1 /*
2 * Copyright (c) 1999, 2013, 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 "asm/macroAssembler.hpp"
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 "interpreter/interpreter.hpp"
33 #include "jvm_linux.h"
34 #include "memory/allocation.inline.hpp"
35 #include "mutex_linux.inline.hpp"
36 #include "os_share_linux.hpp"
37 #include "prims/jniFastGetField.hpp"
38 #include "prims/jvm.h"
39 #include "prims/jvm_misc.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/extendedPC.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/interfaceSupport.hpp"
44 #include "runtime/java.hpp"
45 #include "runtime/javaCalls.hpp"
46 #include "runtime/mutexLocker.hpp"
47 #include "runtime/osThread.hpp"
48 #include "runtime/sharedRuntime.hpp"
49 #include "runtime/stubRoutines.hpp"
50 #include "runtime/thread.inline.hpp"
51 #include "runtime/timer.hpp"
52 #include "services/memTracker.hpp"
53 #include "utilities/events.hpp"
54 #include "utilities/vmError.hpp"
55
56 // put OS-includes here
57 # include <sys/types.h>
58 # include <sys/mman.h>
59 # include <pthread.h>
60 # include <signal.h>
61 # include <errno.h>
62 # include <dlfcn.h>
63 # include <stdlib.h>
64 # include <stdio.h>
65 # include <unistd.h>
66 # include <sys/resource.h>
67 # include <pthread.h>
68 # include <sys/stat.h>
69 # include <sys/time.h>
70 # include <sys/utsname.h>
71 # include <sys/socket.h>
72 # include <sys/wait.h>
73 # include <pwd.h>
74 # include <poll.h>
75 # include <ucontext.h>
76 # include <fpu_control.h>
77
78 #ifdef AMD64
79 #define REG_SP REG_RSP
80 #define REG_PC REG_RIP
81 #define REG_FP REG_RBP
82 #define SPELL_REG_SP "rsp"
83 #define SPELL_REG_FP "rbp"
84 #else
85 #define REG_SP REG_UESP
86 #define REG_PC REG_EIP
87 #define REG_FP REG_EBP
88 #define SPELL_REG_SP "esp"
89 #define SPELL_REG_FP "ebp"
90 #endif // AMD64
91
92 address os::current_stack_pointer() {
93 #ifdef SPARC_WORKS
94 register void *esp;
95 __asm__("mov %%"SPELL_REG_SP", %0":"=r"(esp));
96 return (address) ((char*)esp + sizeof(long)*2);
97 #elif defined(__clang__)
98 intptr_t* esp;
99 __asm__ __volatile__ ("mov %%"SPELL_REG_SP", %0":"=r"(esp):);
100 return (address) esp;
101 #else
102 register void *esp __asm__ (SPELL_REG_SP);
103 return (address) esp;
104 #endif
105 }
106
107 char* os::non_memory_address_word() {
108 // Must never look like an address returned by reserve_memory,
109 // even in its subfields (as defined by the CPU immediate fields,
110 // if the CPU splits constants across multiple instructions).
111
112 return (char*) -1;
113 }
114
115 void os::initialize_thread(Thread* thr) {
116 // Nothing to do.
117 }
118
119 address os::Linux::ucontext_get_pc(ucontext_t * uc) {
120 return (address)uc->uc_mcontext.gregs[REG_PC];
121 }
122
123 intptr_t* os::Linux::ucontext_get_sp(ucontext_t * uc) {
124 return (intptr_t*)uc->uc_mcontext.gregs[REG_SP];
125 }
126
127 intptr_t* os::Linux::ucontext_get_fp(ucontext_t * uc) {
128 return (intptr_t*)uc->uc_mcontext.gregs[REG_FP];
129 }
130
131 // For Forte Analyzer AsyncGetCallTrace profiling support - thread
132 // is currently interrupted by SIGPROF.
133 // os::Solaris::fetch_frame_from_ucontext() tries to skip nested signal
134 // frames. Currently we don't do that on Linux, so it's the same as
135 // os::fetch_frame_from_context().
