 #endif
3847
3848 #define shm_warning_format(format, ...) \
3849 do { \
3850 if (UseLargePages && \
3851 (!FLAG_IS_DEFAULT(UseLargePages) || \
3852 !FLAG_IS_DEFAULT(UseSHM) || \
3853 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3854 warning(format, __VA_ARGS__); \
3855 } \
3856 } while (0)
3857
3858 #define shm_warning(str) shm_warning_format("%s", str)
3859
3860 #define shm_warning_with_errno(str) \
3861 do { \
3862 int err = errno; \
3863 shm_warning_format(str " (error = %d)", err); \
3864 } while (0)
3865
shmat_with_alignment(int shmid,size_t bytes,size_t alignment)3866 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3867 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3868
3869 if (!is_aligned(alignment, SHMLBA)) {
3870 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3871 return NULL;
3872 }
3873
3874 // To ensure that we get 'alignment' aligned memory from shmat,
3875 // we pre-reserve aligned virtual memory and then attach to that.
3876
3877 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3878 if (pre_reserved_addr == NULL) {
3879 // Couldn't pre-reserve aligned memory.
3880 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3881 return NULL;
3882 }
3883
3884 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3885 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3886
3887 if ((intptr_t)addr == -1) {
3888 int err = errno;
3889 shm_warning_with_errno("Failed to attach shared memory.");
3890
3891 assert(err != EACCES, "Unexpected error");
3892 assert(err != EIDRM, "Unexpected error");
3893 assert(err != EINVAL, "Unexpected error");
3894
3895 // Since we don't know if the kernel unmapped the pre-reserved memory area
3896 // we can't unmap it, since that would potentially unmap memory that was
3897 // mapped from other threads.
3898 return NULL;
3899 }
3900
3901 return addr;
3902 }
3903
shmat_at_address(int shmid,char * req_addr)3904 static char* shmat_at_address(int shmid, char* req_addr) {
3905 if (!is_aligned(req_addr, SHMLBA)) {
3906 assert(false, "Requested address needs to be SHMLBA aligned");
3907 return NULL;
3908 }
3909
3910 char* addr = (char*)shmat(shmid, req_addr, 0);
3911
3912 if ((intptr_t)addr == -1) {
3913 shm_warning_with_errno("Failed to attach shared memory.");
3914 return NULL;
3915 }
3916
3917 return addr;
3918 }
3919
shmat_large_pages(int shmid,size_t bytes,size_t alignment,char * req_addr)3920 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3921 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3922 if (req_addr != NULL) {
3923 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3924 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3925 return shmat_at_address(shmid, req_addr);
3926 }
3927
3928 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3929 // return large page size aligned memory addresses when req_addr == NULL.
3930 // However, if the alignment is larger than the large page size, we have
3931 // to manually ensure that the memory returned is 'alignment' aligned.
3932 if (alignment > os::large_page_size()) {
3933 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3934 return shmat_with_alignment(shmid, bytes, alignment);
3935 } else {
3936 return shmat_at_address(shmid, NULL);
3937 }
3938 }
3939
reserve_memory_special_shm(size_t bytes,size_t alignment,char * req_addr,bool exec)3940 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3941 char* req_addr, bool exec) {
3942 // "exec" is passed in but not used. Creating the shared image for
3943 // the code cache doesn't have an SHM_X executable permission to check.
3944 assert(UseLargePages && UseSHM, "only for SHM large pages");
3945 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3946 assert(is_aligned(req_addr, alignment), "Unaligned address");
3947
3948 if (!is_aligned(bytes, os::large_page_size())) {
3949 return NULL; // Fallback to small pages.
3950 }
3951
3952 // Create a large shared memory region to attach to based on size.
3953 // Currently, size is the total size of the heap.
3954 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3955 if (shmid == -1) {
3956 // Possible reasons for shmget failure:
3957 // 1. shmmax is too small for Java heap.
3958 // > check shmmax value: cat /proc/sys/kernel/shmmax
3959 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3960 // 2. not enough large page memory.
3961 // > check available large pages: cat /proc/meminfo
3962 // > increase amount of large pages:
3963 // echo new_value > /proc/sys/vm/nr_hugepages
3964 // Note 1: different Linux may use different name for this property,
3965 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3966 // Note 2: it's possible there's enough physical memory available but
3967 // they are so fragmented after a long run that they can't
3968 // coalesce into large pages. Try to reserve large pages when
3969 // the system is still "fresh".
3970 shm_warning_with_errno("Failed to reserve shared memory.");
3971 return NULL;
3972 }
3973
3974 // Attach to the region.
3975 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3976
3977 // Remove shmid. If shmat() is successful, the actual shared memory segment
3978 // will be deleted when it's detached by shmdt() or when the process
3979 // terminates. If shmat() is not successful this will remove the shared
3980 // segment immediately.
3981 shmctl(shmid, IPC_RMID, NULL);
3982
3983 return addr;
3984 }
3985
warn_on_large_pages_failure(char * req_addr,size_t bytes,int error)3986 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3987 int error) {
3988 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3989
3990 bool warn_on_failure = UseLargePages &&
3991 (!FLAG_IS_DEFAULT(UseLargePages) ||
3992 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3993 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3994
3995 if (warn_on_failure) {
3996 char msg[128];
3997 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3998 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3999 warning("%s", msg);
4000 }
4001 }
4002
reserve_memory_special_huge_tlbfs_only(size_t bytes,char * req_addr,bool exec)4003 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
4004 char* req_addr,
4005 bool exec) {
4006 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4007 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4008 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4009
4010 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4011 char* addr = (char*)::mmap(req_addr, bytes, prot,
4012 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
4013 -1, 0);
4014
4015 if (addr == MAP_FAILED) {
4016 warn_on_large_pages_failure(req_addr, bytes, errno);
4017 return NULL;
4018 }
4019
4020 assert(is_aligned(addr, os::large_page_size()), "Must be");
4021
4022 return addr;
4023 }
4024
4025 // Reserve memory using mmap(MAP_HUGETLB).
4026 // - bytes shall be a multiple of alignment.
4027 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4028 // - alignment sets the alignment at which memory shall be allocated.
4029 // It must be a multiple of allocation granularity.
4030 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4031 // req_addr or NULL.
reserve_memory_special_huge_tlbfs_mixed(size_t bytes,size_t alignment,char * req_addr,bool exec)4032 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4033 size_t alignment,
4034 char* req_addr,
4035 bool exec) {
4036 size_t large_page_size = os::large_page_size();
4037 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4038
4039 assert(is_aligned(req_addr, alignment), "Must be");
4040 assert(is_aligned(bytes, alignment), "Must be");
4041
4042 // First reserve - but not commit - the address range in small pages.
4043 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4044
4045 if (start == NULL) {
4046 return NULL;
4047 }
4048
4049 assert(is_aligned(start, alignment), "Must be");
4050
4051 char* end = start + bytes;
4052
4053 // Find the regions of the allocated chunk that can be promoted to large pages.
4054 char* lp_start = align_up(start, large_page_size);
4055 char* lp_end = align_down(end, large_page_size);
4056
4057 size_t lp_bytes = lp_end - lp_start;
4058
4059 assert(is_aligned(lp_bytes, large_page_size), "Must be");
4060
4061 if (lp_bytes == 0) {
4062 // The mapped region doesn't even span the start and the end of a large page.
4063 // Fall back to allocate a non-special area.
4064 ::munmap(start, end - start);
4065 return NULL;
4066 }
4067
4068 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4069
4070 void* result;
4071
4072 // Commit small-paged leading area.
4073 if (start != lp_start) {
4074 result = ::mmap(start, lp_start - start, prot,
4075 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4076 -1, 0);
4077 if (result == MAP_FAILED) {
4078 ::munmap(lp_start, end - lp_start);
4079 return NULL;
4080 }
4081 }
4082
4083 // Commit large-paged area.
4084 result = ::mmap(lp_start, lp_bytes, prot,
4085 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4086 -1, 0);
4087 if (result == MAP_FAILED) {
4088 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4089 // If the mmap above fails, the large pages region will be unmapped and we
4090 // have regions before and after with small pages. Release these regions.
4091 //
4092 // | mapped | unmapped | mapped |
4093 // ^ ^ ^ ^
4094 // start lp_start lp_end end
4095 //
4096 ::munmap(start, lp_start - start);
4097 ::munmap(lp_end, end - lp_end);
4098 return NULL;
4099 }
4100
4101 // Commit small-paged trailing area.
4102 if (lp_end != end) {
4103 result = ::mmap(lp_end, end - lp_end, prot,
4104 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4105 -1, 0);
4106 if (result == MAP_FAILED) {
4107 ::munmap(start, lp_end - start);
4108 return NULL;
4109 }
4110 }
4111
4112 return start;
4113 }
4114
reserve_memory_special_huge_tlbfs(size_t bytes,size_t alignment,char * req_addr,bool exec)4115 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4116 size_t alignment,
4117 char* req_addr,
4118 bool exec) {
4119 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4120 assert(is_aligned(req_addr, alignment), "Must be");
4121 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4122 assert(is_power_of_2(os::large_page_size()), "Must be");
4123 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4124
4125 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4126 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4127 } else {
4128 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4129 }
4130 }
4131
reserve_memory_special(size_t bytes,size_t alignment,char * req_addr,bool exec)4132 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4133 char* req_addr, bool exec) {
4134 assert(UseLargePages, "only for large pages");
4135
4136 char* addr;
4137 if (UseSHM) {
4138 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4139 } else {
4140 assert(UseHugeTLBFS, "must be");
4141 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4142 }
4143
4144 if (addr != NULL) {
4145 if (UseNUMAInterleaving) {
4146 numa_make_global(addr, bytes);
4147 }
4148
4149 // The memory is committed
4150 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4151 }
4152
4153 return addr;
4154 }
4155
release_memory_special_shm(char * base,size_t bytes)4156 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4157 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4158 return shmdt(base) == 0;
4159 }
4160
release_memory_special_huge_tlbfs(char * base,size_t bytes)4161 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4162 return pd_release_memory(base, bytes);
4163 }
4164
release_memory_special(char * base,size_t bytes)4165 bool os::release_memory_special(char* base, size_t bytes) {
4166 bool res;
4167 if (MemTracker::tracking_level() > NMT_minimal) {
4168 Tracker tkr(Tracker::release);
4169 res = os::Linux::release_memory_special_impl(base, bytes);
4170 if (res) {
4171 tkr.record((address)base, bytes);
4172 }
4173
4174 } else {
4175 res = os::Linux::release_memory_special_impl(base, bytes);
4176 }
4177 return res;
4178 }
4179
release_memory_special_impl(char * base,size_t bytes)4180 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4181 assert(UseLargePages, "only for large pages");
4182 bool res;
4183
4184 if (UseSHM) {
4185 res = os::Linux::release_memory_special_shm(base, bytes);
4186 } else {
4187 assert(UseHugeTLBFS, "must be");
4188 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4189 }
4190 return res;
4191 }
4192
large_page_size()4193 size_t os::large_page_size() {
4194 return _large_page_size;
4195 }
4196
4197 // With SysV SHM the entire memory region must be allocated as shared
4198 // memory.
