1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp Wed Apr 27 01:25:04 2016 +0800
1.3 @@ -0,0 +1,3043 @@
1.4 +/*
1.5 + * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
1.23 + * or visit www.oracle.com if you need additional information or have any
1.24 + * questions.
1.25 + *
1.26 + */
1.27 +
1.28 +#include "precompiled.hpp"
1.29 +#include "asm/macroAssembler.hpp"
1.30 +#include "asm/macroAssembler.inline.hpp"
1.31 +#include "interpreter/interpreter.hpp"
1.32 +#include "nativeInst_x86.hpp"
1.33 +#include "oops/instanceOop.hpp"
1.34 +#include "oops/method.hpp"
1.35 +#include "oops/objArrayKlass.hpp"
1.36 +#include "oops/oop.inline.hpp"
1.37 +#include "prims/methodHandles.hpp"
1.38 +#include "runtime/frame.inline.hpp"
1.39 +#include "runtime/handles.inline.hpp"
1.40 +#include "runtime/sharedRuntime.hpp"
1.41 +#include "runtime/stubCodeGenerator.hpp"
1.42 +#include "runtime/stubRoutines.hpp"
1.43 +#include "runtime/thread.inline.hpp"
1.44 +#include "utilities/top.hpp"
1.45 +#ifdef COMPILER2
1.46 +#include "opto/runtime.hpp"
1.47 +#endif
1.48 +
1.49 +// Declaration and definition of StubGenerator (no .hpp file).
1.50 +// For a more detailed description of the stub routine structure
1.51 +// see the comment in stubRoutines.hpp
1.52 +
1.53 +#define __ _masm->
1.54 +#define a__ ((Assembler*)_masm)->
1.55 +
1.56 +#ifdef PRODUCT
1.57 +#define BLOCK_COMMENT(str) /* nothing */
1.58 +#else
1.59 +#define BLOCK_COMMENT(str) __ block_comment(str)
1.60 +#endif
1.61 +
1.62 +#define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
1.63 +
1.64 +const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
1.65 +const int FPU_CNTRL_WRD_MASK = 0xFFFF;
1.66 +
1.67 +// -------------------------------------------------------------------------------------------------------------------------
1.68 +// Stub Code definitions
1.69 +
1.70 +static address handle_unsafe_access() {
1.71 + JavaThread* thread = JavaThread::current();
1.72 + address pc = thread->saved_exception_pc();
1.73 + // pc is the instruction which we must emulate
1.74 + // doing a no-op is fine: return garbage from the load
1.75 + // therefore, compute npc
1.76 + address npc = Assembler::locate_next_instruction(pc);
1.77 +
1.78 + // request an async exception
1.79 + thread->set_pending_unsafe_access_error();
1.80 +
1.81 + // return address of next instruction to execute
1.82 + return npc;
1.83 +}
1.84 +
1.85 +class StubGenerator: public StubCodeGenerator {
1.86 + private:
1.87 +
1.88 +#ifdef PRODUCT
1.89 +#define inc_counter_np(counter) ((void)0)
1.90 +#else
1.91 + void inc_counter_np_(int& counter) {
1.92 + __ incrementl(ExternalAddress((address)&counter));
1.93 + }
1.94 +#define inc_counter_np(counter) \
1.95 + BLOCK_COMMENT("inc_counter " #counter); \
1.96 + inc_counter_np_(counter);
1.97 +#endif //PRODUCT
1.98 +
1.99 + void inc_copy_counter_np(BasicType t) {
1.100 +#ifndef PRODUCT
1.101 + switch (t) {
1.102 + case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
1.103 + case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
1.104 + case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
1.105 + case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
1.106 + case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
1.107 + }
1.108 + ShouldNotReachHere();
1.109 +#endif //PRODUCT
1.110 + }
1.111 +
1.112 + //------------------------------------------------------------------------------------------------------------------------
1.113 + // Call stubs are used to call Java from C
1.114 + //
1.115 + // [ return_from_Java ] <--- rsp
1.116 + // [ argument word n ]
1.117 + // ...
1.118 + // -N [ argument word 1 ]
1.119 + // -7 [ Possible padding for stack alignment ]
1.120 + // -6 [ Possible padding for stack alignment ]
1.121 + // -5 [ Possible padding for stack alignment ]
1.122 + // -4 [ mxcsr save ] <--- rsp_after_call
1.123 + // -3 [ saved rbx, ]
1.124 + // -2 [ saved rsi ]
1.125 + // -1 [ saved rdi ]
1.126 + // 0 [ saved rbp, ] <--- rbp,
1.127 + // 1 [ return address ]
1.128 + // 2 [ ptr. to call wrapper ]
1.129 + // 3 [ result ]
1.130 + // 4 [ result_type ]
1.131 + // 5 [ method ]
1.132 + // 6 [ entry_point ]
1.133 + // 7 [ parameters ]
1.134 + // 8 [ parameter_size ]
1.135 + // 9 [ thread ]
1.136 +
1.137 +
1.138 + address generate_call_stub(address& return_address) {
1.139 + StubCodeMark mark(this, "StubRoutines", "call_stub");
1.140 + address start = __ pc();
1.141 +
1.142 + // stub code parameters / addresses
1.143 + assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
1.144 + bool sse_save = false;
1.145 + const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
1.146 + const int locals_count_in_bytes (4*wordSize);
1.147 + const Address mxcsr_save (rbp, -4 * wordSize);
1.148 + const Address saved_rbx (rbp, -3 * wordSize);
1.149 + const Address saved_rsi (rbp, -2 * wordSize);
1.150 + const Address saved_rdi (rbp, -1 * wordSize);
1.151 + const Address result (rbp, 3 * wordSize);
1.152 + const Address result_type (rbp, 4 * wordSize);
1.153 + const Address method (rbp, 5 * wordSize);
1.154 + const Address entry_point (rbp, 6 * wordSize);
1.155 + const Address parameters (rbp, 7 * wordSize);
1.156 + const Address parameter_size(rbp, 8 * wordSize);
1.157 + const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
1.158 + sse_save = UseSSE > 0;
1.159 +
1.160 + // stub code
1.161 + __ enter();
1.162 + __ movptr(rcx, parameter_size); // parameter counter
1.163 + __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
1.164 + __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
1.165 + __ subptr(rsp, rcx);
1.166 + __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
1.167 +
1.168 + // save rdi, rsi, & rbx, according to C calling conventions
1.169 + __ movptr(saved_rdi, rdi);
1.170 + __ movptr(saved_rsi, rsi);
1.171 + __ movptr(saved_rbx, rbx);
1.172 + // save and initialize %mxcsr
1.173 + if (sse_save) {
1.174 + Label skip_ldmx;
1.175 + __ stmxcsr(mxcsr_save);
1.176 + __ movl(rax, mxcsr_save);
1.177 + __ andl(rax, MXCSR_MASK); // Only check control and mask bits
1.178 + ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
1.179 + __ cmp32(rax, mxcsr_std);
1.180 + __ jcc(Assembler::equal, skip_ldmx);
1.181 + __ ldmxcsr(mxcsr_std);
1.182 + __ bind(skip_ldmx);
1.183 + }
1.184 +
1.185 + // make sure the control word is correct.
1.186 + __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1.187 +
1.188 +#ifdef ASSERT
1.189 + // make sure we have no pending exceptions
1.190 + { Label L;
1.191 + __ movptr(rcx, thread);
1.192 + __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1.193 + __ jcc(Assembler::equal, L);
1.194 + __ stop("StubRoutines::call_stub: entered with pending exception");
1.195 + __ bind(L);
1.196 + }
1.197 +#endif
1.198 +
1.199 + // pass parameters if any
1.200 + BLOCK_COMMENT("pass parameters if any");
1.201 + Label parameters_done;
1.202 + __ movl(rcx, parameter_size); // parameter counter
1.203 + __ testl(rcx, rcx);
1.204 + __ jcc(Assembler::zero, parameters_done);
1.205 +
1.206 + // parameter passing loop
1.207 +
1.208 + Label loop;
1.209 + // Copy Java parameters in reverse order (receiver last)
1.210 + // Note that the argument order is inverted in the process
1.211 + // source is rdx[rcx: N-1..0]
1.212 + // dest is rsp[rbx: 0..N-1]
1.213 +
1.214 + __ movptr(rdx, parameters); // parameter pointer
1.215 + __ xorptr(rbx, rbx);
1.216 +
1.217 + __ BIND(loop);
1.218 +
1.219 + // get parameter
1.220 + __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
1.221 + __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
1.222 + Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
1.223 + __ increment(rbx);
1.224 + __ decrement(rcx);
1.225 + __ jcc(Assembler::notZero, loop);
1.226 +
1.227 + // call Java function
1.228 + __ BIND(parameters_done);
1.229 + __ movptr(rbx, method); // get Method*
1.230 + __ movptr(rax, entry_point); // get entry_point
1.231 + __ mov(rsi, rsp); // set sender sp
1.232 + BLOCK_COMMENT("call Java function");
1.233 + __ call(rax);
1.234 +
1.235 + BLOCK_COMMENT("call_stub_return_address:");
1.236 + return_address = __ pc();
1.237 +
1.238 +#ifdef COMPILER2
1.239 + {
1.240 + Label L_skip;
1.241 + if (UseSSE >= 2) {
1.242 + __ verify_FPU(0, "call_stub_return");
1.243 + } else {
1.244 + for (int i = 1; i < 8; i++) {
1.245 + __ ffree(i);
1.246 + }
1.247 +
1.248 + // UseSSE <= 1 so double result should be left on TOS
1.249 + __ movl(rsi, result_type);
1.250 + __ cmpl(rsi, T_DOUBLE);
1.251 + __ jcc(Assembler::equal, L_skip);
1.252 + if (UseSSE == 0) {
1.253 + // UseSSE == 0 so float result should be left on TOS
1.254 + __ cmpl(rsi, T_FLOAT);
1.255 + __ jcc(Assembler::equal, L_skip);
1.256 + }
1.257 + __ ffree(0);
1.258 + }
1.259 + __ BIND(L_skip);
1.260 + }
1.261 +#endif // COMPILER2
1.262 +
1.263 + // store result depending on type
1.264 + // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
1.265 + __ movptr(rdi, result);
1.266 + Label is_long, is_float, is_double, exit;
1.267 + __ movl(rsi, result_type);
1.268 + __ cmpl(rsi, T_LONG);
1.269 + __ jcc(Assembler::equal, is_long);
1.270 + __ cmpl(rsi, T_FLOAT);
1.271 + __ jcc(Assembler::equal, is_float);
1.272 + __ cmpl(rsi, T_DOUBLE);
1.273 + __ jcc(Assembler::equal, is_double);
1.274 +
1.275 + // handle T_INT case
1.276 + __ movl(Address(rdi, 0), rax);
1.277 + __ BIND(exit);
1.278 +
1.279 + // check that FPU stack is empty
1.280 + __ verify_FPU(0, "generate_call_stub");
1.281 +
1.282 + // pop parameters
1.283 + __ lea(rsp, rsp_after_call);
1.284 +
1.285 + // restore %mxcsr
1.286 + if (sse_save) {
1.287 + __ ldmxcsr(mxcsr_save);
1.288 + }
1.289 +
1.290 + // restore rdi, rsi and rbx,
1.291 + __ movptr(rbx, saved_rbx);
1.292 + __ movptr(rsi, saved_rsi);
1.293 + __ movptr(rdi, saved_rdi);
1.294 + __ addptr(rsp, 4*wordSize);
1.295 +
1.296 + // return
1.297 + __ pop(rbp);
1.298 + __ ret(0);
1.299 +
1.300 + // handle return types different from T_INT
1.301 + __ BIND(is_long);
1.302 + __ movl(Address(rdi, 0 * wordSize), rax);
1.303 + __ movl(Address(rdi, 1 * wordSize), rdx);
1.304 + __ jmp(exit);
1.305 +
1.306 + __ BIND(is_float);
1.307 + // interpreter uses xmm0 for return values
1.308 + if (UseSSE >= 1) {
1.309 + __ movflt(Address(rdi, 0), xmm0);
1.310 + } else {
1.311 + __ fstp_s(Address(rdi, 0));
1.312 + }
1.313 + __ jmp(exit);
1.314 +
1.315 + __ BIND(is_double);
1.316 + // interpreter uses xmm0 for return values
1.317 + if (UseSSE >= 2) {
1.318 + __ movdbl(Address(rdi, 0), xmm0);
1.319 + } else {
1.320 + __ fstp_d(Address(rdi, 0));
1.321 + }
1.322 + __ jmp(exit);
1.323 +
1.324 + return start;
1.325 + }
1.326 +
1.327 +
1.328 + //------------------------------------------------------------------------------------------------------------------------
1.329 + // Return point for a Java call if there's an exception thrown in Java code.