136 ExtendedPC os::Linux::fetch_frame_from_ucontext(Thread* thread,
137 ucontext_t* uc, intptr_t** ret_sp, intptr_t** ret_fp) {
138
139 assert(thread != NULL, "just checking");
140 assert(ret_sp != NULL, "just checking");
141 assert(ret_fp != NULL, "just checking");
142
143 return os::fetch_frame_from_context(uc, ret_sp, ret_fp);
144 }
145
146 ExtendedPC os::fetch_frame_from_context(void* ucVoid,
147 intptr_t** ret_sp, intptr_t** ret_fp) {
148
149 ExtendedPC epc;
150 ucontext_t* uc = (ucontext_t*)ucVoid;
151
152 if (uc != NULL) {
153 epc = ExtendedPC(os::Linux::ucontext_get_pc(uc));
154 if (ret_sp) *ret_sp = os::Linux::ucontext_get_sp(uc);
155 if (ret_fp) *ret_fp = os::Linux::ucontext_get_fp(uc);
156 } else {
157 // construct empty ExtendedPC for return value checking
158 epc = ExtendedPC(NULL);
159 if (ret_sp) *ret_sp = (intptr_t *)NULL;
160 if (ret_fp) *ret_fp = (intptr_t *)NULL;
161 }
162
163 return epc;
164 }
165
166 frame os::fetch_frame_from_context(void* ucVoid) {
167 intptr_t* sp;
168 intptr_t* fp;
169 ExtendedPC epc = fetch_frame_from_context(ucVoid, &sp, &fp);
170 return frame(sp, fp, epc.pc());
171 }
172
173 // By default, gcc always save frame pointer (%ebp/%rbp) on stack. It may get
174 // turned off by -fomit-frame-pointer,
175 frame os::get_sender_for_C_frame(frame* fr) {
176 return frame(fr->sender_sp(), fr->link(), fr->sender_pc());
177 }
178
179 intptr_t* _get_previous_fp() {
180 #ifdef SPARC_WORKS
181 register intptr_t **ebp;
182 __asm__("mov %%"SPELL_REG_FP", %0":"=r"(ebp));
183 #elif defined(__clang__)
184 intptr_t **ebp;
185 __asm__ __volatile__ ("mov %%"SPELL_REG_FP", %0":"=r"(ebp):);
186 #else
187 register intptr_t **ebp __asm__ (SPELL_REG_FP);
188 #endif
189 return (intptr_t*) *ebp; // we want what it points to.
190 }
191
192
193 frame os::current_frame() {
194 intptr_t* fp = _get_previous_fp();
195 frame myframe((intptr_t*)os::current_stack_pointer(),
196 (intptr_t*)fp,
197 CAST_FROM_FN_PTR(address, os::current_frame));
198 if (os::is_first_C_frame(&myframe)) {
199 // stack is not walkable
200 return frame();
201 } else {
202 return os::get_sender_for_C_frame(&myframe);
203 }
204 }
205
206 // Utility functions
207
208 // From IA32 System Programming Guide
209 enum {
210 trap_page_fault = 0xE
211 };
212
213 extern "C" JNIEXPORT int
214 JVM_handle_linux_signal(int sig,
215 siginfo_t* info,
216 void* ucVoid,
217 int abort_if_unrecognized) {
218 ucontext_t* uc = (ucontext_t*) ucVoid;
219
220 Thread* t = ThreadLocalStorage::get_thread_slow();
221
222 // Must do this before SignalHandlerMark, if crash protection installed we will longjmp away
223 // (no destructors can be run)
224 os::WatcherThreadCrashProtection::check_crash_protection(sig, t);
225
226 SignalHandlerMark shm(t);
227
228 // Note: it's not uncommon that JNI code uses signal/sigset to install
229 // then restore certain signal handler (e.g. to temporarily block SIGPIPE,
230 // or have a SIGILL handler when detecting CPU type). When that happens,
231 // JVM_handle_linux_signal() might be invoked with junk info/ucVoid. To
232 // avoid unnecessary crash when libjsig is not preloaded, try handle signals
233 // that do not require siginfo/ucontext first.
234
235 if (sig == SIGPIPE || sig == SIGXFSZ) {
236 // allow chained handler to go first
237 if (os::Linux::chained_handler(sig, info, ucVoid)) {
238 return true;
239 } else {
240 if (PrintMiscellaneous && (WizardMode || Verbose)) {
241 char buf[64];
242 warning("Ignoring %s - see bugs 4229104 or 646499219",
243 os::exception_name(sig, buf, sizeof(buf)));
244 }
245 return true;
246 }
247 }
248
249 JavaThread* thread = NULL;
250 VMThread* vmthread = NULL;
251 if (os::Linux::signal_handlers_are_installed) {
252 if (t != NULL ){
253 if(t->is_Java_thread()) {
254 thread = (JavaThread*)t;
255 }
256 else if(t->is_VM_thread()){
257 vmthread = (VMThread *)t;