4199 // HugeTLBFS allows application to commit large page memory on demand.
4200 // However, when committing memory with HugeTLBFS fails, the region
4201 // that was supposed to be committed will lose the old reservation
4202 // and allow other threads to steal that memory region. Because of this
4203 // behavior we can't commit HugeTLBFS memory.
can_commit_large_page_memory()4204 bool os::can_commit_large_page_memory() {
4205 return UseTransparentHugePages;
4206 }
4207
can_execute_large_page_memory()4208 bool os::can_execute_large_page_memory() {
4209 return UseTransparentHugePages || UseHugeTLBFS;
4210 }
4211
pd_attempt_reserve_memory_at(size_t bytes,char * requested_addr,int file_desc)4212 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4213 assert(file_desc >= 0, "file_desc is not valid");
4214 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4215 if (result != NULL) {
4216 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4217 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4218 }
4219 }
4220 return result;
4221 }
4222
4223 // Reserve memory at an arbitrary address, only if that area is
4224 // available (and not reserved for something else).
4225
pd_attempt_reserve_memory_at(size_t bytes,char * requested_addr)4226 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4227 // Assert only that the size is a multiple of the page size, since
4228 // that's all that mmap requires, and since that's all we really know
4229 // about at this low abstraction level. If we need higher alignment,
4230 // we can either pass an alignment to this method or verify alignment
4231 // in one of the methods further up the call chain. See bug 5044738.
4232 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4233
4234 // Repeatedly allocate blocks until the block is allocated at the
4235 // right spot.
4236
4237 // Linux mmap allows caller to pass an address as hint; give it a try first,
4238 // if kernel honors the hint then we can return immediately.
4239 char * addr = anon_mmap(requested_addr, bytes, false);
4240 if (addr == requested_addr) {
4241 return requested_addr;
4242 }
4243
4244 if (addr != NULL) {
4245 // mmap() is successful but it fails to reserve at the requested address
4246 anon_munmap(addr, bytes);
4247 }
4248
4249 return NULL;
4250 }
4251
4252 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
infinite_sleep()4253 void os::infinite_sleep() {
4254 while (true) { // sleep forever ...
4255 ::sleep(100); // ... 100 seconds at a time
4256 }
4257 }
4258
4259 // Used to convert frequent JVM_Yield() to nops
dont_yield()4260 bool os::dont_yield() {
4261 return DontYieldALot;
4262 }
4263
4264 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4265 // actually give up the CPU. Since skip buddy (v2.6.28):
4266 //
4267 // * Sets the yielding task as skip buddy for current CPU's run queue.
4268 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4269 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4270 //
4271 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4272 // getting re-scheduled immediately.
4273 //
naked_yield()4274 void os::naked_yield() {
4275 sched_yield();
4276 }
4277
4278 ////////////////////////////////////////////////////////////////////////////////
4279 // thread priority support
4280
4281 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4282 // only supports dynamic priority, static priority must be zero. For real-time
4283 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4284 // However, for large multi-threaded applications, SCHED_RR is not only slower
4285 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4286 // of 5 runs - Sep 2005).
4287 //
4288 // The following code actually changes the niceness of kernel-thread/LWP. It
4289 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4290 // not the entire user process, and user level threads are 1:1 mapped to kernel
4291 // threads. It has always been the case, but could change in the future. For
4292 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4293 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4294 // (e.g., root privilege or CAP_SYS_NICE capability).
4295
4296 int os::java_to_os_priority[CriticalPriority + 1] = {
4297 19, // 0 Entry should never be used
4298
4299 4, // 1 MinPriority
4300 3, // 2
4301 2, // 3
4302
4303 1, // 4
4304 0, // 5 NormPriority
4305 -1, // 6
4306
4307 -2, // 7
4308 -3, // 8
4309 -4, // 9 NearMaxPriority
4310
4311 -5, // 10 MaxPriority
4312
4313 -5 // 11 CriticalPriority
4314 };
4315
prio_init()4316 static int prio_init() {
4317 if (ThreadPriorityPolicy == 1) {
4318 if (geteuid() != 0) {
4319 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4320 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4321 "e.g., being the root user. If the necessary permission is not " \
4322 "possessed, changes to priority will be silently ignored.");
4323 }
4324 }
4325 }
4326 if (UseCriticalJavaThreadPriority) {
4327 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4328 }
4329 return 0;
4330 }
4331
set_native_priority(Thread * thread,int newpri)4332 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4333 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4334
4335 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4336 return (ret == 0) ? OS_OK : OS_ERR;
4337 }
4338
get_native_priority(const Thread * const thread,int * priority_ptr)4339 OSReturn os::get_native_priority(const Thread* const thread,
4340 int *priority_ptr) {
4341 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4342 *priority_ptr = java_to_os_priority[NormPriority];
4343 return OS_OK;
4344 }
4345
4346 errno = 0;
4347 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4348 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4349 }
4350
4351 ////////////////////////////////////////////////////////////////////////////////
4352 // suspend/resume support
4353
4354 // The low-level signal-based suspend/resume support is a remnant from the
4355 // old VM-suspension that used to be for java-suspension, safepoints etc,
4356 // within hotspot. Currently used by JFR's OSThreadSampler
4357 //
4358 // The remaining code is greatly simplified from the more general suspension
4359 // code that used to be used.
4360 //
4361 // The protocol is quite simple:
4362 // - suspend:
4363 // - sends a signal to the target thread
4364 // - polls the suspend state of the osthread using a yield loop
4365 // - target thread signal handler (SR_handler) sets suspend state
4366 // and blocks in sigsuspend until continued
4367 // - resume:
4368 // - sets target osthread state to continue
4369 // - sends signal to end the sigsuspend loop in the SR_handler
4370 //
4371 // Note that the SR_lock plays no role in this suspend/resume protocol,
4372 // but is checked for NULL in SR_handler as a thread termination indicator.
4373 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4374 //
4375 // Note that resume_clear_context() and suspend_save_context() are needed
4376 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4377 // which in part is used by:
4378 // - Forte Analyzer: AsyncGetCallTrace()
4379 // - StackBanging: get_frame_at_stack_banging_point()
4380
resume_clear_context(OSThread * osthread)4381 static void resume_clear_context(OSThread *osthread) {
4382 osthread->set_ucontext(NULL);
4383 osthread->set_siginfo(NULL);
4384 }
4385
suspend_save_context(OSThread * osthread,siginfo_t * siginfo,ucontext_t * context)4386 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4387 ucontext_t* context) {
4388 osthread->set_ucontext(context);
4389 osthread->set_siginfo(siginfo);
4390 }
4391
4392 // Handler function invoked when a thread's execution is suspended or
4393 // resumed. We have to be careful that only async-safe functions are
4394 // called here (Note: most pthread functions are not async safe and
4395 // should be avoided.)
4396 //
4397 // Note: sigwait() is a more natural fit than sigsuspend() from an
4398 // interface point of view, but sigwait() prevents the signal hander
4399 // from being run. libpthread would get very confused by not having
4400 // its signal handlers run and prevents sigwait()'s use with the
4401 // mutex granting granting signal.
4402 //
4403 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4404 //
SR_handler(int sig,siginfo_t * siginfo,ucontext_t * context)4405 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4406 // Save and restore errno to avoid confusing native code with EINTR
4407 // after sigsuspend.
4408 int old_errno = errno;
4409
4410 Thread* thread = Thread::current_or_null_safe();
4411 assert(thread != NULL, "Missing current thread in SR_handler");
4412
4413 // On some systems we have seen signal delivery get "stuck" until the signal
4414 // mask is changed as part of thread termination. Check that the current thread
4415 // has not already terminated (via SR_lock()) - else the following assertion
4416 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4417 // destructor has completed.