1.330 + // The exception is caught and transformed into a pending exception stored in
1.331 + // JavaThread that can be tested from within the VM.
1.332 + //
1.333 + // Note: Usually the parameters are removed by the callee. In case of an exception
1.334 + // crossing an activation frame boundary, that is not the case if the callee
1.335 + // is compiled code => need to setup the rsp.
1.336 + //
1.337 + // rax,: exception oop
1.338 +
1.339 + address generate_catch_exception() {
1.340 + StubCodeMark mark(this, "StubRoutines", "catch_exception");
1.341 + const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
1.342 + const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
1.343 + address start = __ pc();
1.344 +
1.345 + // get thread directly
1.346 + __ movptr(rcx, thread);
1.347 +#ifdef ASSERT
1.348 + // verify that threads correspond
1.349 + { Label L;
1.350 + __ get_thread(rbx);
1.351 + __ cmpptr(rbx, rcx);
1.352 + __ jcc(Assembler::equal, L);
1.353 + __ stop("StubRoutines::catch_exception: threads must correspond");
1.354 + __ bind(L);
1.355 + }
1.356 +#endif
1.357 + // set pending exception
1.358 + __ verify_oop(rax);
1.359 + __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
1.360 + __ lea(Address(rcx, Thread::exception_file_offset ()),
1.361 + ExternalAddress((address)__FILE__));
1.362 + __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
1.363 + // complete return to VM
1.364 + assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
1.365 + __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
1.366 +
1.367 + return start;
1.368 + }
1.369 +
1.370 +
1.371 + //------------------------------------------------------------------------------------------------------------------------
1.372 + // Continuation point for runtime calls returning with a pending exception.
1.373 + // The pending exception check happened in the runtime or native call stub.
1.374 + // The pending exception in Thread is converted into a Java-level exception.
1.375 + //
1.376 + // Contract with Java-level exception handlers:
1.377 + // rax: exception
1.378 + // rdx: throwing pc
1.379 + //
1.380 + // NOTE: At entry of this stub, exception-pc must be on stack !!
1.381 +
1.382 + address generate_forward_exception() {
1.383 + StubCodeMark mark(this, "StubRoutines", "forward exception");
1.384 + address start = __ pc();
1.385 + const Register thread = rcx;
1.386 +
1.387 + // other registers used in this stub
1.388 + const Register exception_oop = rax;
1.389 + const Register handler_addr = rbx;
1.390 + const Register exception_pc = rdx;
1.391 +
1.392 + // Upon entry, the sp points to the return address returning into Java
1.393 + // (interpreted or compiled) code; i.e., the return address becomes the
1.394 + // throwing pc.
1.395 + //
1.396 + // Arguments pushed before the runtime call are still on the stack but
1.397 + // the exception handler will reset the stack pointer -> ignore them.
1.398 + // A potential result in registers can be ignored as well.
1.399 +
1.400 +#ifdef ASSERT
1.401 + // make sure this code is only executed if there is a pending exception
1.402 + { Label L;
1.403 + __ get_thread(thread);
1.404 + __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1.405 + __ jcc(Assembler::notEqual, L);
1.406 + __ stop("StubRoutines::forward exception: no pending exception (1)");
1.407 + __ bind(L);
1.408 + }
1.409 +#endif
1.410 +
1.411 + // compute exception handler into rbx,
1.412 + __ get_thread(thread);
1.413 + __ movptr(exception_pc, Address(rsp, 0));
1.414 + BLOCK_COMMENT("call exception_handler_for_return_address");
1.415 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
1.416 + __ mov(handler_addr, rax);
1.417 +
1.418 + // setup rax & rdx, remove return address & clear pending exception
1.419 + __ get_thread(thread);
1.420 + __ pop(exception_pc);
1.421 + __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
1.422 + __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1.423 +
1.424 +#ifdef ASSERT
1.425 + // make sure exception is set
1.426 + { Label L;
1.427 + __ testptr(exception_oop, exception_oop);
1.428 + __ jcc(Assembler::notEqual, L);
1.429 + __ stop("StubRoutines::forward exception: no pending exception (2)");
1.430 + __ bind(L);
1.431 + }
1.432 +#endif
1.433 +
1.434 + // Verify that there is really a valid exception in RAX.
1.435 + __ verify_oop(exception_oop);
1.436 +
1.437 + // continue at exception handler (return address removed)
1.438 + // rax: exception
1.439 + // rbx: exception handler
1.440 + // rdx: throwing pc
1.441 + __ jmp(handler_addr);
1.442 +
1.443 + return start;
1.444 + }
1.445 +
1.446 +
1.447 + //----------------------------------------------------------------------------------------------------
1.448 + // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
1.449 + //
1.450 + // xchg exists as far back as 8086, lock needed for MP only
1.451 + // Stack layout immediately after call:
1.452 + //
1.453 + // 0 [ret addr ] <--- rsp
1.454 + // 1 [ ex ]
1.455 + // 2 [ dest ]
1.456 + //
1.457 + // Result: *dest <- ex, return (old *dest)
1.458 + //
1.459 + // Note: win32 does not currently use this code
1.460 +
1.461 + address generate_atomic_xchg() {
1.462 + StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
1.463 + address start = __ pc();
1.464 +
1.465 + __ push(rdx);
1.466 + Address exchange(rsp, 2 * wordSize);
1.467 + Address dest_addr(rsp, 3 * wordSize);
1.468 + __ movl(rax, exchange);
1.469 + __ movptr(rdx, dest_addr);
1.470 + __ xchgl(rax, Address(rdx, 0));
1.471 + __ pop(rdx);
1.472 + __ ret(0);
1.473 +
1.474 + return start;
1.475 + }
1.476 +
1.477 + //----------------------------------------------------------------------------------------------------
1.478 + // Support for void verify_mxcsr()
1.479 + //
1.480 + // This routine is used with -Xcheck:jni to verify that native
1.481 + // JNI code does not return to Java code without restoring the
1.482 + // MXCSR register to our expected state.
1.483 +
1.484 +
1.485 + address generate_verify_mxcsr() {
1.486 + StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
1.487 + address start = __ pc();
1.488 +
1.489 + const Address mxcsr_save(rsp, 0);
1.490 +
1.491 + if (CheckJNICalls && UseSSE > 0 ) {
1.492 + Label ok_ret;
1.493 + ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
1.494 + __ push(rax);
1.495 + __ subptr(rsp, wordSize); // allocate a temp location
1.496 + __ stmxcsr(mxcsr_save);
1.497 + __ movl(rax, mxcsr_save);
1.498 + __ andl(rax, MXCSR_MASK);
1.499 + __ cmp32(rax, mxcsr_std);
1.500 + __ jcc(Assembler::equal, ok_ret);
1.501 +
1.502 + __ warn("MXCSR changed by native JNI code.");
1.503 +
1.504 + __ ldmxcsr(mxcsr_std);
1.505 +
1.506 + __ bind(ok_ret);
1.507 + __ addptr(rsp, wordSize);
1.508 + __ pop(rax);
1.509 + }
1.510 +
1.511 + __ ret(0);
1.512 +
1.513 + return start;
1.514 + }
1.515 +
1.516 +
1.517 + //---------------------------------------------------------------------------
1.518 + // Support for void verify_fpu_cntrl_wrd()
1.519 + //
1.520 + // This routine is used with -Xcheck:jni to verify that native
1.521 + // JNI code does not return to Java code without restoring the
1.522 + // FP control word to our expected state.