258 }
259 }
260 }
261 /*
262 NOTE: does not seem to work on linux.
263 if (info == NULL || info->si_code <= 0 || info->si_code == SI_NOINFO) {
264 // can't decode this kind of signal
265 info = NULL;
266 } else {
267 assert(sig == info->si_signo, "bad siginfo");
268 }
269 */
270 // decide if this trap can be handled by a stub
271 address stub = NULL;
272
273 address pc = NULL;
274
275 //%note os_trap_1
276 if (info != NULL && uc != NULL && thread != NULL) {
277 pc = (address) os::Linux::ucontext_get_pc(uc);
278
279 if (StubRoutines::is_safefetch_fault(pc)) {
280 uc->uc_mcontext.gregs[REG_PC] = intptr_t(StubRoutines::continuation_for_safefetch_fault(pc));
281 return 1;
282 }
283
284 #ifndef AMD64
285 // Halt if SI_KERNEL before more crashes get misdiagnosed as Java bugs
286 // This can happen in any running code (currently more frequently in
287 // interpreter code but has been seen in compiled code)
288 if (sig == SIGSEGV && info->si_addr == 0 && info->si_code == SI_KERNEL) {
289 fatal("An irrecoverable SI_KERNEL SIGSEGV has occurred due "
290 "to unstable signal handling in this distribution.");
291 }
292 #endif // AMD64
293
294 // Handle ALL stack overflow variations here
295 if (sig == SIGSEGV) {
296 address addr = (address) info->si_addr;
297
298 // check if fault address is within thread stack
299 if (addr < thread->stack_base() &&
300 addr >= thread->stack_base() - thread->stack_size()) {
301 // stack overflow
302 if (thread->in_stack_yellow_zone(addr)) {
303 thread->disable_stack_yellow_zone();
304 if (thread->thread_state() == _thread_in_Java) {
305 // Throw a stack overflow exception. Guard pages will be reenabled
306 // while unwinding the stack.
307 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::STACK_OVERFLOW);
308 } else {
309 // Thread was in the vm or native code. Return and try to finish.
310 return 1;
311 }
312 } else if (thread->in_stack_red_zone(addr)) {
313 // Fatal red zone violation. Disable the guard pages and fall through
314 // to handle_unexpected_exception way down below.
315 thread->disable_stack_red_zone();
316 tty->print_raw_cr("An irrecoverable stack overflow has occurred.");
317
318 // This is a likely cause, but hard to verify. Let's just print
319 // it as a hint.
320 tty->print_raw_cr("Please check if any of your loaded .so files has "
321 "enabled executable stack (see man page execstack(8))");
322 } else {
323 // Accessing stack address below sp may cause SEGV if current
324 // thread has MAP_GROWSDOWN stack. This should only happen when
325 // current thread was created by user code with MAP_GROWSDOWN flag
326 // and then attached to VM. See notes in os_linux.cpp.
327 if (thread->osthread()->expanding_stack() == 0) {
328 thread->osthread()->set_expanding_stack();
329 if (os::Linux::manually_expand_stack(thread, addr)) {
330 thread->osthread()->clear_expanding_stack();
331 return 1;
332 }
333 thread->osthread()->clear_expanding_stack();
334 } else {
335 fatal("recursive segv. expanding stack.");
336 }
337 }
338 }
339 }
340
341 if (thread->thread_state() == _thread_in_Java) {
342 // Java thread running in Java code => find exception handler if any
343 // a fault inside compiled code, the interpreter, or a stub
344
345 if (sig == SIGSEGV && os::is_poll_address((address)info->si_addr)) {
346 stub = SharedRuntime::get_poll_stub(pc);
347 } else if (sig == SIGBUS /* && info->si_code == BUS_OBJERR */) {
348 // BugId 4454115: A read from a MappedByteBuffer can fault
349 // here if the underlying file has been truncated.
350 // Do not crash the VM in such a case.
351 CodeBlob* cb = CodeCache::find_blob_unsafe(pc);
352 nmethod* nm = (cb != NULL && cb->is_nmethod()) ? (nmethod*)cb : NULL;
353 if (nm != NULL && nm->has_unsafe_access()) {
354 stub = StubRoutines::handler_for_unsafe_access();
355 }
356 }
357 else
358
359 #ifdef AMD64
360 if (sig == SIGFPE &&
361 (info->si_code == FPE_INTDIV || info->si_code == FPE_FLTDIV)) {
362 stub =
363 SharedRuntime::
364 continuation_for_implicit_exception(thread,
365 pc,
366 SharedRuntime::
367 IMPLICIT_DIVIDE_BY_ZERO);
368 #else
369 if (sig == SIGFPE /* && info->si_code == FPE_INTDIV */) {
370 // HACK: si_code does not work on linux 2.2.12-20!!!