4418
4419 if (thread->SR_lock() == NULL) {
4420 return;
4421 }
4422
4423 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4424
4425 OSThread* osthread = thread->osthread();
4426
4427 os::SuspendResume::State current = osthread->sr.state();
4428 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4429 suspend_save_context(osthread, siginfo, context);
4430
4431 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4432 os::SuspendResume::State state = osthread->sr.suspended();
4433 if (state == os::SuspendResume::SR_SUSPENDED) {
4434 sigset_t suspend_set; // signals for sigsuspend()
4435 sigemptyset(&suspend_set);
4436 // get current set of blocked signals and unblock resume signal
4437 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4438 sigdelset(&suspend_set, SR_signum);
4439
4440 sr_semaphore.signal();
4441 // wait here until we are resumed
4442 while (1) {
4443 sigsuspend(&suspend_set);
4444
4445 os::SuspendResume::State result = osthread->sr.running();
4446 if (result == os::SuspendResume::SR_RUNNING) {
4447 sr_semaphore.signal();
4448 break;
4449 }
4450 }
4451
4452 } else if (state == os::SuspendResume::SR_RUNNING) {
4453 // request was cancelled, continue
4454 } else {
4455 ShouldNotReachHere();
4456 }
4457
4458 resume_clear_context(osthread);
4459 } else if (current == os::SuspendResume::SR_RUNNING) {
4460 // request was cancelled, continue
4461 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4462 // ignore
4463 } else {
4464 // ignore
4465 }
4466
4467 errno = old_errno;
4468 }
4469
SR_initialize()4470 static int SR_initialize() {
4471 struct sigaction act;
4472 char *s;
4473
4474 // Get signal number to use for suspend/resume
4475 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4476 int sig = ::strtol(s, 0, 10);
4477 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4478 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4479 SR_signum = sig;
4480 } else {
4481 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4482 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4483 }
4484 }
4485
4486 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4487 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4488
4489 sigemptyset(&SR_sigset);
4490 sigaddset(&SR_sigset, SR_signum);
4491
4492 // Set up signal handler for suspend/resume
4493 act.sa_flags = SA_RESTART|SA_SIGINFO;
4494 act.sa_handler = (void (*)(int)) SR_handler;
4495
4496 // SR_signum is blocked by default.
4497 // 4528190 - We also need to block pthread restart signal (32 on all
4498 // supported Linux platforms). Note that LinuxThreads need to block
4499 // this signal for all threads to work properly. So we don't have
4500 // to use hard-coded signal number when setting up the mask.
4501 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4502
4503 if (sigaction(SR_signum, &act, 0) == -1) {
4504 return -1;
4505 }
4506
4507 // Save signal flag
4508 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4509 return 0;
4510 }
4511
sr_notify(OSThread * osthread)4512 static int sr_notify(OSThread* osthread) {
4513 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4514 assert_status(status == 0, status, "pthread_kill");
4515 return status;
4516 }
4517
4518 // "Randomly" selected value for how long we want to spin
4519 // before bailing out on suspending a thread, also how often
4520 // we send a signal to a thread we want to resume
4521 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4522 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4523
4524 // returns true on success and false on error - really an error is fatal
4525 // but this seems the normal response to library errors
do_suspend(OSThread * osthread)4526 static bool do_suspend(OSThread* osthread) {
4527 assert(osthread->sr.is_running(), "thread should be running");
4528 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4529
4530 // mark as suspended and send signal
4531 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4532 // failed to switch, state wasn't running?
4533 ShouldNotReachHere();
4534 return false;
4535 }
4536
4537 if (sr_notify(osthread) != 0) {
4538 ShouldNotReachHere();
4539 }
4540
4541 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4542 while (true) {
4543 if (sr_semaphore.timedwait(2)) {
4544 break;
4545 } else {
4546 // timeout
4547 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4548 if (cancelled == os::SuspendResume::SR_RUNNING) {
4549 return false;
4550 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4551 // make sure that we consume the signal on the semaphore as well
4552 sr_semaphore.wait();
4553 break;
4554 } else {
4555 ShouldNotReachHere();
4556 return false;
4557 }
4558 }
4559 }
4560
4561 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4562 return true;
4563 }
4564
do_resume(OSThread * osthread)4565 static void do_resume(OSThread* osthread) {
4566 assert(osthread->sr.is_suspended(), "thread should be suspended");
4567 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4568
4569 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4570 // failed to switch to WAKEUP_REQUEST
4571 ShouldNotReachHere();
4572 return;
4573 }
4574
4575 while (true) {
4576 if (sr_notify(osthread) == 0) {
4577 if (sr_semaphore.timedwait(2)) {
4578 if (osthread->sr.is_running()) {
4579 return;
4580 }
4581 }
4582 } else {
4583 ShouldNotReachHere();
4584 }
4585 }
4586
4587 guarantee(osthread->sr.is_running(), "Must be running!");
4588 }
4589
4590 ///////////////////////////////////////////////////////////////////////////////////
4591 // signal handling (except suspend/resume)
4592
4593 // This routine may be used by user applications as a "hook" to catch signals.
4594 // The user-defined signal handler must pass unrecognized signals to this
4595 // routine, and if it returns true (non-zero), then the signal handler must
4596 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4597 // routine will never retun false (zero), but instead will execute a VM panic
4598 // routine kill the process.
4599 //
4600 // If this routine returns false, it is OK to call it again. This allows
4601 // the user-defined signal handler to perform checks either before or after
4602 // the VM performs its own checks. Naturally, the user code would be making
4603 // a serious error if it tried to handle an exception (such as a null check
4604 // or breakpoint) that the VM was generating for its own correct operation.
4605 //
4606 // This routine may recognize any of the following kinds of signals:
4607 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4608 // It should be consulted by handlers for any of those signals.
4609 //
4610 // The caller of this routine must pass in the three arguments supplied
4611 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4612 // field of the structure passed to sigaction(). This routine assumes that
4613 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4614 //
4615 // Note that the VM will print warnings if it detects conflicting signal
4616 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4617 //
4618 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4619 siginfo_t* siginfo,
4620 void* ucontext,
4621 int abort_if_unrecognized);
4622
signalHandler(int sig,siginfo_t * info,void * uc)4623 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4624 assert(info != NULL && uc != NULL, "it must be old kernel");
4625 int orig_errno = errno; // Preserve errno value over signal handler.
4626 JVM_handle_linux_signal(sig, info, uc, true);
4627 errno = orig_errno;
4628 }
4629
4630
4631 // This boolean allows users to forward their own non-matching signals
4632 // to JVM_handle_linux_signal, harmlessly.
4633 bool os::Linux::signal_handlers_are_installed = false;
4634
4635 // For signal-chaining
4636 bool os::Linux::libjsig_is_loaded = false;
4637 typedef struct sigaction *(*get_signal_t)(int);
4638 get_signal_t os::Linux::get_signal_action = NULL;
4639
get_chained_signal_action(int sig)4640 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4641 struct sigaction *actp = NULL;
4642
4643 if (libjsig_is_loaded) {
4644 // Retrieve the old signal handler from libjsig
4645 actp = (*get_signal_action)(sig);
4646 }
4647 if (actp == NULL) {
4648 // Retrieve the preinstalled signal handler from jvm
4649 actp = os::Posix::get_preinstalled_handler(sig);
4650 }
4651
4652 return actp;
4653 }
4654
call_chained_handler(struct sigaction * actp,int sig,siginfo_t * siginfo,void * context)4655 static bool call_chained_handler(struct sigaction *actp, int sig,
4656 siginfo_t *siginfo, void *context) {
4657 // Call the old signal handler
4658 if (actp->sa_handler == SIG_DFL) {
4659 // It's more reasonable to let jvm treat it as an unexpected exception
4660 // instead of taking the default action.
4661 return false;
4662 } else if (actp->sa_handler != SIG_IGN) {
4663 if ((actp->sa_flags & SA_NODEFER) == 0) {
4664 // automaticlly block the signal
4665 sigaddset(&(actp->sa_mask), sig);
4666 }
4667
4668 sa_handler_t hand = NULL;
4669 sa_sigaction_t sa = NULL;
4670 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4671 // retrieve the chained handler
4672 if (siginfo_flag_set) {
4673 sa = actp->sa_sigaction;
4674 } else {
4675 hand = actp->sa_handler;
4676 }
4677
4678 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4679 actp->sa_handler = SIG_DFL;
4680 }
4681
4682 // try to honor the signal mask
4683 sigset_t oset;
4684 sigemptyset(&oset);
4685 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4686
4687 // call into the chained handler
4688 if (siginfo_flag_set) {
4689 (*sa)(sig, siginfo, context);
4690 } else {
4691 (*hand)(sig);
4692 }
4693
4694 // restore the signal mask
4695 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4696 }
4697 // Tell jvm's signal handler the signal is taken care of.
4698 return true;
4699 }
4700
chained_handler(int sig,siginfo_t * siginfo,void * context)4701 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4702 bool chained = false;
4703 // signal-chaining
4704 if (UseSignalChaining) {
4705 struct sigaction *actp = get_chained_signal_action(sig);
4706 if (actp != NULL) {
4707 chained = call_chained_handler(actp, sig, siginfo, context);
4708 }
4709 }
4710 return chained;
4711 }
4712
4713 // for diagnostic
4714 int sigflags[NSIG];
4715
get_our_sigflags(int sig)4716 int os::Linux::get_our_sigflags(int sig) {
4717 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4718 return sigflags[sig];
4719 }
4720
set_our_sigflags(int sig,int flags)4721 void os::Linux::set_our_sigflags(int sig, int flags) {
4722 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4723 if (sig > 0 && sig < NSIG) {
4724 sigflags[sig] = flags;
4725 }
4726 }
4727
set_signal_handler(int sig,bool set_installed)4728 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4729 // Check for overwrite.
4730 struct sigaction oldAct;
4731 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4732
4733 void* oldhand = oldAct.sa_sigaction
4734 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4735 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4736 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4737 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4738 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4739 if (AllowUserSignalHandlers || !set_installed) {
4740 // Do not overwrite; user takes responsibility to forward to us.
4741 return;
4742 } else if (UseSignalChaining) {
4743 // save the old handler in jvm
4744 os::Posix::save_preinstalled_handler(sig, oldAct);
4745 // libjsig also interposes the sigaction() call below and saves the
4746 // old sigaction on it own.