1.523 +
1.524 + address generate_verify_fpu_cntrl_wrd() {
1.525 + StubCodeMark mark(this, "StubRoutines", "verify_spcw");
1.526 + address start = __ pc();
1.527 +
1.528 + const Address fpu_cntrl_wrd_save(rsp, 0);
1.529 +
1.530 + if (CheckJNICalls) {
1.531 + Label ok_ret;
1.532 + __ push(rax);
1.533 + __ subptr(rsp, wordSize); // allocate a temp location
1.534 + __ fnstcw(fpu_cntrl_wrd_save);
1.535 + __ movl(rax, fpu_cntrl_wrd_save);
1.536 + __ andl(rax, FPU_CNTRL_WRD_MASK);
1.537 + ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
1.538 + __ cmp32(rax, fpu_std);
1.539 + __ jcc(Assembler::equal, ok_ret);
1.540 +
1.541 + __ warn("Floating point control word changed by native JNI code.");
1.542 +
1.543 + __ fldcw(fpu_std);
1.544 +
1.545 + __ bind(ok_ret);
1.546 + __ addptr(rsp, wordSize);
1.547 + __ pop(rax);
1.548 + }
1.549 +
1.550 + __ ret(0);
1.551 +
1.552 + return start;
1.553 + }
1.554 +
1.555 + //---------------------------------------------------------------------------
1.556 + // Wrapper for slow-case handling of double-to-integer conversion
1.557 + // d2i or f2i fast case failed either because it is nan or because
1.558 + // of under/overflow.
1.559 + // Input: FPU TOS: float value
1.560 + // Output: rax, (rdx): integer (long) result
1.561 +
1.562 + address generate_d2i_wrapper(BasicType t, address fcn) {
1.563 + StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
1.564 + address start = __ pc();
1.565 +
1.566 + // Capture info about frame layout
1.567 + enum layout { FPUState_off = 0,
1.568 + rbp_off = FPUStateSizeInWords,
1.569 + rdi_off,
1.570 + rsi_off,
1.571 + rcx_off,
1.572 + rbx_off,
1.573 + saved_argument_off,
1.574 + saved_argument_off2, // 2nd half of double
1.575 + framesize
1.576 + };
1.577 +
1.578 + assert(FPUStateSizeInWords == 27, "update stack layout");
1.579 +
1.580 + // Save outgoing argument to stack across push_FPU_state()
1.581 + __ subptr(rsp, wordSize * 2);
1.582 + __ fstp_d(Address(rsp, 0));
1.583 +
1.584 + // Save CPU & FPU state
1.585 + __ push(rbx);
1.586 + __ push(rcx);
1.587 + __ push(rsi);
1.588 + __ push(rdi);
1.589 + __ push(rbp);
1.590 + __ push_FPU_state();
1.591 +
1.592 + // push_FPU_state() resets the FP top of stack
1.593 + // Load original double into FP top of stack
1.594 + __ fld_d(Address(rsp, saved_argument_off * wordSize));
1.595 + // Store double into stack as outgoing argument
1.596 + __ subptr(rsp, wordSize*2);
1.597 + __ fst_d(Address(rsp, 0));
1.598 +
1.599 + // Prepare FPU for doing math in C-land
1.600 + __ empty_FPU_stack();
1.601 + // Call the C code to massage the double. Result in EAX
1.602 + if (t == T_INT)
1.603 + { BLOCK_COMMENT("SharedRuntime::d2i"); }
1.604 + else if (t == T_LONG)
1.605 + { BLOCK_COMMENT("SharedRuntime::d2l"); }
1.606 + __ call_VM_leaf( fcn, 2 );
1.607 +
1.608 + // Restore CPU & FPU state
1.609 + __ pop_FPU_state();
1.610 + __ pop(rbp);
1.611 + __ pop(rdi);
1.612 + __ pop(rsi);
1.613 + __ pop(rcx);
1.614 + __ pop(rbx);
1.615 + __ addptr(rsp, wordSize * 2);
1.616 +
1.617 + __ ret(0);
1.618 +
1.619 + return start;
1.620 + }
1.621 +
1.622 +
1.623 + //---------------------------------------------------------------------------
1.624 + // The following routine generates a subroutine to throw an asynchronous
1.625 + // UnknownError when an unsafe access gets a fault that could not be
1.626 + // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
1.627 + address generate_handler_for_unsafe_access() {
1.628 + StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
1.629 + address start = __ pc();
1.630 +
1.631 + __ push(0); // hole for return address-to-be
1.632 + __ pusha(); // push registers
1.633 + Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
1.634 + BLOCK_COMMENT("call handle_unsafe_access");
1.635 + __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
1.636 + __ movptr(next_pc, rax); // stuff next address
1.637 + __ popa();
1.638 + __ ret(0); // jump to next address
1.639 +
1.640 + return start;
1.641 + }
1.642 +
1.643 +
1.644 + //----------------------------------------------------------------------------------------------------
1.645 + // Non-destructive plausibility checks for oops
1.646 +
1.647 + address generate_verify_oop() {
1.648 + StubCodeMark mark(this, "StubRoutines", "verify_oop");
1.649 + address start = __ pc();
1.650 +
1.651 + // Incoming arguments on stack after saving rax,:
1.652 + //
1.653 + // [tos ]: saved rdx
1.654 + // [tos + 1]: saved EFLAGS
1.655 + // [tos + 2]: return address
1.656 + // [tos + 3]: char* error message
1.657 + // [tos + 4]: oop object to verify
1.658 + // [tos + 5]: saved rax, - saved by caller and bashed
1.659 +
1.660 + Label exit, error;
1.661 + __ pushf();
1.662 + __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1.663 + __ push(rdx); // save rdx
1.664 + // make sure object is 'reasonable'
1.665 + __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
1.666 + __ testptr(rax, rax);
1.667 + __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
1.668 +
1.669 + // Check if the oop is in the right area of memory
1.670 + const int oop_mask = Universe::verify_oop_mask();
1.671 + const int oop_bits = Universe::verify_oop_bits();
1.672 + __ mov(rdx, rax);
1.673 + __ andptr(rdx, oop_mask);
1.674 + __ cmpptr(rdx, oop_bits);
1.675 + __ jcc(Assembler::notZero, error);
1.676 +
1.677 + // make sure klass is 'reasonable', which is not zero.
1.678 + __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
1.679 + __ testptr(rax, rax);
1.680 + __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1.681 +
1.682 + // return if everything seems ok
1.683 + __ bind(exit);
1.684 + __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
1.685 + __ pop(rdx); // restore rdx
1.686 + __ popf(); // restore EFLAGS
1.687 + __ ret(3 * wordSize); // pop arguments
1.688 +
1.689 + // handle errors
1.690 + __ bind(error);
1.691 + __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
1.692 + __ pop(rdx); // get saved rdx back
1.693 + __ popf(); // get saved EFLAGS off stack -- will be ignored
1.694 + __ pusha(); // push registers (eip = return address & msg are already pushed)
1.695 + BLOCK_COMMENT("call MacroAssembler::debug");
1.696 + __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
1.697 + __ popa();
1.698 + __ ret(3 * wordSize); // pop arguments
1.699 + return start;
1.700 + }
1.701 +
1.702 + //
1.703 + // Generate pre-barrier for array stores
1.704 + //
1.705 + // Input:
1.706 + // start - starting address
1.707 + // count - element count
1.708 + void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
1.709 + assert_different_registers(start, count);
1.710 + BarrierSet* bs = Universe::heap()->barrier_set();
1.711 + switch (bs->kind()) {
1.712 + case BarrierSet::G1SATBCT:
1.713 + case BarrierSet::G1SATBCTLogging:
1.714 + // With G1, don't generate the call if we statically know that the target in uninitialized
1.715 + if (!uninitialized_target) {
1.716 + __ pusha(); // push registers
1.717 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
1.718 + start, count);
1.719 + __ popa();
1.720 + }
1.721 + break;
1.722 + case BarrierSet::CardTableModRef:
1.723 + case BarrierSet::CardTableExtension:
1.724 + case BarrierSet::ModRef:
1.725 + break;
1.726 + default :
1.727 + ShouldNotReachHere();
1.728 +
1.729 + }
1.730 + }
1.731 +
1.732 +
1.733 + //
1.734 + // Generate a post-barrier for an array store
1.735 + //
1.736 + // start - starting address
1.737 + // count - element count
1.738 + //
1.739 + // The two input registers are overwritten.