371 int op = pc[0];
372 if (op == 0xDB) {
373 // FIST
374 // TODO: The encoding of D2I in i486.ad can cause an exception
375 // prior to the fist instruction if there was an invalid operation
376 // pending. We want to dismiss that exception. From the win_32
377 // side it also seems that if it really was the fist causing
378 // the exception that we do the d2i by hand with different
379 // rounding. Seems kind of weird.
380 // NOTE: that we take the exception at the NEXT floating point instruction.
381 assert(pc[0] == 0xDB, "not a FIST opcode");
382 assert(pc[1] == 0x14, "not a FIST opcode");
383 assert(pc[2] == 0x24, "not a FIST opcode");
384 return true;
385 } else if (op == 0xF7) {
386 // IDIV
387 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_DIVIDE_BY_ZERO);
388 } else {
389 // TODO: handle more cases if we are using other x86 instructions
390 // that can generate SIGFPE signal on linux.
391 tty->print_cr("unknown opcode 0x%X with SIGFPE.", op);
392 fatal("please update this code.");
393 }
394 #endif // AMD64
395 } else if (sig == SIGSEGV &&
396 !MacroAssembler::needs_explicit_null_check((intptr_t)info->si_addr)) {
397 // Determination of interpreter/vtable stub/compiled code null exception
398 stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_NULL);
399 }
400 } else if (thread->thread_state() == _thread_in_vm &&
401 sig == SIGBUS && /* info->si_code == BUS_OBJERR && */
402 thread->doing_unsafe_access()) {
403 stub = StubRoutines::handler_for_unsafe_access();
404 }
405
406 // jni_fast_Get<Primitive>Field can trap at certain pc's if a GC kicks in
407 // and the heap gets shrunk before the field access.
408 if ((sig == SIGSEGV) || (sig == SIGBUS)) {
409 address addr = JNI_FastGetField::find_slowcase_pc(pc);
410 if (addr != (address)-1) {
411 stub = addr;
412 }
413 }
414
415 // Check to see if we caught the safepoint code in the
416 // process of write protecting the memory serialization page.
417 // It write enables the page immediately after protecting it
418 // so we can just return to retry the write.
419 if ((sig == SIGSEGV) &&
420 os::is_memory_serialize_page(thread, (address) info->si_addr)) {
421 // Block current thread until the memory serialize page permission restored.
422 os::block_on_serialize_page_trap();
423 return true;
424 }
425 }
426
427 #ifndef AMD64
428 // Execution protection violation
429 //
430 // This should be kept as the last step in the triage. We don't
431 // have a dedicated trap number for a no-execute fault, so be
432 // conservative and allow other handlers the first shot.
433 //
434 // Note: We don't test that info->si_code == SEGV_ACCERR here.
435 // this si_code is so generic that it is almost meaningless; and
436 // the si_code for this condition may change in the future.
437 // Furthermore, a false-positive should be harmless.
438 if (UnguardOnExecutionViolation > 0 &&
439 (sig == SIGSEGV || sig == SIGBUS) &&
440 uc->uc_mcontext.gregs[REG_TRAPNO] == trap_page_fault) {
441 int page_size = os::vm_page_size();
442 address addr = (address) info->si_addr;
443 address pc = os::Linux::ucontext_get_pc(uc);
444 // Make sure the pc and the faulting address are sane.
445 //
446 // If an instruction spans a page boundary, and the page containing
447 // the beginning of the instruction is executable but the following
448 // page is not, the pc and the faulting address might be slightly
449 // different - we still want to unguard the 2nd page in this case.
450 //
451 // 15 bytes seems to be a (very) safe value for max instruction size.