4747 } else {
4748 fatal("Encountered unexpected pre-existing sigaction handler "
4749 "%#lx for signal %d.", (long)oldhand, sig);
4750 }
4751 }
4752
4753 struct sigaction sigAct;
4754 sigfillset(&(sigAct.sa_mask));
4755 sigAct.sa_handler = SIG_DFL;
4756 if (!set_installed) {
4757 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4758 } else {
4759 sigAct.sa_sigaction = signalHandler;
4760 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4761 }
4762 // Save flags, which are set by ours
4763 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4764 sigflags[sig] = sigAct.sa_flags;
4765
4766 int ret = sigaction(sig, &sigAct, &oldAct);
4767 assert(ret == 0, "check");
4768
4769 void* oldhand2 = oldAct.sa_sigaction
4770 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4771 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4772 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4773 }
4774
4775 // install signal handlers for signals that HotSpot needs to
4776 // handle in order to support Java-level exception handling.
4777
install_signal_handlers()4778 void os::Linux::install_signal_handlers() {
4779 if (!signal_handlers_are_installed) {
4780 signal_handlers_are_installed = true;
4781
4782 // signal-chaining
4783 typedef void (*signal_setting_t)();
4784 signal_setting_t begin_signal_setting = NULL;
4785 signal_setting_t end_signal_setting = NULL;
4786 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4787 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4788 if (begin_signal_setting != NULL) {
4789 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4790 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4791 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4792 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4793 libjsig_is_loaded = true;
4794 assert(UseSignalChaining, "should enable signal-chaining");
4795 }
4796 if (libjsig_is_loaded) {
4797 // Tell libjsig jvm is setting signal handlers
4798 (*begin_signal_setting)();
4799 }
4800
4801 set_signal_handler(SIGSEGV, true);
4802 set_signal_handler(SIGPIPE, true);
4803 set_signal_handler(SIGBUS, true);
4804 set_signal_handler(SIGILL, true);
4805 set_signal_handler(SIGFPE, true);
4806 #if defined(PPC64)
4807 set_signal_handler(SIGTRAP, true);
4808 #endif
4809 set_signal_handler(SIGXFSZ, true);
4810
4811 if (libjsig_is_loaded) {
4812 // Tell libjsig jvm finishes setting signal handlers
4813 (*end_signal_setting)();
4814 }
4815
4816 // We don't activate signal checker if libjsig is in place, we trust ourselves
4817 // and if UserSignalHandler is installed all bets are off.
4818 // Log that signal checking is off only if -verbose:jni is specified.
4819 if (CheckJNICalls) {
4820 if (libjsig_is_loaded) {
4821 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled");
4822 check_signals = false;
4823 }
4824 if (AllowUserSignalHandlers) {
4825 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4826 check_signals = false;
4827 }
4828 }
4829 }
4830 }
4831
4832 // This is the fastest way to get thread cpu time on Linux.
4833 // Returns cpu time (user+sys) for any thread, not only for current.
4834 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4835 // It might work on 2.6.10+ with a special kernel/glibc patch.
4836 // For reference, please, see IEEE Std 1003.1-2004:
4837 // http://www.unix.org/single_unix_specification
4838
fast_thread_cpu_time(clockid_t clockid)4839 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4840 struct timespec tp;
4841 int rc = os::Posix::clock_gettime(clockid, &tp);
4842 assert(rc == 0, "clock_gettime is expected to return 0 code");
4843
4844 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4845 }
4846
4847 /////
4848 // glibc on Linux platform uses non-documented flag
4849 // to indicate, that some special sort of signal
4850 // trampoline is used.
4851 // We will never set this flag, and we should
4852 // ignore this flag in our diagnostic
4853 #ifdef SIGNIFICANT_SIGNAL_MASK
4854 #undef SIGNIFICANT_SIGNAL_MASK
4855 #endif
4856 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4857
get_signal_handler_name(address handler,char * buf,int buflen)4858 static const char* get_signal_handler_name(address handler,
4859 char* buf, int buflen) {
4860 int offset = 0;
4861 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4862 if (found) {
4863 // skip directory names
4864 const char *p1, *p2;
4865 p1 = buf;
4866 size_t len = strlen(os::file_separator());
4867 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4868 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4869 } else {
4870 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4871 }
4872 return buf;
4873 }
4874
print_signal_handler(outputStream * st,int sig,char * buf,size_t buflen)4875 static void print_signal_handler(outputStream* st, int sig,
4876 char* buf, size_t buflen) {
4877 struct sigaction sa;
4878
4879 sigaction(sig, NULL, &sa);
4880
4881 // See comment for SIGNIFICANT_SIGNAL_MASK define
4882 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4883
4884 st->print("%s: ", os::exception_name(sig, buf, buflen));
4885
4886 address handler = (sa.sa_flags & SA_SIGINFO)
4887 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4888 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4889
4890 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4891 st->print("SIG_DFL");
4892 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4893 st->print("SIG_IGN");
4894 } else {
4895 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4896 }
4897
4898 st->print(", sa_mask[0]=");
4899 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4900
4901 address rh = VMError::get_resetted_sighandler(sig);
4902 // May be, handler was resetted by VMError?
4903 if (rh != NULL) {
4904 handler = rh;
4905 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4906 }
4907
4908 st->print(", sa_flags=");
4909 os::Posix::print_sa_flags(st, sa.sa_flags);
4910
4911 // Check: is it our handler?
4912 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4913 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4914 // It is our signal handler
4915 // check for flags, reset system-used one!
4916 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4917 st->print(
4918 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4919 os::Linux::get_our_sigflags(sig));
4920 }
4921 }
4922 st->cr();
4923 }
4924
4925
4926 #define DO_SIGNAL_CHECK(sig) \
4927 do { \
4928 if (!sigismember(&check_signal_done, sig)) { \
4929 os::Linux::check_signal_handler(sig); \
4930 } \
4931 } while (0)
4932
4933 // This method is a periodic task to check for misbehaving JNI applications
4934 // under CheckJNI, we can add any periodic checks here
4935
run_periodic_checks()4936 void os::run_periodic_checks() {
4937 if (check_signals == false) return;
4938
4939 // SEGV and BUS if overridden could potentially prevent
4940 // generation of hs*.log in the event of a crash, debugging
4941 // such a case can be very challenging, so we absolutely
4942 // check the following for a good measure:
4943 DO_SIGNAL_CHECK(SIGSEGV);
4944 DO_SIGNAL_CHECK(SIGILL);
4945 DO_SIGNAL_CHECK(SIGFPE);
4946 DO_SIGNAL_CHECK(SIGBUS);
4947 DO_SIGNAL_CHECK(SIGPIPE);
4948 DO_SIGNAL_CHECK(SIGXFSZ);
4949 #if defined(PPC64)
4950 DO_SIGNAL_CHECK(SIGTRAP);
4951 #endif
4952
4953 // ReduceSignalUsage allows the user to override these handlers
4954 // see comments at the very top and jvm_md.h
4955 if (!ReduceSignalUsage) {
4956 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4957 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4958 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4959 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4960 }
4961
4962 DO_SIGNAL_CHECK(SR_signum);
4963 }
4964
4965 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4966
4967 static os_sigaction_t os_sigaction = NULL;
4968
check_signal_handler(int sig)4969 void os::Linux::check_signal_handler(int sig) {
4970 char buf[O_BUFLEN];
4971 address jvmHandler = NULL;
4972
4973
4974 struct sigaction act;
4975 if (os_sigaction == NULL) {
4976 // only trust the default sigaction, in case it has been interposed
4977 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4978 if (os_sigaction == NULL) return;
4979 }
4980
4981 os_sigaction(sig, (struct sigaction*)NULL, &act);
4982
4983
4984 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4985
4986 address thisHandler = (act.sa_flags & SA_SIGINFO)
4987 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4988 : CAST_FROM_FN_PTR(address, act.sa_handler);
4989
4990
4991 switch (sig) {
4992 case SIGSEGV:
4993 case SIGBUS:
4994 case SIGFPE:
4995 case SIGPIPE:
4996 case SIGILL:
4997 case SIGXFSZ:
4998 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4999 break;
5000
5001 case SHUTDOWN1_SIGNAL:
5002 case SHUTDOWN2_SIGNAL:
5003 case SHUTDOWN3_SIGNAL:
5004 case BREAK_SIGNAL:
5005 jvmHandler = (address)user_handler();
5006 break;
5007
5008 default:
5009 if (sig == SR_signum) {
5010 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5011 } else {
5012 return;
5013 }
5014 break;
5015 }
5016
5017 if (thisHandler != jvmHandler) {
5018 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5019 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5020 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5021 // No need to check this sig any longer
5022 sigaddset(&check_signal_done, sig);
5023 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5024 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5025 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5026 exception_name(sig, buf, O_BUFLEN));
5027 }
5028 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5029 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5030 tty->print("expected:");
5031 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5032 tty->cr();
5033 tty->print(" found:");
5034 os::Posix::print_sa_flags(tty, act.sa_flags);
5035 tty->cr();
5036 // No need to check this sig any longer
5037 sigaddset(&check_signal_done, sig);
5038 }
5039
5040 // Dump all the signal
5041 if (sigismember(&check_signal_done, sig)) {
5042 print_signal_handlers(tty, buf, O_BUFLEN);
5043 }
5044 }
5045
5046 extern void report_error(char* file_name, int line_no, char* title,
5047 char* format, ...);
5048
5049 // this is called _before_ most of the global arguments have been parsed
init(void)5050 void os::init(void) {
5051 char dummy; // used to get a guess on initial stack address
5052
5053 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5054
5055 init_random(1234567);
5056
5057 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5058 if (Linux::page_size() == -1) {
5059 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5060 os::strerror(errno));
5061 }
5062 init_page_sizes((size_t) Linux::page_size());
5063
5064 Linux::initialize_system_info();
5065
5066 os::Linux::CPUPerfTicks pticks;
5067 bool res = os::Linux::get_tick_information(&pticks, -1);
5068
5069 if (res && pticks.has_steal_ticks) {
5070 has_initial_tick_info = true;
5071 initial_total_ticks = pticks.total;
5072 initial_steal_ticks = pticks.steal;
5073 }
5074
5075 // _main_thread points to the thread that created/loaded the JVM.