1.740 + //
1.741 + void gen_write_ref_array_post_barrier(Register start, Register count) {
1.742 + BarrierSet* bs = Universe::heap()->barrier_set();
1.743 + assert_different_registers(start, count);
1.744 + switch (bs->kind()) {
1.745 + case BarrierSet::G1SATBCT:
1.746 + case BarrierSet::G1SATBCTLogging:
1.747 + {
1.748 + __ pusha(); // push registers
1.749 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
1.750 + start, count);
1.751 + __ popa();
1.752 + }
1.753 + break;
1.754 +
1.755 + case BarrierSet::CardTableModRef:
1.756 + case BarrierSet::CardTableExtension:
1.757 + {
1.758 + CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1.759 + assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1.760 +
1.761 + Label L_loop;
1.762 + const Register end = count; // elements count; end == start+count-1
1.763 + assert_different_registers(start, end);
1.764 +
1.765 + __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
1.766 + __ shrptr(start, CardTableModRefBS::card_shift);
1.767 + __ shrptr(end, CardTableModRefBS::card_shift);
1.768 + __ subptr(end, start); // end --> count
1.769 + __ BIND(L_loop);
1.770 + intptr_t disp = (intptr_t) ct->byte_map_base;
1.771 + Address cardtable(start, count, Address::times_1, disp);
1.772 + __ movb(cardtable, 0);
1.773 + __ decrement(count);
1.774 + __ jcc(Assembler::greaterEqual, L_loop);
1.775 + }
1.776 + break;
1.777 + case BarrierSet::ModRef:
1.778 + break;
1.779 + default :
1.780 + ShouldNotReachHere();
1.781 +
1.782 + }
1.783 + }
1.784 +
1.785 +
1.786 + // Copy 64 bytes chunks
1.787 + //
1.788 + // Inputs:
1.789 + // from - source array address
1.790 + // to_from - destination array address - from
1.791 + // qword_count - 8-bytes element count, negative
1.792 + //
1.793 + void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
1.794 + assert( UseSSE >= 2, "supported cpu only" );
1.795 + Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
1.796 + // Copy 64-byte chunks
1.797 + __ jmpb(L_copy_64_bytes);
1.798 + __ align(OptoLoopAlignment);
1.799 + __ BIND(L_copy_64_bytes_loop);
1.800 +
1.801 + if (UseUnalignedLoadStores) {
1.802 + if (UseAVX >= 2) {
1.803 + __ vmovdqu(xmm0, Address(from, 0));
1.804 + __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
1.805 + __ vmovdqu(xmm1, Address(from, 32));
1.806 + __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
1.807 + } else {
1.808 + __ movdqu(xmm0, Address(from, 0));
1.809 + __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
1.810 + __ movdqu(xmm1, Address(from, 16));
1.811 + __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
1.812 + __ movdqu(xmm2, Address(from, 32));
1.813 + __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
1.814 + __ movdqu(xmm3, Address(from, 48));
1.815 + __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
1.816 + }
1.817 + } else {
1.818 + __ movq(xmm0, Address(from, 0));
1.819 + __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
1.820 + __ movq(xmm1, Address(from, 8));
1.821 + __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
1.822 + __ movq(xmm2, Address(from, 16));
1.823 + __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
1.824 + __ movq(xmm3, Address(from, 24));
1.825 + __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
1.826 + __ movq(xmm4, Address(from, 32));
1.827 + __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
1.828 + __ movq(xmm5, Address(from, 40));
1.829 + __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
1.830 + __ movq(xmm6, Address(from, 48));
1.831 + __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
1.832 + __ movq(xmm7, Address(from, 56));
1.833 + __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
1.834 + }
1.835 +
1.836 + __ addl(from, 64);
1.837 + __ BIND(L_copy_64_bytes);
1.838 + __ subl(qword_count, 8);
1.839 + __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
1.840 +
1.841 + if (UseUnalignedLoadStores && (UseAVX >= 2)) {
1.842 + // clean upper bits of YMM registers
1.843 + __ vzeroupper();
1.844 + }
1.845 + __ addl(qword_count, 8);
1.846 + __ jccb(Assembler::zero, L_exit);
1.847 + //
1.848 + // length is too short, just copy qwords
1.849 + //
1.850 + __ BIND(L_copy_8_bytes);
1.851 + __ movq(xmm0, Address(from, 0));
1.852 + __ movq(Address(from, to_from, Address::times_1), xmm0);
1.853 + __ addl(from, 8);
1.854 + __ decrement(qword_count);
1.855 + __ jcc(Assembler::greater, L_copy_8_bytes);
1.856 + __ BIND(L_exit);
1.857 + }
1.858 +
1.859 + // Copy 64 bytes chunks
1.860 + //
1.861 + // Inputs:
1.862 + // from - source array address
1.863 + // to_from - destination array address - from
1.864 + // qword_count - 8-bytes element count, negative
1.865 + //
1.866 + void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
1.867 + assert( VM_Version::supports_mmx(), "supported cpu only" );
1.868 + Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
1.869 + // Copy 64-byte chunks
1.870 + __ jmpb(L_copy_64_bytes);
1.871 + __ align(OptoLoopAlignment);
1.872 + __ BIND(L_copy_64_bytes_loop);
1.873 + __ movq(mmx0, Address(from, 0));
1.874 + __ movq(mmx1, Address(from, 8));
1.875 + __ movq(mmx2, Address(from, 16));
1.876 + __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
1.877 + __ movq(mmx3, Address(from, 24));
1.878 + __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
1.879 + __ movq(mmx4, Address(from, 32));
1.880 + __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
1.881 + __ movq(mmx5, Address(from, 40));
1.882 + __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
1.883 + __ movq(mmx6, Address(from, 48));
1.884 + __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
1.885 + __ movq(mmx7, Address(from, 56));
1.886 + __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
1.887 + __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
1.888 + __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
1.889 + __ addptr(from, 64);
1.890 + __ BIND(L_copy_64_bytes);
1.891 + __ subl(qword_count, 8);
1.892 + __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
1.893 + __ addl(qword_count, 8);
1.894 + __ jccb(Assembler::zero, L_exit);
1.895 + //
1.896 + // length is too short, just copy qwords
1.897 + //
1.898 + __ BIND(L_copy_8_bytes);
1.899 + __ movq(mmx0, Address(from, 0));
1.900 + __ movq(Address(from, to_from, Address::times_1), mmx0);
1.901 + __ addptr(from, 8);
1.902 + __ decrement(qword_count);
1.903 + __ jcc(Assembler::greater, L_copy_8_bytes);
1.904 + __ BIND(L_exit);
1.905 + __ emms();
1.906 + }
1.907 +
1.908 + address generate_disjoint_copy(BasicType t, bool aligned,
1.909 + Address::ScaleFactor sf,
1.910 + address* entry, const char *name,
1.911 + bool dest_uninitialized = false) {
1.912 + __ align(CodeEntryAlignment);
1.913 + StubCodeMark mark(this, "StubRoutines", name);
1.914 + address start = __ pc();
1.915 +
1.916 + Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1.917 + Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
1.918 +
1.919 + int shift = Address::times_ptr - sf;
1.920 +
1.921 + const Register from = rsi; // source array address
1.922 + const Register to = rdi; // destination array address
1.923 + const Register count = rcx; // elements count
1.924 + const Register to_from = to; // (to - from)
1.925 + const Register saved_to = rdx; // saved destination array address
1.926 +
1.927 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.928 + __ push(rsi);
1.929 + __ push(rdi);
1.930 + __ movptr(from , Address(rsp, 12+ 4));
1.931 + __ movptr(to , Address(rsp, 12+ 8));
1.932 + __ movl(count, Address(rsp, 12+ 12));
1.933 +
1.934 + if (entry != NULL) {
1.935 + *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1.936 + BLOCK_COMMENT("Entry:");
1.937 + }
1.938 +
1.939 + if (t == T_OBJECT) {
1.940 + __ testl(count, count);
1.941 + __ jcc(Assembler::zero, L_0_count);
1.942 + gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1.943 + __ mov(saved_to, to); // save 'to'
1.944 + }
1.945 +
1.946 + __ subptr(to, from); // to --> to_from
1.947 + __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1.948 + __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1.949 + if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
1.950 + // align source address at 4 bytes address boundary
1.951 + if (t == T_BYTE) {
1.952 + // One byte misalignment happens only for byte arrays
1.953 + __ testl(from, 1);
1.954 + __ jccb(Assembler::zero, L_skip_align1);
1.955 + __ movb(rax, Address(from, 0));
1.956 + __ movb(Address(from, to_from, Address::times_1, 0), rax);
1.957 + __ increment(from);
1.958 + __ decrement(count);
1.959 + __ BIND(L_skip_align1);
1.960 + }
1.961 + // Two bytes misalignment happens only for byte and short (char) arrays
1.962 + __ testl(from, 2);
1.963 + __ jccb(Assembler::zero, L_skip_align2);
1.964 + __ movw(rax, Address(from, 0));
1.965 + __ movw(Address(from, to_from, Address::times_1, 0), rax);
1.966 + __ addptr(from, 2);
1.967 + __ subl(count, 1<<(shift-1));
1.968 + __ BIND(L_skip_align2);
1.969 + }
1.970 + if (!VM_Version::supports_mmx()) {
1.971 + __ mov(rax, count); // save 'count'
1.972 + __ shrl(count, shift); // bytes count
1.973 + __ addptr(to_from, from);// restore 'to'
1.974 + __ rep_mov();
1.975 + __ subptr(to_from, from);// restore 'to_from'
1.976 + __ mov(count, rax); // restore 'count'
1.977 + __ jmpb(L_copy_2_bytes); // all dwords were copied
1.978 + } else {
1.979 + if (!UseUnalignedLoadStores) {
1.980 + // align to 8 bytes, we know we are 4 byte aligned to start
1.981 + __ testptr(from, 4);
1.982 + __ jccb(Assembler::zero, L_copy_64_bytes);
1.983 + __ movl(rax, Address(from, 0));
1.984 + __ movl(Address(from, to_from, Address::times_1, 0), rax);
1.985 + __ addptr(from, 4);
1.986 + __ subl(count, 1<<shift);
1.987 + }
1.988 + __ BIND(L_copy_64_bytes);
1.989 + __ mov(rax, count);
1.990 + __ shrl(rax, shift+1); // 8 bytes chunk count
1.991 + //
1.992 + // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
1.993 + //
1.994 + if (UseXMMForArrayCopy) {
1.995 + xmm_copy_forward(from, to_from, rax);
1.996 + } else {
1.997 + mmx_copy_forward(from, to_from, rax);
1.998 + }
1.999 + }
1.1000 + // copy tailing dword
1.1001 + __ BIND(L_copy_4_bytes);
1.1002 + __ testl(count, 1<<shift);
1.1003 + __ jccb(Assembler::zero, L_copy_2_bytes);
1.1004 + __ movl(rax, Address(from, 0));
1.1005 + __ movl(Address(from, to_from, Address::times_1, 0), rax);
1.1006 + if (t == T_BYTE || t == T_SHORT) {
1.1007 + __ addptr(from, 4);
1.1008 + __ BIND(L_copy_2_bytes);
1.1009 + // copy tailing word
1.1010 + __ testl(count, 1<<(shift-1));
1.1011 + __ jccb(Assembler::zero, L_copy_byte);
1.1012 + __ movw(rax, Address(from, 0));
1.1013 + __ movw(Address(from, to_from, Address::times_1, 0), rax);
1.1014 + if (t == T_BYTE) {
1.1015 + __ addptr(from, 2);
1.1016 + __ BIND(L_copy_byte);
1.1017 + // copy tailing byte
1.1018 + __ testl(count, 1);
1.1019 + __ jccb(Assembler::zero, L_exit);
1.1020 + __ movb(rax, Address(from, 0));
1.1021 + __ movb(Address(from, to_from, Address::times_1, 0), rax);
1.1022 + __ BIND(L_exit);
1.1023 + } else {
1.1024 + __ BIND(L_copy_byte);
1.1025 + }
1.1026 + } else {
1.1027 + __ BIND(L_copy_2_bytes);
1.1028 + }
1.1029 +
1.1030 + if (t == T_OBJECT) {
1.1031 + __ movl(count, Address(rsp, 12+12)); // reread 'count'
1.1032 + __ mov(to, saved_to); // restore 'to'
1.1033 + gen_write_ref_array_post_barrier(to, count);
1.1034 + __ BIND(L_0_count);
1.1035 + }
1.1036 + inc_copy_counter_np(t);
1.1037 + __ pop(rdi);
1.1038 + __ pop(rsi);
1.1039 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1040 + __ xorptr(rax, rax); // return 0
1.1041 + __ ret(0);
1.1042 + return start;
1.1043 + }
1.1044 +
1.1045 +
1.1046 + address generate_fill(BasicType t, bool aligned, const char *name) {
1.1047 + __ align(CodeEntryAlignment);
1.1048 + StubCodeMark mark(this, "StubRoutines", name);
1.1049 + address start = __ pc();
1.1050 +
1.1051 + BLOCK_COMMENT("Entry:");
1.1052 +
1.1053 + const Register to = rdi; // source array address
1.1054 + const Register value = rdx; // value
1.1055 + const Register count = rsi; // elements count
1.1056 +
1.1057 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1058 + __ push(rsi);
1.1059 + __ push(rdi);
1.1060 + __ movptr(to , Address(rsp, 12+ 4));
1.1061 + __ movl(value, Address(rsp, 12+ 8));
1.1062 + __ movl(count, Address(rsp, 12+ 12));
1.1063 +
1.1064 + __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1.1065 +
1.1066 + __ pop(rdi);
1.1067 + __ pop(rsi);
1.1068 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1069 + __ ret(0);
1.1070 + return start;
1.1071 + }
1.1072 +
1.1073 + address generate_conjoint_copy(BasicType t, bool aligned,
1.1074 + Address::ScaleFactor sf,
1.1075 + address nooverlap_target,
1.1076 + address* entry, const char *name,
1.1077 + bool dest_uninitialized = false) {
1.1078 + __ align(CodeEntryAlignment);
1.1079 + StubCodeMark mark(this, "StubRoutines", name);
1.1080 + address start = __ pc();
1.1081 +
1.1082 + Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1.1083 + Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1.1084 +
1.1085 + int shift = Address::times_ptr - sf;
1.1086 +
1.1087 + const Register src = rax; // source array address
1.1088 + const Register dst = rdx; // destination array address
1.1089 + const Register from = rsi; // source array address
1.1090 + const Register to = rdi; // destination array address
1.1091 + const Register count = rcx; // elements count
1.1092 + const Register end = rax; // array end address
1.1093 +
1.1094 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1095 + __ push(rsi);
1.1096 + __ push(rdi);
1.1097 + __ movptr(src , Address(rsp, 12+ 4)); // from
1.1098 + __ movptr(dst , Address(rsp, 12+ 8)); // to
1.1099 + __ movl2ptr(count, Address(rsp, 12+12)); // count
1.1100 +
1.1101 + if (entry != NULL) {
1.1102 + *entry = __ pc(); // Entry point from generic arraycopy stub.