452 bool pc_is_near_addr =
453 (pointer_delta((void*) addr, (void*) pc, sizeof(char)) < 15);
454 bool instr_spans_page_boundary =
455 (align_size_down((intptr_t) pc ^ (intptr_t) addr,
456 (intptr_t) page_size) > 0);
457
458 if (pc == addr || (pc_is_near_addr && instr_spans_page_boundary)) {
459 static volatile address last_addr =
460 (address) os::non_memory_address_word();
461
462 // In conservative mode, don't unguard unless the address is in the VM
463 if (addr != last_addr &&
464 (UnguardOnExecutionViolation > 1 || os::address_is_in_vm(addr))) {
465
466 // Set memory to RWX and retry
467 address page_start =
468 (address) align_size_down((intptr_t) addr, (intptr_t) page_size);
469 bool res = os::protect_memory((char*) page_start, page_size,
470 os::MEM_PROT_RWX);
471
472 if (PrintMiscellaneous && Verbose) {
473 char buf[256];
474 jio_snprintf(buf, sizeof(buf), "Execution protection violation "
475 "at " INTPTR_FORMAT
476 ", unguarding " INTPTR_FORMAT ": %s, errno=%d", addr,
477 page_start, (res ? "success" : "failed"), errno);
478 tty->print_raw_cr(buf);
479 }
480 stub = pc;
481
482 // Set last_addr so if we fault again at the same address, we don't end
483 // up in an endless loop.
484 //
485 // There are two potential complications here. Two threads trapping at
486 // the same address at the same time could cause one of the threads to
487 // think it already unguarded, and abort the VM. Likely very rare.
488 //
489 // The other race involves two threads alternately trapping at
490 // different addresses and failing to unguard the page, resulting in
491 // an endless loop. This condition is probably even more unlikely than
492 // the first.
493 //
494 // Although both cases could be avoided by using locks or thread local
495 // last_addr, these solutions are unnecessary complication: this
496 // handler is a best-effort safety net, not a complete solution. It is
497 // disabled by default and should only be used as a workaround in case
498 // we missed any no-execute-unsafe VM code.
499
500 last_addr = addr;
501 }
502 }
503 }
504 #endif // !AMD64
505
506 if (stub != NULL) {
507 // save all thread context in case we need to restore it
508 if (thread != NULL) thread->set_saved_exception_pc(pc);
509
510 uc->uc_mcontext.gregs[REG_PC] = (greg_t)stub;
511 return true;
512 }
513
514 // signal-chaining
515 if (os::Linux::chained_handler(sig, info, ucVoid)) {
516 return true;
517 }
518
519 if (!abort_if_unrecognized) {
520 // caller wants another chance, so give it to him
521 return false;
522 }
523
524 if (pc == NULL && uc != NULL) {
525 pc = os::Linux::ucontext_get_pc(uc);
526 }
527
528 // unmask current signal
529 sigset_t newset;
530 sigemptyset(&newset);
531 sigaddset(&newset, sig);
532 sigprocmask(SIG_UNBLOCK, &newset, NULL);
533
534 VMError err(t, sig, pc, info, ucVoid);
535 err.report_and_die();
536
537 ShouldNotReachHere();
538 }
539
540 void os::Linux::init_thread_fpu_state(void) {
541 #ifndef AMD64
542 // set fpu to 53 bit precision
543 set_fpu_control_word(0x27f);
544 #endif // !AMD64
545 }
546
547 int os::Linux::get_fpu_control_word(void) {
548 #ifdef AMD64
549 return 0;
550 #else
551 int fpu_control;
552 _FPU_GETCW(fpu_control);
553 return fpu_control & 0xffff;
554 #endif // AMD64
555 }
556
557 void os::Linux::set_fpu_control_word(int fpu_control) {
558 #ifndef AMD64
559 _FPU_SETCW(fpu_control);
560 #endif // !AMD64
561 }
562
563 // Check that the linux kernel version is 2.4 or higher since earlier
564 // versions do not support SSE without patches.
565 bool os::supports_sse() {
566 #ifdef AMD64
567 return true;
568 #else
569 struct utsname uts;
570 if( uname(&uts) != 0 ) return false; // uname fails?
571 char *minor_string;
572 int major = strtol(uts.release,&minor_string,10);
573 int minor = strtol(minor_string+1,NULL,10);
574 bool result = (major > 2 || (major==2 && minor >= 4));
575 #ifndef PRODUCT
576 if (PrintMiscellaneous && Verbose) {
577 tty->print("OS version is %d.%d, which %s support SSE/SSE2\n",
578 major,minor, result ? "DOES" : "does NOT");
579 }
580 #endif
581 return result;
582 #endif // AMD64
583 }
584
585 bool os::is_allocatable(size_t bytes) {
586 #ifdef AMD64
587 // unused on amd64?