5076 Linux::_main_thread = pthread_self();
5077
5078 // retrieve entry point for pthread_setname_np
5079 Linux::_pthread_setname_np =
5080 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5081
5082 os::Posix::init();
5083
5084 initial_time_count = javaTimeNanos();
5085
5086 // Always warn if no monotonic clock available
5087 if (!os::Posix::supports_monotonic_clock()) {
5088 warning("No monotonic clock was available - timed services may " \
5089 "be adversely affected if the time-of-day clock changes");
5090 }
5091 }
5092
5093 // To install functions for atexit system call
5094 extern "C" {
perfMemory_exit_helper()5095 static void perfMemory_exit_helper() {
5096 perfMemory_exit();
5097 }
5098 }
5099
pd_init_container_support()5100 void os::pd_init_container_support() {
5101 OSContainer::init();
5102 }
5103
numa_init()5104 void os::Linux::numa_init() {
5105
5106 // Java can be invoked as
5107 // 1. Without numactl and heap will be allocated/configured on all nodes as
5108 // per the system policy.
5109 // 2. With numactl --interleave:
5110 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5111 // API for membind case bitmask is reset.
5112 // Interleave is only hint and Kernel can fallback to other nodes if
5113 // no memory is available on the target nodes.
5114 // 3. With numactl --membind:
5115 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5116 // interleave case returns bitmask of all nodes.
5117 // numa_all_nodes_ptr holds bitmask of all nodes.
5118 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5119 // bitmask when externally configured to run on all or fewer nodes.
5120
5121 if (!Linux::libnuma_init()) {
5122 UseNUMA = false;
5123 } else {
5124 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5125 // If there's only one node (they start from 0) or if the process
5126 // is bound explicitly to a single node using membind, disable NUMA.
5127 UseNUMA = false;
5128 } else {
5129
5130 LogTarget(Info,os) log;
5131 LogStream ls(log);
5132
5133 Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5134
5135 struct bitmask* bmp = Linux::_numa_membind_bitmask;
5136 const char* numa_mode = "membind";
5137
5138 if (Linux::is_running_in_interleave_mode()) {
5139 bmp = Linux::_numa_interleave_bitmask;
5140 numa_mode = "interleave";
5141 }
5142
5143 ls.print("UseNUMA is enabled and invoked in '%s' mode."
5144 " Heap will be configured using NUMA memory nodes:", numa_mode);
5145
5146 for (int node = 0; node <= Linux::numa_max_node(); node++) {
5147 if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5148 ls.print(" %d", node);
5149 }
5150 }
5151 }
5152 }
5153
5154 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5155 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5156 // we can make the adaptive lgrp chunk resizing work. If the user specified both
5157 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5158 // and disable adaptive resizing.
5159 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5160 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5161 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5162 UseAdaptiveSizePolicy = false;
5163 UseAdaptiveNUMAChunkSizing = false;
5164 }
5165 }
5166
5167 if (!UseNUMA && ForceNUMA) {
5168 UseNUMA = true;
5169 }
5170 }
5171
5172 // this is called _after_ the global arguments have been parsed
init_2(void)5173 jint os::init_2(void) {
5174
5175 // This could be set after os::Posix::init() but all platforms
5176 // have to set it the same so we have to mirror Solaris.
5177 DEBUG_ONLY(os::set_mutex_init_done();)
5178
5179 os::Posix::init_2();
5180
5181 Linux::fast_thread_clock_init();
5182
5183 // initialize suspend/resume support - must do this before signal_sets_init()
5184 if (SR_initialize() != 0) {
5185 perror("SR_initialize failed");
5186 return JNI_ERR;
5187 }
5188
5189 Linux::signal_sets_init();
5190 Linux::install_signal_handlers();
5191 // Initialize data for jdk.internal.misc.Signal
5192 if (!ReduceSignalUsage) {
5193 jdk_misc_signal_init();
5194 }
5195
5196 if (AdjustStackSizeForTLS) {
5197 get_minstack_init();
5198 }
5199
5200 // Check and sets minimum stack sizes against command line options
5201 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5202 return JNI_ERR;
5203 }
5204
5205 #if defined(IA32)
5206 // Need to ensure we've determined the process's initial stack to
5207 // perform the workaround
5208 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5209 workaround_expand_exec_shield_cs_limit();
5210 #else
5211 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5212 if (!suppress_primordial_thread_resolution) {
5213 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5214 }
5215 #endif
5216
5217 Linux::libpthread_init();
5218 Linux::sched_getcpu_init();
5219 log_info(os)("HotSpot is running with %s, %s",
5220 Linux::glibc_version(), Linux::libpthread_version());
5221
5222 if (UseNUMA) {
5223 Linux::numa_init();
5224 }
5225
5226 if (MaxFDLimit) {
5227 // set the number of file descriptors to max. print out error
5228 // if getrlimit/setrlimit fails but continue regardless.
5229 struct rlimit nbr_files;
5230 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5231 if (status != 0) {
5232 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5233 } else {
5234 nbr_files.rlim_cur = nbr_files.rlim_max;
5235 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5236 if (status != 0) {
5237 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5238 }
5239 }
5240 }
5241
5242 // at-exit methods are called in the reverse order of their registration.
5243 // atexit functions are called on return from main or as a result of a
5244 // call to exit(3C). There can be only 32 of these functions registered
5245 // and atexit() does not set errno.
5246
5247 if (PerfAllowAtExitRegistration) {
5248 // only register atexit functions if PerfAllowAtExitRegistration is set.
5249 // atexit functions can be delayed until process exit time, which
5250 // can be problematic for embedded VM situations. Embedded VMs should
5251 // call DestroyJavaVM() to assure that VM resources are released.
5252
5253 // note: perfMemory_exit_helper atexit function may be removed in
5254 // the future if the appropriate cleanup code can be added to the
5255 // VM_Exit VMOperation's doit method.
5256 if (atexit(perfMemory_exit_helper) != 0) {
5257 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5258 }
5259 }
5260
5261 // initialize thread priority policy
5262 prio_init();
5263
5264 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5265 set_coredump_filter(DAX_SHARED_BIT);
5266 }
5267
5268 if (DumpPrivateMappingsInCore) {
5269 set_coredump_filter(FILE_BACKED_PVT_BIT);
5270 }
5271
5272 if (DumpSharedMappingsInCore) {
5273 set_coredump_filter(FILE_BACKED_SHARED_BIT);
5274 }
5275
5276 return JNI_OK;
5277 }
5278
5279 // Mark the polling page as unreadable
make_polling_page_unreadable(void)5280 void os::make_polling_page_unreadable(void) {
5281 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5282 fatal("Could not disable polling page");
5283 }
5284 }
5285
5286 // Mark the polling page as readable
make_polling_page_readable(void)5287 void os::make_polling_page_readable(void) {
5288 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5289 fatal("Could not enable polling page");
5290 }
5291 }
5292
5293 // older glibc versions don't have this macro (which expands to
5294 // an optimized bit-counting function) so we have to roll our own
5295 #ifndef CPU_COUNT
5296
_cpu_count(const cpu_set_t * cpus)5297 static int _cpu_count(const cpu_set_t* cpus) {
5298 int count = 0;
5299 // only look up to the number of configured processors
5300 for (int i = 0; i < os::processor_count(); i++) {
5301 if (CPU_ISSET(i, cpus)) {
5302 count++;
5303 }
5304 }
5305 return count;
5306 }
5307
5308 #define CPU_COUNT(cpus) _cpu_count(cpus)
5309
5310 #endif // CPU_COUNT
5311
5312 // Get the current number of available processors for this process.
5313 // This value can change at any time during a process's lifetime.
5314 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5315 // If it appears there may be more than 1024 processors then we do a
5316 // dynamic check - see 6515172 for details.
5317 // If anything goes wrong we fallback to returning the number of online
5318 // processors - which can be greater than the number available to the process.
active_processor_count()5319 int os::Linux::active_processor_count() {
5320 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5321 cpu_set_t* cpus_p = &cpus;
5322 int cpus_size = sizeof(cpu_set_t);
5323
5324 int configured_cpus = os::processor_count(); // upper bound on available cpus
5325 int cpu_count = 0;
5326
5327 // old build platforms may not support dynamic cpu sets
5328 #ifdef CPU_ALLOC
5329
5330 // To enable easy testing of the dynamic path on different platforms we
5331 // introduce a diagnostic flag: UseCpuAllocPath
5332 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5333 // kernel may use a mask bigger than cpu_set_t
5334 log_trace(os)("active_processor_count: using dynamic path %s"
5335 "- configured processors: %d",
5336 UseCpuAllocPath ? "(forced) " : "",
5337 configured_cpus);
5338 cpus_p = CPU_ALLOC(configured_cpus);
5339 if (cpus_p != NULL) {
5340 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5341 // zero it just to be safe
5342 CPU_ZERO_S(cpus_size, cpus_p);
5343 }
5344 else {
5345 // failed to allocate so fallback to online cpus
5346 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5347 log_trace(os)("active_processor_count: "
5348 "CPU_ALLOC failed (%s) - using "
5349 "online processor count: %d",
5350 os::strerror(errno), online_cpus);
5351 return online_cpus;
5352 }
5353 }
5354 else {
5355 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5356 configured_cpus);
5357 }
5358 #else // CPU_ALLOC
5359 // these stubs won't be executed
5360 #define CPU_COUNT_S(size, cpus) -1
5361 #define CPU_FREE(cpus)
5362
5363 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5364 configured_cpus);
5365 #endif // CPU_ALLOC
5366
5367 // pid 0 means the current thread - which we have to assume represents the process
5368 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5369 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5370 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5371 }
5372 else {
5373 cpu_count = CPU_COUNT(cpus_p);
5374 }
5375 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5376 }
5377 else {
5378 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5379 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5380 "which may exceed available processors", os::strerror(errno), cpu_count);
5381 }
5382
5383 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5384 CPU_FREE(cpus_p);
5385 }
5386
5387 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5388 return cpu_count;
5389 }
5390
5391 // Determine the active processor count from one of
5392 // three different sources:
5393 //
5394 // 1. User option -XX:ActiveProcessorCount
5395 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5396 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5397 //
5398 // Option 1, if specified, will always override.