1.1103 + BLOCK_COMMENT("Entry:");
1.1104 + }
1.1105 +
1.1106 + // nooverlap_target expects arguments in rsi and rdi.
1.1107 + __ mov(from, src);
1.1108 + __ mov(to , dst);
1.1109 +
1.1110 + // arrays overlap test: dispatch to disjoint stub if necessary.
1.1111 + RuntimeAddress nooverlap(nooverlap_target);
1.1112 + __ cmpptr(dst, src);
1.1113 + __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1.1114 + __ jump_cc(Assembler::belowEqual, nooverlap);
1.1115 + __ cmpptr(dst, end);
1.1116 + __ jump_cc(Assembler::aboveEqual, nooverlap);
1.1117 +
1.1118 + if (t == T_OBJECT) {
1.1119 + __ testl(count, count);
1.1120 + __ jcc(Assembler::zero, L_0_count);
1.1121 + gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1.1122 + }
1.1123 +
1.1124 + // copy from high to low
1.1125 + __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1.1126 + __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1.1127 + if (t == T_BYTE || t == T_SHORT) {
1.1128 + // Align the end of destination array at 4 bytes address boundary
1.1129 + __ lea(end, Address(dst, count, sf, 0));
1.1130 + if (t == T_BYTE) {
1.1131 + // One byte misalignment happens only for byte arrays
1.1132 + __ testl(end, 1);
1.1133 + __ jccb(Assembler::zero, L_skip_align1);
1.1134 + __ decrement(count);
1.1135 + __ movb(rdx, Address(from, count, sf, 0));
1.1136 + __ movb(Address(to, count, sf, 0), rdx);
1.1137 + __ BIND(L_skip_align1);
1.1138 + }
1.1139 + // Two bytes misalignment happens only for byte and short (char) arrays
1.1140 + __ testl(end, 2);
1.1141 + __ jccb(Assembler::zero, L_skip_align2);
1.1142 + __ subptr(count, 1<<(shift-1));
1.1143 + __ movw(rdx, Address(from, count, sf, 0));
1.1144 + __ movw(Address(to, count, sf, 0), rdx);
1.1145 + __ BIND(L_skip_align2);
1.1146 + __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1.1147 + __ jcc(Assembler::below, L_copy_4_bytes);
1.1148 + }
1.1149 +
1.1150 + if (!VM_Version::supports_mmx()) {
1.1151 + __ std();
1.1152 + __ mov(rax, count); // Save 'count'
1.1153 + __ mov(rdx, to); // Save 'to'
1.1154 + __ lea(rsi, Address(from, count, sf, -4));
1.1155 + __ lea(rdi, Address(to , count, sf, -4));
1.1156 + __ shrptr(count, shift); // bytes count
1.1157 + __ rep_mov();
1.1158 + __ cld();
1.1159 + __ mov(count, rax); // restore 'count'
1.1160 + __ andl(count, (1<<shift)-1); // mask the number of rest elements
1.1161 + __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1.1162 + __ mov(to, rdx); // restore 'to'
1.1163 + __ jmpb(L_copy_2_bytes); // all dword were copied
1.1164 + } else {
1.1165 + // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1.1166 + __ testptr(end, 4);
1.1167 + __ jccb(Assembler::zero, L_copy_8_bytes);
1.1168 + __ subl(count, 1<<shift);
1.1169 + __ movl(rdx, Address(from, count, sf, 0));
1.1170 + __ movl(Address(to, count, sf, 0), rdx);
1.1171 + __ jmpb(L_copy_8_bytes);
1.1172 +
1.1173 + __ align(OptoLoopAlignment);
1.1174 + // Move 8 bytes
1.1175 + __ BIND(L_copy_8_bytes_loop);
1.1176 + if (UseXMMForArrayCopy) {
1.1177 + __ movq(xmm0, Address(from, count, sf, 0));
1.1178 + __ movq(Address(to, count, sf, 0), xmm0);
1.1179 + } else {
1.1180 + __ movq(mmx0, Address(from, count, sf, 0));
1.1181 + __ movq(Address(to, count, sf, 0), mmx0);
1.1182 + }
1.1183 + __ BIND(L_copy_8_bytes);
1.1184 + __ subl(count, 2<<shift);
1.1185 + __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1.1186 + __ addl(count, 2<<shift);
1.1187 + if (!UseXMMForArrayCopy) {
1.1188 + __ emms();
1.1189 + }
1.1190 + }
1.1191 + __ BIND(L_copy_4_bytes);
1.1192 + // copy prefix qword
1.1193 + __ testl(count, 1<<shift);
1.1194 + __ jccb(Assembler::zero, L_copy_2_bytes);
1.1195 + __ movl(rdx, Address(from, count, sf, -4));
1.1196 + __ movl(Address(to, count, sf, -4), rdx);
1.1197 +
1.1198 + if (t == T_BYTE || t == T_SHORT) {
1.1199 + __ subl(count, (1<<shift));
1.1200 + __ BIND(L_copy_2_bytes);
1.1201 + // copy prefix dword
1.1202 + __ testl(count, 1<<(shift-1));
1.1203 + __ jccb(Assembler::zero, L_copy_byte);
1.1204 + __ movw(rdx, Address(from, count, sf, -2));
1.1205 + __ movw(Address(to, count, sf, -2), rdx);
1.1206 + if (t == T_BYTE) {
1.1207 + __ subl(count, 1<<(shift-1));
1.1208 + __ BIND(L_copy_byte);
1.1209 + // copy prefix byte
1.1210 + __ testl(count, 1);
1.1211 + __ jccb(Assembler::zero, L_exit);
1.1212 + __ movb(rdx, Address(from, 0));
1.1213 + __ movb(Address(to, 0), rdx);
1.1214 + __ BIND(L_exit);
1.1215 + } else {
1.1216 + __ BIND(L_copy_byte);
1.1217 + }
1.1218 + } else {
1.1219 + __ BIND(L_copy_2_bytes);
1.1220 + }
1.1221 + if (t == T_OBJECT) {
1.1222 + __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1.1223 + gen_write_ref_array_post_barrier(to, count);
1.1224 + __ BIND(L_0_count);
1.1225 + }
1.1226 + inc_copy_counter_np(t);
1.1227 + __ pop(rdi);
1.1228 + __ pop(rsi);
1.1229 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1230 + __ xorptr(rax, rax); // return 0
1.1231 + __ ret(0);
1.1232 + return start;
1.1233 + }
1.1234 +
1.1235 +
1.1236 + address generate_disjoint_long_copy(address* entry, const char *name) {
1.1237 + __ align(CodeEntryAlignment);
1.1238 + StubCodeMark mark(this, "StubRoutines", name);
1.1239 + address start = __ pc();
1.1240 +
1.1241 + Label L_copy_8_bytes, L_copy_8_bytes_loop;
1.1242 + const Register from = rax; // source array address
1.1243 + const Register to = rdx; // destination array address
1.1244 + const Register count = rcx; // elements count
1.1245 + const Register to_from = rdx; // (to - from)
1.1246 +
1.1247 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1248 + __ movptr(from , Address(rsp, 8+0)); // from
1.1249 + __ movptr(to , Address(rsp, 8+4)); // to
1.1250 + __ movl2ptr(count, Address(rsp, 8+8)); // count
1.1251 +
1.1252 + *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1.1253 + BLOCK_COMMENT("Entry:");
1.1254 +
1.1255 + __ subptr(to, from); // to --> to_from
1.1256 + if (VM_Version::supports_mmx()) {
1.1257 + if (UseXMMForArrayCopy) {
1.1258 + xmm_copy_forward(from, to_from, count);
1.1259 + } else {
1.1260 + mmx_copy_forward(from, to_from, count);
1.1261 + }
1.1262 + } else {
1.1263 + __ jmpb(L_copy_8_bytes);
1.1264 + __ align(OptoLoopAlignment);
1.1265 + __ BIND(L_copy_8_bytes_loop);
1.1266 + __ fild_d(Address(from, 0));
1.1267 + __ fistp_d(Address(from, to_from, Address::times_1));
1.1268 + __ addptr(from, 8);
1.1269 + __ BIND(L_copy_8_bytes);
1.1270 + __ decrement(count);
1.1271 + __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1.1272 + }
1.1273 + inc_copy_counter_np(T_LONG);
1.1274 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1275 + __ xorptr(rax, rax); // return 0
1.1276 + __ ret(0);
1.1277 + return start;
1.1278 + }
1.1279 +
1.1280 + address generate_conjoint_long_copy(address nooverlap_target,
1.1281 + address* entry, const char *name) {
1.1282 + __ align(CodeEntryAlignment);
1.1283 + StubCodeMark mark(this, "StubRoutines", name);
1.1284 + address start = __ pc();
1.1285 +
1.1286 + Label L_copy_8_bytes, L_copy_8_bytes_loop;
1.1287 + const Register from = rax; // source array address
1.1288 + const Register to = rdx; // destination array address
1.1289 + const Register count = rcx; // elements count
1.1290 + const Register end_from = rax; // source array end address
1.1291 +
1.1292 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1293 + __ movptr(from , Address(rsp, 8+0)); // from
1.1294 + __ movptr(to , Address(rsp, 8+4)); // to
1.1295 + __ movl2ptr(count, Address(rsp, 8+8)); // count
1.1296 +
1.1297 + *entry = __ pc(); // Entry point from generic arraycopy stub.