588 return true;
589 #else
590
591 if (bytes < 2 * G) {
592 return true;
593 }
594
595 char* addr = reserve_memory(bytes, NULL);
596
597 if (addr != NULL) {
598 release_memory(addr, bytes);
599 }
600
601 return addr != NULL;
602 #endif // AMD64
603 }
604
605 ////////////////////////////////////////////////////////////////////////////////
606 // thread stack
607
608 #ifdef AMD64
609 size_t os::Linux::min_stack_allowed = 64 * K;
610
611 // amd64: pthread on amd64 is always in floating stack mode
612 bool os::Linux::supports_variable_stack_size() { return true; }
613 #else
614 size_t os::Linux::min_stack_allowed = (48 DEBUG_ONLY(+4))*K;
615
616 #ifdef __GNUC__
617 #define GET_GS() ({int gs; __asm__ volatile("movw %%gs, %w0":"=q"(gs)); gs&0xffff;})
618 #endif
619
620 // Test if pthread library can support variable thread stack size. LinuxThreads
621 // in fixed stack mode allocates 2M fixed slot for each thread. LinuxThreads
622 // in floating stack mode and NPTL support variable stack size.
623 bool os::Linux::supports_variable_stack_size() {
624 if (os::Linux::is_NPTL()) {
625 // NPTL, yes
626 return true;
627
628 } else {
629 // Note: We can't control default stack size when creating a thread.
630 // If we use non-default stack size (pthread_attr_setstacksize), both
631 // floating stack and non-floating stack LinuxThreads will return the
632 // same value. This makes it impossible to implement this function by
633 // detecting thread stack size directly.
634 //
635 // An alternative approach is to check %gs. Fixed-stack LinuxThreads
636 // do not use %gs, so its value is 0. Floating-stack LinuxThreads use
637 // %gs (either as LDT selector or GDT selector, depending on kernel)
638 // to access thread specific data.
639 //
640 // Note that %gs is a reserved glibc register since early 2001, so
641 // applications are not allowed to change its value (Ulrich Drepper from
642 // Redhat confirmed that all known offenders have been modified to use
643 // either %fs or TSD). In the worst case scenario, when VM is embedded in
644 // a native application that plays with %gs, we might see non-zero %gs
645 // even LinuxThreads is running in fixed stack mode. As the result, we'll
646 // return true and skip _thread_safety_check(), so we may not be able to
647 // detect stack-heap collisions. But otherwise it's harmless.
648 //
649 #ifdef __GNUC__
650 return (GET_GS() != 0);
651 #else
652 return false;
653 #endif
654 }
655 }
656 #endif // AMD64
657
658 // return default stack size for thr_type
659 size_t os::Linux::default_stack_size(os::ThreadType thr_type) {
660 // default stack size (compiler thread needs larger stack)
661 #ifdef AMD64
662 size_t s = (thr_type == os::compiler_thread ? 4 * M : 1 * M);
663 #else
664 size_t s = (thr_type == os::compiler_thread ? 2 * M : 512 * K);
665 #endif // AMD64
666 return s;
667 }
668
669 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
670 // Creating guard page is very expensive. Java thread has HotSpot
671 // guard page, only enable glibc guard page for non-Java threads.
672 return (thr_type == java_thread ? 0 : page_size());
673 }
674
675 // Java thread:
676 //
677 // Low memory addresses
678 // +------------------------+
679 // | |\ JavaThread created by VM does not have glibc
680 // | glibc guard page | - guard, attached Java thread usually has
681 // | |/ 1 page glibc guard.
682 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
683 // | |\
684 // | HotSpot Guard Pages | - red and yellow pages
685 // | |/
686 // +------------------------+ JavaThread::stack_yellow_zone_base()
687 // | |\
688 // | Normal Stack | -
689 // | |/
690 // P2 +------------------------+ Thread::stack_base()
691 //
692 // Non-Java thread:
693 //
694 // Low memory addresses
695 // +------------------------+
696 // | |\
697 // | glibc guard page | - usually 1 page
698 // | |/
699 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
700 // | |\
701 // | Normal Stack | -
702 // | |/
703 // P2 +------------------------+ Thread::stack_base()
704 //
705 // ** P1 (aka bottom) and size ( P2 = P1 - size) are the address and stack size returned from
706 // pthread_attr_getstack()
707
708 static void current_stack_region(address * bottom, size_t * size) {
709 if (os::Linux::is_initial_thread()) {
710 // initial thread needs special handling because pthread_getattr_np()
711 // may return bogus value.