5399 // If the cgroup subsystem is active and configured, we
5400 // will return the min of the cgroup and option 2 results.
5401 // This is required since tools, such as numactl, that
5402 // alter cpu affinity do not update cgroup subsystem
5403 // cpuset configuration files.
active_processor_count()5404 int os::active_processor_count() {
5405 // User has overridden the number of active processors
5406 if (ActiveProcessorCount > 0) {
5407 log_trace(os)("active_processor_count: "
5408 "active processor count set by user : %d",
5409 ActiveProcessorCount);
5410 return ActiveProcessorCount;
5411 }
5412
5413 int active_cpus;
5414 if (OSContainer::is_containerized()) {
5415 active_cpus = OSContainer::active_processor_count();
5416 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5417 active_cpus);
5418 } else {
5419 active_cpus = os::Linux::active_processor_count();
5420 }
5421
5422 return active_cpus;
5423 }
5424
processor_id()5425 uint os::processor_id() {
5426 const int id = Linux::sched_getcpu();
5427 assert(id >= 0 && id < _processor_count, "Invalid processor id");
5428 return (uint)id;
5429 }
5430
set_native_thread_name(const char * name)5431 void os::set_native_thread_name(const char *name) {
5432 if (Linux::_pthread_setname_np) {
5433 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5434 snprintf(buf, sizeof(buf), "%s", name);
5435 buf[sizeof(buf) - 1] = '\0';
5436 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5437 // ERANGE should not happen; all other errors should just be ignored.
5438 assert(rc != ERANGE, "pthread_setname_np failed");
5439 }
5440 }
5441
bind_to_processor(uint processor_id)5442 bool os::bind_to_processor(uint processor_id) {
5443 // Not yet implemented.
5444 return false;
5445 }
5446
5447 ///
5448
internal_do_task()5449 void os::SuspendedThreadTask::internal_do_task() {
5450 if (do_suspend(_thread->osthread())) {
5451 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5452 do_task(context);
5453 do_resume(_thread->osthread());
5454 }
5455 }
5456
5457 ////////////////////////////////////////////////////////////////////////////////
5458 // debug support
5459
find(address addr,outputStream * st)5460 bool os::find(address addr, outputStream* st) {
5461 Dl_info dlinfo;
5462 memset(&dlinfo, 0, sizeof(dlinfo));
5463 if (dladdr(addr, &dlinfo) != 0) {
5464 st->print(PTR_FORMAT ": ", p2i(addr));
5465 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5466 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5467 p2i(addr) - p2i(dlinfo.dli_saddr));
5468 } else if (dlinfo.dli_fbase != NULL) {
5469 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5470 } else {
5471 st->print("<absolute address>");
5472 }
5473 if (dlinfo.dli_fname != NULL) {
5474 st->print(" in %s", dlinfo.dli_fname);
5475 }
5476 if (dlinfo.dli_fbase != NULL) {
5477 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5478 }
5479 st->cr();
5480
5481 if (Verbose) {
5482 // decode some bytes around the PC
5483 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5484 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5485 address lowest = (address) dlinfo.dli_sname;
5486 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5487 if (begin < lowest) begin = lowest;
5488 Dl_info dlinfo2;
5489 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5490 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5491 end = (address) dlinfo2.dli_saddr;
5492 }
5493 Disassembler::decode(begin, end, st);
5494 }
5495 return true;
5496 }
5497 return false;
5498 }
5499
5500 ////////////////////////////////////////////////////////////////////////////////
5501 // misc
5502
5503 // This does not do anything on Linux. This is basically a hook for being
5504 // able to use structured exception handling (thread-local exception filters)
5505 // on, e.g., Win32.
5506 void
os_exception_wrapper(java_call_t f,JavaValue * value,const methodHandle & method,JavaCallArguments * args,Thread * thread)5507 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5508 JavaCallArguments* args, Thread* thread) {
5509 f(value, method, args, thread);
5510 }
5511
print_statistics()5512 void os::print_statistics() {
5513 }
5514
message_box(const char * title,const char * message)5515 bool os::message_box(const char* title, const char* message) {
5516 int i;
5517 fdStream err(defaultStream::error_fd());
5518 for (i = 0; i < 78; i++) err.print_raw("=");
5519 err.cr();
5520 err.print_raw_cr(title);
5521 for (i = 0; i < 78; i++) err.print_raw("-");
5522 err.cr();
5523 err.print_raw_cr(message);
5524 for (i = 0; i < 78; i++) err.print_raw("=");
5525 err.cr();
5526
5527 char buf[16];
5528 // Prevent process from exiting upon "read error" without consuming all CPU
5529 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5530
5531 return buf[0] == 'y' || buf[0] == 'Y';
5532 }
5533
5534 // Is a (classpath) directory empty?
dir_is_empty(const char * path)5535 bool os::dir_is_empty(const char* path) {
5536 DIR *dir = NULL;
5537 struct dirent *ptr;
5538
5539 dir = opendir(path);
5540 if (dir == NULL) return true;
5541
5542 // Scan the directory
5543 bool result = true;
5544 while (result && (ptr = readdir(dir)) != NULL) {
5545 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5546 result = false;
5547 }
5548 }
5549 closedir(dir);
5550 return result;
5551 }
5552
5553 // This code originates from JDK's sysOpen and open64_w
5554 // from src/solaris/hpi/src/system_md.c
5555
open(const char * path,int oflag,int mode)5556 int os::open(const char *path, int oflag, int mode) {
5557 if (strlen(path) > MAX_PATH - 1) {
5558 errno = ENAMETOOLONG;
5559 return -1;
5560 }
5561
5562 // All file descriptors that are opened in the Java process and not
5563 // specifically destined for a subprocess should have the close-on-exec
5564 // flag set. If we don't set it, then careless 3rd party native code
5565 // might fork and exec without closing all appropriate file descriptors
5566 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5567 // turn might:
5568 //
5569 // - cause end-of-file to fail to be detected on some file
5570 // descriptors, resulting in mysterious hangs, or
5571 //
5572 // - might cause an fopen in the subprocess to fail on a system
5573 // suffering from bug 1085341.
5574 //
5575 // (Yes, the default setting of the close-on-exec flag is a Unix
5576 // design flaw)
5577 //
5578 // See:
5579 // 1085341: 32-bit stdio routines should support file descriptors >255
5580 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5581 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5582 //
5583 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5584 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5585 // because it saves a system call and removes a small window where the flag
5586 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5587 // and we fall back to using FD_CLOEXEC (see below).
5588 #ifdef O_CLOEXEC
5589 oflag |= O_CLOEXEC;
5590 #endif
5591
5592 int fd = ::open64(path, oflag, mode);
5593 if (fd == -1) return -1;
5594
5595 //If the open succeeded, the file might still be a directory
5596 {
5597 struct stat64 buf64;
5598 int ret = ::fstat64(fd, &buf64);
5599 int st_mode = buf64.st_mode;
5600
5601 if (ret != -1) {
5602 if ((st_mode & S_IFMT) == S_IFDIR) {
5603 errno = EISDIR;
5604 ::close(fd);
5605 return -1;
5606 }
5607 } else {
5608 ::close(fd);
5609 return -1;
5610 }
5611 }
5612
5613 #ifdef FD_CLOEXEC
5614 // Validate that the use of the O_CLOEXEC flag on open above worked.
5615 // With recent kernels, we will perform this check exactly once.
5616 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5617 if (!O_CLOEXEC_is_known_to_work) {
5618 int flags = ::fcntl(fd, F_GETFD);
5619 if (flags != -1) {
5620 if ((flags & FD_CLOEXEC) != 0)
5621 O_CLOEXEC_is_known_to_work = 1;
5622 else
5623 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5624 }
5625 }
5626 #endif
5627
5628 return fd;
5629 }
5630
5631
5632 // create binary file, rewriting existing file if required
create_binary_file(const char * path,bool rewrite_existing)5633 int os::create_binary_file(const char* path, bool rewrite_existing) {
5634 int oflags = O_WRONLY | O_CREAT;
5635 if (!rewrite_existing) {
5636 oflags |= O_EXCL;
5637 }
5638 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5639 }
5640
5641 // return current position of file pointer
current_file_offset(int fd)5642 jlong os::current_file_offset(int fd) {
5643 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5644 }
5645
5646 // move file pointer to the specified offset
seek_to_file_offset(int fd,jlong offset)5647 jlong os::seek_to_file_offset(int fd, jlong offset) {
5648 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5649 }
5650
5651 // This code originates from JDK's sysAvailable
5652 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5653
available(int fd,jlong * bytes)5654 int os::available(int fd, jlong *bytes) {
5655 jlong cur, end;
5656 int mode;
5657 struct stat64 buf64;
5658
5659 if (::fstat64(fd, &buf64) >= 0) {
5660 mode = buf64.st_mode;
5661 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5662 int n;
5663 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5664 *bytes = n;
5665 return 1;
5666 }
5667 }
5668 }
5669 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5670 return 0;
5671 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5672 return 0;
5673 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5674 return 0;
5675 }
5676 *bytes = end - cur;
5677 return 1;
5678 }
5679
5680 // Map a block of memory.
pd_map_memory(int fd,const char * file_name,size_t file_offset,char * addr,size_t bytes,bool read_only,bool allow_exec)5681 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5682 char *addr, size_t bytes, bool read_only,
5683 bool allow_exec) {
5684 int prot;
5685 int flags = MAP_PRIVATE;
5686
5687 if (read_only) {
5688 prot = PROT_READ;
5689 } else {
5690 prot = PROT_READ | PROT_WRITE;
5691 }
5692
5693 if (allow_exec) {
5694 prot |= PROT_EXEC;
5695 }
5696
5697 if (addr != NULL) {
5698 flags |= MAP_FIXED;
5699 }
5700
5701 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5702 fd, file_offset);
5703 if (mapped_address == MAP_FAILED) {
5704 return NULL;
5705 }
5706 return mapped_address;
5707 }
5708
5709
5710 // Remap a block of memory.
pd_remap_memory(int fd,const char * file_name,size_t file_offset,char * addr,size_t bytes,bool read_only,bool allow_exec)5711 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5712 char *addr, size_t bytes, bool read_only,
5713 bool allow_exec) {
5714 // same as map_memory() on this OS
5715 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5716 allow_exec);
5717 }
5718
5719
5720 // Unmap a block of memory.
pd_unmap_memory(char * addr,size_t bytes)5721 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5722 return munmap(addr, bytes) == 0;
5723 }
5724
5725 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5726
fast_cpu_time(Thread * thread)5727 static jlong fast_cpu_time(Thread *thread) {
5728 clockid_t clockid;
5729 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5730 &clockid);
5731 if (rc == 0) {
5732 return os::Linux::fast_thread_cpu_time(clockid);
5733 } else {
5734 // It's possible to encounter a terminated native thread that failed
5735 // to detach itself from the VM - which should result in ESRCH.