1.1298 + BLOCK_COMMENT("Entry:");
1.1299 +
1.1300 + // arrays overlap test
1.1301 + __ cmpptr(to, from);
1.1302 + RuntimeAddress nooverlap(nooverlap_target);
1.1303 + __ jump_cc(Assembler::belowEqual, nooverlap);
1.1304 + __ lea(end_from, Address(from, count, Address::times_8, 0));
1.1305 + __ cmpptr(to, end_from);
1.1306 + __ movptr(from, Address(rsp, 8)); // from
1.1307 + __ jump_cc(Assembler::aboveEqual, nooverlap);
1.1308 +
1.1309 + __ jmpb(L_copy_8_bytes);
1.1310 +
1.1311 + __ align(OptoLoopAlignment);
1.1312 + __ BIND(L_copy_8_bytes_loop);
1.1313 + if (VM_Version::supports_mmx()) {
1.1314 + if (UseXMMForArrayCopy) {
1.1315 + __ movq(xmm0, Address(from, count, Address::times_8));
1.1316 + __ movq(Address(to, count, Address::times_8), xmm0);
1.1317 + } else {
1.1318 + __ movq(mmx0, Address(from, count, Address::times_8));
1.1319 + __ movq(Address(to, count, Address::times_8), mmx0);
1.1320 + }
1.1321 + } else {
1.1322 + __ fild_d(Address(from, count, Address::times_8));
1.1323 + __ fistp_d(Address(to, count, Address::times_8));
1.1324 + }
1.1325 + __ BIND(L_copy_8_bytes);
1.1326 + __ decrement(count);
1.1327 + __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1.1328 +
1.1329 + if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1.1330 + __ emms();
1.1331 + }
1.1332 + inc_copy_counter_np(T_LONG);
1.1333 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1334 + __ xorptr(rax, rax); // return 0
1.1335 + __ ret(0);
1.1336 + return start;
1.1337 + }
1.1338 +
1.1339 +
1.1340 + // Helper for generating a dynamic type check.
1.1341 + // The sub_klass must be one of {rbx, rdx, rsi}.
1.1342 + // The temp is killed.
1.1343 + void generate_type_check(Register sub_klass,
1.1344 + Address& super_check_offset_addr,
1.1345 + Address& super_klass_addr,
1.1346 + Register temp,
1.1347 + Label* L_success, Label* L_failure) {
1.1348 + BLOCK_COMMENT("type_check:");
1.1349 +
1.1350 + Label L_fallthrough;
1.1351 +#define LOCAL_JCC(assembler_con, label_ptr) \
1.1352 + if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1.1353 + else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1.1354 +
1.1355 + // The following is a strange variation of the fast path which requires
1.1356 + // one less register, because needed values are on the argument stack.
1.1357 + // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1.1358 + // L_success, L_failure, NULL);
1.1359 + assert_different_registers(sub_klass, temp);
1.1360 +
1.1361 + int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1.1362 +
1.1363 + // if the pointers are equal, we are done (e.g., String[] elements)
1.1364 + __ cmpptr(sub_klass, super_klass_addr);
1.1365 + LOCAL_JCC(Assembler::equal, L_success);
1.1366 +
1.1367 + // check the supertype display:
1.1368 + __ movl2ptr(temp, super_check_offset_addr);
1.1369 + Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1.1370 + __ movptr(temp, super_check_addr); // load displayed supertype
1.1371 + __ cmpptr(temp, super_klass_addr); // test the super type
1.1372 + LOCAL_JCC(Assembler::equal, L_success);
1.1373 +
1.1374 + // if it was a primary super, we can just fail immediately
1.1375 + __ cmpl(super_check_offset_addr, sc_offset);
1.1376 + LOCAL_JCC(Assembler::notEqual, L_failure);
1.1377 +
1.1378 + // The repne_scan instruction uses fixed registers, which will get spilled.
1.1379 + // We happen to know this works best when super_klass is in rax.
1.1380 + Register super_klass = temp;
1.1381 + __ movptr(super_klass, super_klass_addr);
1.1382 + __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1.1383 + L_success, L_failure);
1.1384 +
1.1385 + __ bind(L_fallthrough);
1.1386 +
1.1387 + if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1.1388 + if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1.1389 +
1.1390 +#undef LOCAL_JCC
1.1391 + }
1.1392 +
1.1393 + //
1.1394 + // Generate checkcasting array copy stub
1.1395 + //
1.1396 + // Input:
1.1397 + // 4(rsp) - source array address
1.1398 + // 8(rsp) - destination array address
1.1399 + // 12(rsp) - element count, can be zero
1.1400 + // 16(rsp) - size_t ckoff (super_check_offset)
1.1401 + // 20(rsp) - oop ckval (super_klass)
1.1402 + //
1.1403 + // Output:
1.1404 + // rax, == 0 - success
1.1405 + // rax, == -1^K - failure, where K is partial transfer count
1.1406 + //
1.1407 + address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1.1408 + __ align(CodeEntryAlignment);
1.1409 + StubCodeMark mark(this, "StubRoutines", name);
1.1410 + address start = __ pc();
1.1411 +
1.1412 + Label L_load_element, L_store_element, L_do_card_marks, L_done;
1.1413 +
1.1414 + // register use:
1.1415 + // rax, rdx, rcx -- loop control (end_from, end_to, count)
1.1416 + // rdi, rsi -- element access (oop, klass)
1.1417 + // rbx, -- temp
1.1418 + const Register from = rax; // source array address
1.1419 + const Register to = rdx; // destination array address
1.1420 + const Register length = rcx; // elements count
1.1421 + const Register elem = rdi; // each oop copied
1.1422 + const Register elem_klass = rsi; // each elem._klass (sub_klass)
1.1423 + const Register temp = rbx; // lone remaining temp
1.1424 +
1.1425 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1426 +
1.1427 + __ push(rsi);
1.1428 + __ push(rdi);
1.1429 + __ push(rbx);
1.1430 +
1.1431 + Address from_arg(rsp, 16+ 4); // from
1.1432 + Address to_arg(rsp, 16+ 8); // to
1.1433 + Address length_arg(rsp, 16+12); // elements count
1.1434 + Address ckoff_arg(rsp, 16+16); // super_check_offset
1.1435 + Address ckval_arg(rsp, 16+20); // super_klass
1.1436 +
1.1437 + // Load up:
1.1438 + __ movptr(from, from_arg);
1.1439 + __ movptr(to, to_arg);
1.1440 + __ movl2ptr(length, length_arg);
1.1441 +
1.1442 + if (entry != NULL) {
1.1443 + *entry = __ pc(); // Entry point from generic arraycopy stub.
1.1444 + BLOCK_COMMENT("Entry:");
1.1445 + }
1.1446 +
1.1447 + //---------------------------------------------------------------
1.1448 + // Assembler stub will be used for this call to arraycopy
1.1449 + // if the two arrays are subtypes of Object[] but the
1.1450 + // destination array type is not equal to or a supertype
1.1451 + // of the source type. Each element must be separately
1.1452 + // checked.
1.1453 +
1.1454 + // Loop-invariant addresses. They are exclusive end pointers.
1.1455 + Address end_from_addr(from, length, Address::times_ptr, 0);
1.1456 + Address end_to_addr(to, length, Address::times_ptr, 0);
1.1457 +
1.1458 + Register end_from = from; // re-use
1.1459 + Register end_to = to; // re-use
1.1460 + Register count = length; // re-use
1.1461 +
1.1462 + // Loop-variant addresses. They assume post-incremented count < 0.
1.1463 + Address from_element_addr(end_from, count, Address::times_ptr, 0);
1.1464 + Address to_element_addr(end_to, count, Address::times_ptr, 0);
1.1465 + Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1.1466 +
1.1467 + // Copy from low to high addresses, indexed from the end of each array.
1.1468 + gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1.1469 + __ lea(end_from, end_from_addr);
1.1470 + __ lea(end_to, end_to_addr);
1.1471 + assert(length == count, ""); // else fix next line:
1.1472 + __ negptr(count); // negate and test the length
1.1473 + __ jccb(Assembler::notZero, L_load_element);
1.1474 +
1.1475 + // Empty array: Nothing to do.
1.1476 + __ xorptr(rax, rax); // return 0 on (trivial) success
1.1477 + __ jmp(L_done);
1.1478 +
1.1479 + // ======== begin loop ========
1.1480 + // (Loop is rotated; its entry is L_load_element.)
1.1481 + // Loop control:
1.1482 + // for (count = -count; count != 0; count++)
1.1483 + // Base pointers src, dst are biased by 8*count,to last element.
1.1484 + __ align(OptoLoopAlignment);
1.1485 +
1.1486 + __ BIND(L_store_element);
1.1487 + __ movptr(to_element_addr, elem); // store the oop
1.1488 + __ increment(count); // increment the count toward zero
1.1489 + __ jccb(Assembler::zero, L_do_card_marks);
1.1490 +
1.1491 + // ======== loop entry is here ========
1.1492 + __ BIND(L_load_element);
1.1493 + __ movptr(elem, from_element_addr); // load the oop
1.1494 + __ testptr(elem, elem);
1.1495 + __ jccb(Assembler::zero, L_store_element);
1.1496 +
1.1497 + // (Could do a trick here: Remember last successful non-null
1.1498 + // element stored and make a quick oop equality check on it.)