712 *bottom = os::Linux::initial_thread_stack_bottom();
713 *size = os::Linux::initial_thread_stack_size();
714 } else {
715 pthread_attr_t attr;
716
717 int rslt = pthread_getattr_np(pthread_self(), &attr);
718
719 // JVM needs to know exact stack location, abort if it fails
720 if (rslt != 0) {
721 if (rslt == ENOMEM) {
722 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
723 } else {
724 fatal(err_msg("pthread_getattr_np failed with errno = %d", rslt));
725 }
726 }
727
728 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
729 fatal("Can not locate current stack attributes!");
730 }
731
732 pthread_attr_destroy(&attr);
733
734 }
735 assert(os::current_stack_pointer() >= *bottom &&
736 os::current_stack_pointer() < *bottom + *size, "just checking");
737 }
738
739 address os::current_stack_base() {
740 address bottom;
741 size_t size;
742 current_stack_region(&bottom, &size);
743 return (bottom + size);
744 }
745
746 size_t os::current_stack_size() {
747 // stack size includes normal stack and HotSpot guard pages
748 address bottom;
749 size_t size;
750 current_stack_region(&bottom, &size);
751 return size;
752 }
753
754 /////////////////////////////////////////////////////////////////////////////
755 // helper functions for fatal error handler
756
757 void os::print_context(outputStream *st, void *context) {
758 if (context == NULL) return;
759
760 ucontext_t *uc = (ucontext_t*)context;
761 st->print_cr("Registers:");
762 #ifdef AMD64
763 st->print( "RAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RAX]);
764 st->print(", RBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBX]);
765 st->print(", RCX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RCX]);
766 st->print(", RDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDX]);
767 st->cr();
768 st->print( "RSP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSP]);
769 st->print(", RBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBP]);
770 st->print(", RSI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSI]);
771 st->print(", RDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDI]);
772 st->cr();
773 st->print( "R8 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R8]);
774 st->print(", R9 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R9]);
775 st->print(", R10=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R10]);
776 st->print(", R11=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R11]);
777 st->cr();
778 st->print( "R12=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R12]);
779 st->print(", R13=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R13]);
780 st->print(", R14=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R14]);
781 st->print(", R15=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R15]);
782 st->cr();
783 st->print( "RIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RIP]);
784 st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]);
785 st->print(", CSGSFS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_CSGSFS]);
786 st->print(", ERR=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ERR]);
787 st->cr();
788 st->print(" TRAPNO=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_TRAPNO]);
789 #else
790 st->print( "EAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EAX]);
791 st->print(", EBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBX]);
792 st->print(", ECX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ECX]);
793 st->print(", EDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDX]);
794 st->cr();
795 st->print( "ESP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_UESP]);
796 st->print(", EBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBP]);
797 st->print(", ESI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ESI]);
798 st->print(", EDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDI]);
799 st->cr();
800 st->print( "EIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EIP]);
801 st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]);
802 st->print(", CR2=" INTPTR_FORMAT, uc->uc_mcontext.cr2);
803 #endif // AMD64
804 st->cr();
805 st->cr();
806
807 intptr_t *sp = (intptr_t *)os::Linux::ucontext_get_sp(uc);
808 st->print_cr("Top of Stack: (sp=" PTR_FORMAT ")", sp);
809 print_hex_dump(st, (address)sp, (address)(sp + 8*sizeof(intptr_t)), sizeof(intptr_t));
810 st->cr();
811
812 // Note: it may be unsafe to inspect memory near pc. For example, pc may
813 // point to garbage if entry point in an nmethod is corrupted. Leave
814 // this at the end, and hope for the best.