5736 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5737 return -1;
5738 }
5739 }
5740
5741 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5742 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5743 // of a thread.
5744 //
5745 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5746 // the fast estimate available on the platform.
5747
current_thread_cpu_time()5748 jlong os::current_thread_cpu_time() {
5749 if (os::Linux::supports_fast_thread_cpu_time()) {
5750 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5751 } else {
5752 // return user + sys since the cost is the same
5753 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5754 }
5755 }
5756
thread_cpu_time(Thread * thread)5757 jlong os::thread_cpu_time(Thread* thread) {
5758 // consistent with what current_thread_cpu_time() returns
5759 if (os::Linux::supports_fast_thread_cpu_time()) {
5760 return fast_cpu_time(thread);
5761 } else {
5762 return slow_thread_cpu_time(thread, true /* user + sys */);
5763 }
5764 }
5765
current_thread_cpu_time(bool user_sys_cpu_time)5766 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5767 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5768 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5769 } else {
5770 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5771 }
5772 }
5773
thread_cpu_time(Thread * thread,bool user_sys_cpu_time)5774 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5775 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5776 return fast_cpu_time(thread);
5777 } else {
5778 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5779 }
5780 }
5781
5782 // -1 on error.
slow_thread_cpu_time(Thread * thread,bool user_sys_cpu_time)5783 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5784 pid_t tid = thread->osthread()->thread_id();
5785 char *s;
5786 char stat[2048];
5787 int statlen;
5788 char proc_name[64];
5789 int count;
5790 long sys_time, user_time;
5791 char cdummy;
5792 int idummy;
5793 long ldummy;
5794 FILE *fp;
5795
5796 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5797 fp = fopen(proc_name, "r");
5798 if (fp == NULL) return -1;
5799 statlen = fread(stat, 1, 2047, fp);
5800 stat[statlen] = '\0';
5801 fclose(fp);
5802
5803 // Skip pid and the command string. Note that we could be dealing with
5804 // weird command names, e.g. user could decide to rename java launcher
5805 // to "java 1.4.2 :)", then the stat file would look like
5806 // 1234 (java 1.4.2 :)) R ... ...
5807 // We don't really need to know the command string, just find the last
5808 // occurrence of ")" and then start parsing from there. See bug 4726580.
5809 s = strrchr(stat, ')');
5810 if (s == NULL) return -1;
5811
5812 // Skip blank chars
5813 do { s++; } while (s && isspace(*s));
5814
5815 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5816 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5817 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5818 &user_time, &sys_time);
5819 if (count != 13) return -1;
5820 if (user_sys_cpu_time) {
5821 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5822 } else {
5823 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5824 }
5825 }
5826
current_thread_cpu_time_info(jvmtiTimerInfo * info_ptr)5827 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5828 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5829 info_ptr->may_skip_backward = false; // elapsed time not wall time
5830 info_ptr->may_skip_forward = false; // elapsed time not wall time
5831 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5832 }
5833
thread_cpu_time_info(jvmtiTimerInfo * info_ptr)5834 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5835 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5836 info_ptr->may_skip_backward = false; // elapsed time not wall time
5837 info_ptr->may_skip_forward = false; // elapsed time not wall time
5838 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5839 }
5840
is_thread_cpu_time_supported()5841 bool os::is_thread_cpu_time_supported() {
5842 return true;
5843 }
5844
5845 // System loadavg support. Returns -1 if load average cannot be obtained.
5846 // Linux doesn't yet have a (official) notion of processor sets,
5847 // so just return the system wide load average.
loadavg(double loadavg[],int nelem)5848 int os::loadavg(double loadavg[], int nelem) {
5849 return ::getloadavg(loadavg, nelem);
5850 }
5851
pause()5852 void os::pause() {
5853 char filename[MAX_PATH];
5854 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5855 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5856 } else {
5857 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5858 }
5859
5860 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5861 if (fd != -1) {
5862 struct stat buf;
5863 ::close(fd);
5864 while (::stat(filename, &buf) == 0) {
5865 (void)::poll(NULL, 0, 100);
5866 }
5867 } else {
5868 jio_fprintf(stderr,
5869 "Could not open pause file '%s', continuing immediately.\n", filename);
5870 }
5871 }
5872
5873 extern char** environ;
5874
5875 // Run the specified command in a separate process. Return its exit value,
5876 // or -1 on failure (e.g. can't fork a new process).
5877 // Unlike system(), this function can be called from signal handler. It
5878 // doesn't block SIGINT et al.
fork_and_exec(char * cmd,bool use_vfork_if_available)5879 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5880 const char * argv[4] = {"sh", "-c", cmd, NULL};
5881
5882 pid_t pid ;
5883
5884 if (use_vfork_if_available) {
5885 pid = vfork();
5886 } else {
5887 pid = fork();
5888 }
5889
5890 if (pid < 0) {
5891 // fork failed
5892 return -1;
5893
5894 } else if (pid == 0) {
5895 // child process
5896
5897 execve("/bin/sh", (char* const*)argv, environ);
5898
5899 // execve failed
5900 _exit(-1);
5901
5902 } else {
5903 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5904 // care about the actual exit code, for now.
5905
5906 int status;
5907
5908 // Wait for the child process to exit. This returns immediately if
5909 // the child has already exited. */
5910 while (waitpid(pid, &status, 0) < 0) {
5911 switch (errno) {
5912 case ECHILD: return 0;
5913 case EINTR: break;
5914 default: return -1;
5915 }
5916 }
5917
5918 if (WIFEXITED(status)) {
5919 // The child exited normally; get its exit code.
5920 return WEXITSTATUS(status);
5921 } else if (WIFSIGNALED(status)) {
5922 // The child exited because of a signal
5923 // The best value to return is 0x80 + signal number,
5924 // because that is what all Unix shells do, and because
5925 // it allows callers to distinguish between process exit and
5926 // process death by signal.
5927 return 0x80 + WTERMSIG(status);
5928 } else {
5929 // Unknown exit code; pass it through
5930 return status;
5931 }
5932 }
5933 }
5934
5935 // Get the default path to the core file
5936 // Returns the length of the string
get_core_path(char * buffer,size_t bufferSize)5937 int os::get_core_path(char* buffer, size_t bufferSize) {
5938 /*
5939 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5940 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5941 */
5942 const int core_pattern_len = 129;
5943 char core_pattern[core_pattern_len] = {0};
5944
5945 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5946 if (core_pattern_file == -1) {
5947 return -1;
5948 }
5949
5950 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5951 ::close(core_pattern_file);
5952 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5953 return -1;
5954 }
5955 if (core_pattern[ret-1] == '\n') {
5956 core_pattern[ret-1] = '\0';
5957 } else {
5958 core_pattern[ret] = '\0';
5959 }
5960
5961 // Replace the %p in the core pattern with the process id. NOTE: we do this
5962 // only if the pattern doesn't start with "|", and we support only one %p in
5963 // the pattern.
5964 char *pid_pos = strstr(core_pattern, "%p");
5965 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p"
5966 int written;
5967
5968 if (core_pattern[0] == '/') {
5969 if (pid_pos != NULL) {
5970 *pid_pos = '\0';
5971 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5972 current_process_id(), tail);
5973 } else {
5974 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5975 }
5976 } else {
5977 char cwd[PATH_MAX];
5978
5979 const char* p = get_current_directory(cwd, PATH_MAX);
5980 if (p == NULL) {
5981 return -1;
5982 }
5983
5984 if (core_pattern[0] == '|') {
5985 written = jio_snprintf(buffer, bufferSize,
5986 "\"%s\" (or dumping to %s/core.%d)",
5987 &core_pattern[1], p, current_process_id());
5988 } else if (pid_pos != NULL) {
5989 *pid_pos = '\0';
5990 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5991 current_process_id(), tail);
5992 } else {
5993 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5994 }
5995 }
5996
5997 if (written < 0) {
5998 return -1;
5999 }
6000
6001 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6002 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6003
6004 if (core_uses_pid_file != -1) {
6005 char core_uses_pid = 0;
6006 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6007 ::close(core_uses_pid_file);
6008
6009 if (core_uses_pid == '1') {
6010 jio_snprintf(buffer + written, bufferSize - written,
6011 ".%d", current_process_id());
6012 }
6013 }
6014 }
6015
6016 return strlen(buffer);
6017 }
6018
start_debugging(char * buf,int buflen)6019 bool os::start_debugging(char *buf, int buflen) {
6020 int len = (int)strlen(buf);
6021 char *p = &buf[len];
6022
6023 jio_snprintf(p, buflen-len,
6024 "\n\n"
6025 "Do you want to debug the problem?\n\n"
6026 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6027 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6028 "Otherwise, press RETURN to abort...",
6029 os::current_process_id(), os::current_process_id(),
6030 os::current_thread_id(), os::current_thread_id());
6031
6032 bool yes = os::message_box("Unexpected Error", buf);
6033
6034 if (yes) {
6035 // yes, user asked VM to launch debugger
6036 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6037 os::current_process_id(), os::current_process_id());
6038
6039 os::fork_and_exec(buf);
6040 yes = false;
6041 }
6042 return yes;
6043 }
6044
6045
6046 // Java/Compiler thread:
6047 //
6048 // Low memory addresses
6049 // P0 +------------------------+
6050 // | |\ Java thread created by VM does not have glibc
6051 // | glibc guard page | - guard page, attached Java thread usually has
6052 // | |/ 1 glibc guard page.