1.1499 +
1.1500 + __ movptr(elem_klass, elem_klass_addr); // query the object klass
1.1501 + generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1.1502 + &L_store_element, NULL);
1.1503 + // (On fall-through, we have failed the element type check.)
1.1504 + // ======== end loop ========
1.1505 +
1.1506 + // It was a real error; we must depend on the caller to finish the job.
1.1507 + // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1.1508 + // Emit GC store barriers for the oops we have copied (length_arg + count),
1.1509 + // and report their number to the caller.
1.1510 + assert_different_registers(to, count, rax);
1.1511 + Label L_post_barrier;
1.1512 + __ addl(count, length_arg); // transfers = (length - remaining)
1.1513 + __ movl2ptr(rax, count); // save the value
1.1514 + __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1.1515 + __ jccb(Assembler::notZero, L_post_barrier);
1.1516 + __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1.1517 +
1.1518 + // Come here on success only.
1.1519 + __ BIND(L_do_card_marks);
1.1520 + __ xorptr(rax, rax); // return 0 on success
1.1521 + __ movl2ptr(count, length_arg);
1.1522 +
1.1523 + __ BIND(L_post_barrier);
1.1524 + __ movptr(to, to_arg); // reload
1.1525 + gen_write_ref_array_post_barrier(to, count);
1.1526 +
1.1527 + // Common exit point (success or failure).
1.1528 + __ BIND(L_done);
1.1529 + __ pop(rbx);
1.1530 + __ pop(rdi);
1.1531 + __ pop(rsi);
1.1532 + inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1.1533 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1534 + __ ret(0);
1.1535 +
1.1536 + return start;
1.1537 + }
1.1538 +
1.1539 + //
1.1540 + // Generate 'unsafe' array copy stub
1.1541 + // Though just as safe as the other stubs, it takes an unscaled
1.1542 + // size_t argument instead of an element count.
1.1543 + //
1.1544 + // Input:
1.1545 + // 4(rsp) - source array address
1.1546 + // 8(rsp) - destination array address
1.1547 + // 12(rsp) - byte count, can be zero
1.1548 + //
1.1549 + // Output:
1.1550 + // rax, == 0 - success
1.1551 + // rax, == -1 - need to call System.arraycopy
1.1552 + //
1.1553 + // Examines the alignment of the operands and dispatches
1.1554 + // to a long, int, short, or byte copy loop.
1.1555 + //
1.1556 + address generate_unsafe_copy(const char *name,
1.1557 + address byte_copy_entry,
1.1558 + address short_copy_entry,
1.1559 + address int_copy_entry,
1.1560 + address long_copy_entry) {
1.1561 +
1.1562 + Label L_long_aligned, L_int_aligned, L_short_aligned;
1.1563 +
1.1564 + __ align(CodeEntryAlignment);
1.1565 + StubCodeMark mark(this, "StubRoutines", name);
1.1566 + address start = __ pc();
1.1567 +
1.1568 + const Register from = rax; // source array address
1.1569 + const Register to = rdx; // destination array address
1.1570 + const Register count = rcx; // elements count
1.1571 +
1.1572 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1573 + __ push(rsi);
1.1574 + __ push(rdi);
1.1575 + Address from_arg(rsp, 12+ 4); // from
1.1576 + Address to_arg(rsp, 12+ 8); // to
1.1577 + Address count_arg(rsp, 12+12); // byte count
1.1578 +
1.1579 + // Load up:
1.1580 + __ movptr(from , from_arg);
1.1581 + __ movptr(to , to_arg);
1.1582 + __ movl2ptr(count, count_arg);
1.1583 +
1.1584 + // bump this on entry, not on exit:
1.1585 + inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1.1586 +
1.1587 + const Register bits = rsi;
1.1588 + __ mov(bits, from);
1.1589 + __ orptr(bits, to);
1.1590 + __ orptr(bits, count);
1.1591 +
1.1592 + __ testl(bits, BytesPerLong-1);
1.1593 + __ jccb(Assembler::zero, L_long_aligned);
1.1594 +
1.1595 + __ testl(bits, BytesPerInt-1);
1.1596 + __ jccb(Assembler::zero, L_int_aligned);
1.1597 +
1.1598 + __ testl(bits, BytesPerShort-1);
1.1599 + __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1.1600 +
1.1601 + __ BIND(L_short_aligned);
1.1602 + __ shrptr(count, LogBytesPerShort); // size => short_count
1.1603 + __ movl(count_arg, count); // update 'count'
1.1604 + __ jump(RuntimeAddress(short_copy_entry));
1.1605 +
1.1606 + __ BIND(L_int_aligned);
1.1607 + __ shrptr(count, LogBytesPerInt); // size => int_count
1.1608 + __ movl(count_arg, count); // update 'count'
1.1609 + __ jump(RuntimeAddress(int_copy_entry));
1.1610 +
1.1611 + __ BIND(L_long_aligned);
1.1612 + __ shrptr(count, LogBytesPerLong); // size => qword_count
1.1613 + __ movl(count_arg, count); // update 'count'
1.1614 + __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1.1615 + __ pop(rsi);
1.1616 + __ jump(RuntimeAddress(long_copy_entry));
1.1617 +
1.1618 + return start;
1.1619 + }
1.1620 +
1.1621 +
1.1622 + // Perform range checks on the proposed arraycopy.
1.1623 + // Smashes src_pos and dst_pos. (Uses them up for temps.)
1.1624 + void arraycopy_range_checks(Register src,
1.1625 + Register src_pos,
1.1626 + Register dst,
1.1627 + Register dst_pos,
1.1628 + Address& length,
1.1629 + Label& L_failed) {
1.1630 + BLOCK_COMMENT("arraycopy_range_checks:");
1.1631 + const Register src_end = src_pos; // source array end position
1.1632 + const Register dst_end = dst_pos; // destination array end position
1.1633 + __ addl(src_end, length); // src_pos + length
1.1634 + __ addl(dst_end, length); // dst_pos + length
1.1635 +
1.1636 + // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1.1637 + __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1.1638 + __ jcc(Assembler::above, L_failed);
1.1639 +
1.1640 + // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1.1641 + __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1.1642 + __ jcc(Assembler::above, L_failed);
1.1643 +
1.1644 + BLOCK_COMMENT("arraycopy_range_checks done");
1.1645 + }
1.1646 +
1.1647 +
1.1648 + //
1.1649 + // Generate generic array copy stubs
1.1650 + //
1.1651 + // Input:
1.1652 + // 4(rsp) - src oop
1.1653 + // 8(rsp) - src_pos
1.1654 + // 12(rsp) - dst oop
1.1655 + // 16(rsp) - dst_pos
1.1656 + // 20(rsp) - element count
1.1657 + //
1.1658 + // Output:
1.1659 + // rax, == 0 - success
1.1660 + // rax, == -1^K - failure, where K is partial transfer count
1.1661 + //
1.1662 + address generate_generic_copy(const char *name,
1.1663 + address entry_jbyte_arraycopy,
1.1664 + address entry_jshort_arraycopy,
1.1665 + address entry_jint_arraycopy,
1.1666 + address entry_oop_arraycopy,
1.1667 + address entry_jlong_arraycopy,
1.1668 + address entry_checkcast_arraycopy) {
1.1669 + Label L_failed, L_failed_0, L_objArray;
1.1670 +
1.1671 + { int modulus = CodeEntryAlignment;
1.1672 + int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1.1673 + int advance = target - (__ offset() % modulus);
1.1674 + if (advance < 0) advance += modulus;
1.1675 + if (advance > 0) __ nop(advance);
1.1676 + }
1.1677 + StubCodeMark mark(this, "StubRoutines", name);
1.1678 +
1.1679 + // Short-hop target to L_failed. Makes for denser prologue code.
1.1680 + __ BIND(L_failed_0);
1.1681 + __ jmp(L_failed);
1.1682 + assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1.1683 +
1.1684 + __ align(CodeEntryAlignment);
1.1685 + address start = __ pc();
1.1686 +
1.1687 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.1688 + __ push(rsi);
1.1689 + __ push(rdi);
1.1690 +
1.1691 + // bump this on entry, not on exit:
1.1692 + inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1.1693 +
1.1694 + // Input values
1.1695 + Address SRC (rsp, 12+ 4);
1.1696 + Address SRC_POS (rsp, 12+ 8);
1.1697 + Address DST (rsp, 12+12);
1.1698 + Address DST_POS (rsp, 12+16);
1.1699 + Address LENGTH (rsp, 12+20);
1.1700 +
1.1701 + //-----------------------------------------------------------------------
1.1702 + // Assembler stub will be used for this call to arraycopy
1.1703 + // if the following conditions are met:
1.1704 + //
1.1705 + // (1) src and dst must not be null.
1.1706 + // (2) src_pos must not be negative.
1.1707 + // (3) dst_pos must not be negative.
1.1708 + // (4) length must not be negative.
1.1709 + // (5) src klass and dst klass should be the same and not NULL.
1.1710 + // (6) src and dst should be arrays.
1.1711 + // (7) src_pos + length must not exceed length of src.
1.1712 + // (8) dst_pos + length must not exceed length of dst.