815 address pc = os::Linux::ucontext_get_pc(uc);
816 st->print_cr("Instructions: (pc=" PTR_FORMAT ")", pc);
817 print_hex_dump(st, pc - 32, pc + 32, sizeof(char));
818 }
819
820 void os::print_register_info(outputStream *st, void *context) {
821 if (context == NULL) return;
822
823 ucontext_t *uc = (ucontext_t*)context;
824
825 st->print_cr("Register to memory mapping:");
826 st->cr();
827
828 // this is horrendously verbose but the layout of the registers in the
829 // context does not match how we defined our abstract Register set, so
830 // we can't just iterate through the gregs area
831
832 // this is only for the "general purpose" registers
833
834 #ifdef AMD64
835 st->print("RAX="); print_location(st, uc->uc_mcontext.gregs[REG_RAX]);
836 st->print("RBX="); print_location(st, uc->uc_mcontext.gregs[REG_RBX]);
837 st->print("RCX="); print_location(st, uc->uc_mcontext.gregs[REG_RCX]);
838 st->print("RDX="); print_location(st, uc->uc_mcontext.gregs[REG_RDX]);
839 st->print("RSP="); print_location(st, uc->uc_mcontext.gregs[REG_RSP]);
840 st->print("RBP="); print_location(st, uc->uc_mcontext.gregs[REG_RBP]);
841 st->print("RSI="); print_location(st, uc->uc_mcontext.gregs[REG_RSI]);
842 st->print("RDI="); print_location(st, uc->uc_mcontext.gregs[REG_RDI]);
843 st->print("R8 ="); print_location(st, uc->uc_mcontext.gregs[REG_R8]);
844 st->print("R9 ="); print_location(st, uc->uc_mcontext.gregs[REG_R9]);
845 st->print("R10="); print_location(st, uc->uc_mcontext.gregs[REG_R10]);
846 st->print("R11="); print_location(st, uc->uc_mcontext.gregs[REG_R11]);
847 st->print("R12="); print_location(st, uc->uc_mcontext.gregs[REG_R12]);
848 st->print("R13="); print_location(st, uc->uc_mcontext.gregs[REG_R13]);
849 st->print("R14="); print_location(st, uc->uc_mcontext.gregs[REG_R14]);
850 st->print("R15="); print_location(st, uc->uc_mcontext.gregs[REG_R15]);
851 #else
852 st->print("EAX="); print_location(st, uc->uc_mcontext.gregs[REG_EAX]);
853 st->print("EBX="); print_location(st, uc->uc_mcontext.gregs[REG_EBX]);
854 st->print("ECX="); print_location(st, uc->uc_mcontext.gregs[REG_ECX]);
855 st->print("EDX="); print_location(st, uc->uc_mcontext.gregs[REG_EDX]);
856 st->print("ESP="); print_location(st, uc->uc_mcontext.gregs[REG_ESP]);
857 st->print("EBP="); print_location(st, uc->uc_mcontext.gregs[REG_EBP]);
858 st->print("ESI="); print_location(st, uc->uc_mcontext.gregs[REG_ESI]);
859 st->print("EDI="); print_location(st, uc->uc_mcontext.gregs[REG_EDI]);
860 #endif // AMD64
861
862 st->cr();
863 }
864
865 void os::setup_fpu() {
866 #ifndef AMD64
867 address fpu_cntrl = StubRoutines::addr_fpu_cntrl_wrd_std();
868 __asm__ volatile ( "fldcw (%0)" :
869 : "r" (fpu_cntrl) : "memory");
870 #endif // !AMD64
871 }
872
873 #ifndef PRODUCT
874 void os::verify_stack_alignment() {
875 #ifdef AMD64
876 assert(((intptr_t)os::current_stack_pointer() & (StackAlignmentInBytes-1)) == 0, "incorrect stack alignment");
877 #endif
878 }
879 #endif
880
881
882 /*
883 * IA32 only: execute code at a high address in case buggy NX emulation is present. I.e. avoid CS limit
884 * updates (JDK-8023956).
885 */
886 void os::workaround_expand_exec_shield_cs_limit() {
887 #if defined(IA32)
888 size_t page_size = os::vm_page_size();
889 /*
890 * Take the highest VA the OS will give us and exec
891 *
892 * Although using -(pagesz) as mmap hint works on newer kernel as you would
893 * think, older variants affected by this work-around don't (search forward only).
894 *
895 * On the affected distributions, we understand the memory layout to be:
896 *
897 * TASK_LIMIT= 3G, main stack base close to TASK_LIMT.
898 *
899 * A few pages south main stack will do it.
900 *
901 * If we are embedded in an app other than launcher (initial != main stack),
902 * we don't have much control or understanding of the address space, just let it slide.
903 */
904 char* hint = (char*) (Linux::initial_thread_stack_bottom() -
905 ((StackYellowPages + StackRedPages + 1) * page_size));
906 char* codebuf = os::reserve_memory(page_size, hint);
907 if ( (codebuf == NULL) || (!os::commit_memory(codebuf, page_size, true)) ) {
908 return; // No matter, we tried, best effort.
909 }
910
911 MemTracker::record_virtual_memory_type((address)codebuf, mtInternal);
912
913 if (PrintMiscellaneous && (Verbose || WizardMode)) {
914 tty->print_cr("[CS limit NX emulation work-around, exec code at: %p]", codebuf);
915 }
916
917 // Some code to exec: the 'ret' instruction
918 codebuf[0] = 0xC3;
919
920 // Call the code in the codebuf
921 __asm__ volatile("call *%0" : : "r"(codebuf));
922
923 // keep the page mapped so CS limit isn't reduced.
924 #endif
925 }