6053 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6054 // | |\
6055 // | HotSpot Guard Pages | - red, yellow and reserved pages
6056 // | |/
6057 // +------------------------+ JavaThread::stack_reserved_zone_base()
6058 // | |\
6059 // | Normal Stack | -
6060 // | |/
6061 // P2 +------------------------+ Thread::stack_base()
6062 //
6063 // Non-Java thread:
6064 //
6065 // Low memory addresses
6066 // P0 +------------------------+
6067 // | |\
6068 // | glibc guard page | - usually 1 page
6069 // | |/
6070 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6071 // | |\
6072 // | Normal Stack | -
6073 // | |/
6074 // P2 +------------------------+ Thread::stack_base()
6075 //
6076 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6077 // returned from pthread_attr_getstack().
6078 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6079 // of the stack size given in pthread_attr. We work around this for
6080 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6081 //
6082 #ifndef ZERO
current_stack_region(address * bottom,size_t * size)6083 static void current_stack_region(address * bottom, size_t * size) {
6084 if (os::is_primordial_thread()) {
6085 // primordial thread needs special handling because pthread_getattr_np()
6086 // may return bogus value.
6087 *bottom = os::Linux::initial_thread_stack_bottom();
6088 *size = os::Linux::initial_thread_stack_size();
6089 } else {
6090 pthread_attr_t attr;
6091
6092 int rslt = pthread_getattr_np(pthread_self(), &attr);
6093
6094 // JVM needs to know exact stack location, abort if it fails
6095 if (rslt != 0) {
6096 if (rslt == ENOMEM) {
6097 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6098 } else {
6099 fatal("pthread_getattr_np failed with error = %d", rslt);
6100 }
6101 }
6102
6103 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6104 fatal("Cannot locate current stack attributes!");
6105 }
6106
6107 // Work around NPTL stack guard error.
6108 size_t guard_size = 0;
6109 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6110 if (rslt != 0) {
6111 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6112 }
6113 *bottom += guard_size;
6114 *size -= guard_size;
6115
6116 pthread_attr_destroy(&attr);
6117
6118 }
6119 assert(os::current_stack_pointer() >= *bottom &&
6120 os::current_stack_pointer() < *bottom + *size, "just checking");
6121 }
6122
current_stack_base()6123 address os::current_stack_base() {
6124 address bottom;
6125 size_t size;
6126 current_stack_region(&bottom, &size);
6127 return (bottom + size);
6128 }
6129
current_stack_size()6130 size_t os::current_stack_size() {
6131 // This stack size includes the usable stack and HotSpot guard pages
6132 // (for the threads that have Hotspot guard pages).
6133 address bottom;
6134 size_t size;
6135 current_stack_region(&bottom, &size);
6136 return size;
6137 }
6138 #endif
6139
get_mtime(const char * filename)6140 static inline struct timespec get_mtime(const char* filename) {
6141 struct stat st;
6142 int ret = os::stat(filename, &st);
6143 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6144 return st.st_mtim;
6145 }
6146
compare_file_modified_times(const char * file1,const char * file2)6147 int os::compare_file_modified_times(const char* file1, const char* file2) {
6148 struct timespec filetime1 = get_mtime(file1);
6149 struct timespec filetime2 = get_mtime(file2);
6150 int diff = filetime1.tv_sec - filetime2.tv_sec;
6151 if (diff == 0) {
6152 return filetime1.tv_nsec - filetime2.tv_nsec;
6153 }
6154 return diff;
6155 }
6156
supports_map_sync()6157 bool os::supports_map_sync() {
6158 return true;
6159 }
6160
6161 /////////////// Unit tests ///////////////
6162
6163 #ifndef PRODUCT
6164
6165 class TestReserveMemorySpecial : AllStatic {
6166 public:
small_page_write(void * addr,size_t size)6167 static void small_page_write(void* addr, size_t size) {
6168 size_t page_size = os::vm_page_size();
6169
6170 char* end = (char*)addr + size;
6171 for (char* p = (char*)addr; p < end; p += page_size) {
6172 *p = 1;
6173 }
6174 }
6175
test_reserve_memory_special_huge_tlbfs_only(size_t size)6176 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6177 if (!UseHugeTLBFS) {
6178 return;
6179 }
6180
6181 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6182
6183 if (addr != NULL) {
6184 small_page_write(addr, size);
6185
6186 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6187 }
6188 }
6189
test_reserve_memory_special_huge_tlbfs_only()6190 static void test_reserve_memory_special_huge_tlbfs_only() {
6191 if (!UseHugeTLBFS) {
6192 return;
6193 }
6194
6195 size_t lp = os::large_page_size();
6196
6197 for (size_t size = lp; size <= lp * 10; size += lp) {
6198 test_reserve_memory_special_huge_tlbfs_only(size);
6199 }
6200 }
6201
test_reserve_memory_special_huge_tlbfs_mixed()6202 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6203 size_t lp = os::large_page_size();
6204 size_t ag = os::vm_allocation_granularity();
6205
6206 // sizes to test
6207 const size_t sizes[] = {
6208 lp, lp + ag, lp + lp / 2, lp * 2,
6209 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6210 lp * 10, lp * 10 + lp / 2
6211 };
6212 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6213
6214 // For each size/alignment combination, we test three scenarios:
6215 // 1) with req_addr == NULL
6216 // 2) with a non-null req_addr at which we expect to successfully allocate
6217 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6218 // expect the allocation to either fail or to ignore req_addr
6219
6220 // Pre-allocate two areas; they shall be as large as the largest allocation
6221 // and aligned to the largest alignment we will be testing.
6222 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6223 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6224 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6225 -1, 0);
6226 assert(mapping1 != MAP_FAILED, "should work");
6227
6228 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6229 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6230 -1, 0);
6231 assert(mapping2 != MAP_FAILED, "should work");
6232
6233 // Unmap the first mapping, but leave the second mapping intact: the first
6234 // mapping will serve as a value for a "good" req_addr (case 2). The second
6235 // mapping, still intact, as "bad" req_addr (case 3).
6236 ::munmap(mapping1, mapping_size);
6237
6238 // Case 1
6239 for (int i = 0; i < num_sizes; i++) {
6240 const size_t size = sizes[i];
6241 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6242 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6243 if (p != NULL) {
6244 assert(is_aligned(p, alignment), "must be");
6245 small_page_write(p, size);
6246 os::Linux::release_memory_special_huge_tlbfs(p, size);
6247 }
6248 }
6249 }
6250
6251 // Case 2
6252 for (int i = 0; i < num_sizes; i++) {
6253 const size_t size = sizes[i];
6254 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6255 char* const req_addr = align_up(mapping1, alignment);
6256 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6257 if (p != NULL) {
6258 assert(p == req_addr, "must be");
6259 small_page_write(p, size);
6260 os::Linux::release_memory_special_huge_tlbfs(p, size);
6261 }
6262 }
6263 }
6264
6265 // Case 3
6266 for (int i = 0; i < num_sizes; i++) {
6267 const size_t size = sizes[i];
6268 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6269 char* const req_addr = align_up(mapping2, alignment);
6270 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6271 // as the area around req_addr contains already existing mappings, the API should always
6272 // return NULL (as per contract, it cannot return another address)
6273 assert(p == NULL, "must be");
6274 }
6275 }
6276
6277 ::munmap(mapping2, mapping_size);
6278
6279 }
6280
test_reserve_memory_special_huge_tlbfs()6281 static void test_reserve_memory_special_huge_tlbfs() {
6282 if (!UseHugeTLBFS) {
6283 return;
6284 }
6285
6286 test_reserve_memory_special_huge_tlbfs_only();
6287 test_reserve_memory_special_huge_tlbfs_mixed();
6288 }
6289
test_reserve_memory_special_shm(size_t size,size_t alignment)6290 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6291 if (!UseSHM) {
6292 return;
6293 }
6294
6295 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6296
6297 if (addr != NULL) {
6298 assert(is_aligned(addr, alignment), "Check");
6299 assert(is_aligned(addr, os::large_page_size()), "Check");
6300
6301 small_page_write(addr, size);
6302
6303 os::Linux::release_memory_special_shm(addr, size);
6304 }
6305 }
6306
test_reserve_memory_special_shm()6307 static void test_reserve_memory_special_shm() {
6308 size_t lp = os::large_page_size();
6309 size_t ag = os::vm_allocation_granularity();
6310
6311 for (size_t size = ag; size < lp * 3; size += ag) {
6312 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6313 test_reserve_memory_special_shm(size, alignment);
6314 }
6315 }
6316 }
6317
test()6318 static void test() {
6319 test_reserve_memory_special_huge_tlbfs();
6320 test_reserve_memory_special_shm();
6321 }
6322 };
6323
TestReserveMemorySpecial_test()6324 void TestReserveMemorySpecial_test() {
6325 TestReserveMemorySpecial::test();
6326 }
6327
6328 #endif
6329