1.1713 + //
1.1714 +
1.1715 + const Register src = rax; // source array oop
1.1716 + const Register src_pos = rsi;
1.1717 + const Register dst = rdx; // destination array oop
1.1718 + const Register dst_pos = rdi;
1.1719 + const Register length = rcx; // transfer count
1.1720 +
1.1721 + // if (src == NULL) return -1;
1.1722 + __ movptr(src, SRC); // src oop
1.1723 + __ testptr(src, src);
1.1724 + __ jccb(Assembler::zero, L_failed_0);
1.1725 +
1.1726 + // if (src_pos < 0) return -1;
1.1727 + __ movl2ptr(src_pos, SRC_POS); // src_pos
1.1728 + __ testl(src_pos, src_pos);
1.1729 + __ jccb(Assembler::negative, L_failed_0);
1.1730 +
1.1731 + // if (dst == NULL) return -1;
1.1732 + __ movptr(dst, DST); // dst oop
1.1733 + __ testptr(dst, dst);
1.1734 + __ jccb(Assembler::zero, L_failed_0);
1.1735 +
1.1736 + // if (dst_pos < 0) return -1;
1.1737 + __ movl2ptr(dst_pos, DST_POS); // dst_pos
1.1738 + __ testl(dst_pos, dst_pos);
1.1739 + __ jccb(Assembler::negative, L_failed_0);
1.1740 +
1.1741 + // if (length < 0) return -1;
1.1742 + __ movl2ptr(length, LENGTH); // length
1.1743 + __ testl(length, length);
1.1744 + __ jccb(Assembler::negative, L_failed_0);
1.1745 +
1.1746 + // if (src->klass() == NULL) return -1;
1.1747 + Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1.1748 + Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1.1749 + const Register rcx_src_klass = rcx; // array klass
1.1750 + __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1.1751 +
1.1752 +#ifdef ASSERT
1.1753 + // assert(src->klass() != NULL);
1.1754 + BLOCK_COMMENT("assert klasses not null");
1.1755 + { Label L1, L2;
1.1756 + __ testptr(rcx_src_klass, rcx_src_klass);
1.1757 + __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1.1758 + __ bind(L1);
1.1759 + __ stop("broken null klass");
1.1760 + __ bind(L2);
1.1761 + __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1.1762 + __ jccb(Assembler::equal, L1); // this would be broken also
1.1763 + BLOCK_COMMENT("assert done");
1.1764 + }
1.1765 +#endif //ASSERT
1.1766 +
1.1767 + // Load layout helper (32-bits)
1.1768 + //
1.1769 + // |array_tag| | header_size | element_type | |log2_element_size|
1.1770 + // 32 30 24 16 8 2 0
1.1771 + //
1.1772 + // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1.1773 + //
1.1774 +
1.1775 + int lh_offset = in_bytes(Klass::layout_helper_offset());
1.1776 + Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1.1777 +
1.1778 + // Handle objArrays completely differently...
1.1779 + jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1.1780 + __ cmpl(src_klass_lh_addr, objArray_lh);
1.1781 + __ jcc(Assembler::equal, L_objArray);
1.1782 +
1.1783 + // if (src->klass() != dst->klass()) return -1;
1.1784 + __ cmpptr(rcx_src_klass, dst_klass_addr);
1.1785 + __ jccb(Assembler::notEqual, L_failed_0);
1.1786 +
1.1787 + const Register rcx_lh = rcx; // layout helper
1.1788 + assert(rcx_lh == rcx_src_klass, "known alias");
1.1789 + __ movl(rcx_lh, src_klass_lh_addr);
1.1790 +
1.1791 + // if (!src->is_Array()) return -1;
1.1792 + __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1.1793 + __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1.1794 +
1.1795 + // At this point, it is known to be a typeArray (array_tag 0x3).
1.1796 +#ifdef ASSERT
1.1797 + { Label L;
1.1798 + __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1.1799 + __ jcc(Assembler::greaterEqual, L); // signed cmp
1.1800 + __ stop("must be a primitive array");
1.1801 + __ bind(L);
1.1802 + }
1.1803 +#endif
1.1804 +
1.1805 + assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1.1806 + arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1.1807 +
1.1808 + // TypeArrayKlass
1.1809 + //
1.1810 + // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1.1811 + // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1.1812 + //
1.1813 + const Register rsi_offset = rsi; // array offset
1.1814 + const Register src_array = src; // src array offset
1.1815 + const Register dst_array = dst; // dst array offset
1.1816 + const Register rdi_elsize = rdi; // log2 element size
1.1817 +
1.1818 + __ mov(rsi_offset, rcx_lh);
1.1819 + __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1.1820 + __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1.1821 + __ addptr(src_array, rsi_offset); // src array offset
1.1822 + __ addptr(dst_array, rsi_offset); // dst array offset
1.1823 + __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1.1824 +
1.1825 + // next registers should be set before the jump to corresponding stub
1.1826 + const Register from = src; // source array address
1.1827 + const Register to = dst; // destination array address
1.1828 + const Register count = rcx; // elements count
1.1829 + // some of them should be duplicated on stack
1.1830 +#define FROM Address(rsp, 12+ 4)
1.1831 +#define TO Address(rsp, 12+ 8) // Not used now
1.1832 +#define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1.1833 +
1.1834 + BLOCK_COMMENT("scale indexes to element size");
1.1835 + __ movl2ptr(rsi, SRC_POS); // src_pos
1.1836 + __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1.1837 + assert(src_array == from, "");
1.1838 + __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1.1839 + __ movl2ptr(rdi, DST_POS); // dst_pos
1.1840 + __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1.1841 + assert(dst_array == to, "");
1.1842 + __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1.1843 + __ movptr(FROM, from); // src_addr
1.1844 + __ mov(rdi_elsize, rcx_lh); // log2 elsize
1.1845 + __ movl2ptr(count, LENGTH); // elements count
1.1846 +
1.1847 + BLOCK_COMMENT("choose copy loop based on element size");
1.1848 + __ cmpl(rdi_elsize, 0);
1.1849 +
1.1850 + __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1.1851 + __ cmpl(rdi_elsize, LogBytesPerShort);
1.1852 + __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1.1853 + __ cmpl(rdi_elsize, LogBytesPerInt);
1.1854 + __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1.1855 +#ifdef ASSERT
1.1856 + __ cmpl(rdi_elsize, LogBytesPerLong);
1.1857 + __ jccb(Assembler::notEqual, L_failed);
1.1858 +#endif
1.1859 + __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1.1860 + __ pop(rsi);
1.1861 + __ jump(RuntimeAddress(entry_jlong_arraycopy));
1.1862 +
1.1863 + __ BIND(L_failed);
1.1864 + __ xorptr(rax, rax);
1.1865 + __ notptr(rax); // return -1
1.1866 + __ pop(rdi);
1.1867 + __ pop(rsi);
1.1868 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.1869 + __ ret(0);
1.1870 +
1.1871 + // ObjArrayKlass
1.1872 + __ BIND(L_objArray);
1.1873 + // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1.1874 +
1.1875 + Label L_plain_copy, L_checkcast_copy;
1.1876 + // test array classes for subtyping
1.1877 + __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1.1878 + __ jccb(Assembler::notEqual, L_checkcast_copy);
1.1879 +
1.1880 + // Identically typed arrays can be copied without element-wise checks.
1.1881 + assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1.1882 + arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1.1883 +
1.1884 + __ BIND(L_plain_copy);
1.1885 + __ movl2ptr(count, LENGTH); // elements count
1.1886 + __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1.1887 + __ lea(from, Address(src, src_pos, Address::times_ptr,
1.1888 + arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1.1889 + __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1.1890 + __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1.1891 + arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1.1892 + __ movptr(FROM, from); // src_addr
1.1893 + __ movptr(TO, to); // dst_addr
1.1894 + __ movl(COUNT, count); // count
1.1895 + __ jump(RuntimeAddress(entry_oop_arraycopy));
1.1896 +
1.1897 + __ BIND(L_checkcast_copy);
1.1898 + // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1.1899 + {
1.1900 + // Handy offsets:
1.1901 + int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1.1902 + int sco_offset = in_bytes(Klass::super_check_offset_offset());
1.1903 +
1.1904 + Register rsi_dst_klass = rsi;
1.1905 + Register rdi_temp = rdi;
1.1906 + assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1.1907 + assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1.1908 + Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1.1909 +
1.1910 + // Before looking at dst.length, make sure dst is also an objArray.
1.1911 + __ movptr(rsi_dst_klass, dst_klass_addr);
1.1912 + __ cmpl(dst_klass_lh_addr, objArray_lh);
1.1913 + __ jccb(Assembler::notEqual, L_failed);
1.1914 +
1.1915 + // It is safe to examine both src.length and dst.length.
1.1916 + __ movl2ptr(src_pos, SRC_POS); // reload rsi
1.1917 + arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1.1918 + // (Now src_pos and dst_pos are killed, but not src and dst.)
1.1919 +
1.1920 + // We'll need this temp (don't forget to pop it after the type check).
1.1921 + __ push(rbx);
1.1922 + Register rbx_src_klass = rbx;
1.1923 +
1.1924 + __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1.1925 + __ movptr(rsi_dst_klass, dst_klass_addr);
1.1926 + Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1.1927 + Label L_fail_array_check;
1.1928 + generate_type_check(rbx_src_klass,
1.1929 + super_check_offset_addr, dst_klass_addr,
1.1930 + rdi_temp, NULL, &L_fail_array_check);
1.1931 + // (On fall-through, we have passed the array type check.)
1.1932 + __ pop(rbx);
1.1933 + __ jmp(L_plain_copy);
1.1934 +
1.1935 + __ BIND(L_fail_array_check);
1.1936 + // Reshuffle arguments so we can call checkcast_arraycopy:
1.1937 +
1.1938 + // match initial saves for checkcast_arraycopy
1.1939 + // push(rsi); // already done; see above
1.1940 + // push(rdi); // already done; see above
1.1941 + // push(rbx); // already done; see above
1.1942 +
1.1943 + // Marshal outgoing arguments now, freeing registers.
1.1944 + Address from_arg(rsp, 16+ 4); // from
1.1945 + Address to_arg(rsp, 16+ 8); // to
1.1946 + Address length_arg(rsp, 16+12); // elements count
1.1947 + Address ckoff_arg(rsp, 16+16); // super_check_offset
1.1948 + Address ckval_arg(rsp, 16+20); // super_klass
1.1949 +
1.1950 + Address SRC_POS_arg(rsp, 16+ 8);
1.1951 + Address DST_POS_arg(rsp, 16+16);
1.1952 + Address LENGTH_arg(rsp, 16+20);
1.1953 + // push rbx, changed the incoming offsets (why not just use rbp,??)
1.1954 + // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1.1955 +
1.1956 + __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1.1957 + __ movl2ptr(length, LENGTH_arg); // reload elements count
1.1958 + __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1.1959 + __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1.1960 +
1.1961 + __ movptr(ckval_arg, rbx); // destination element type
1.1962 + __ movl(rbx, Address(rbx, sco_offset));
1.1963 + __ movl(ckoff_arg, rbx); // corresponding class check offset
1.1964 +
1.1965 + __ movl(length_arg, length); // outgoing length argument
1.1966 +
1.1967 + __ lea(from, Address(src, src_pos, Address::times_ptr,
1.1968 + arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.1969 + __ movptr(from_arg, from);
1.1970 +
1.1971 + __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1.1972 + arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.1973 + __ movptr(to_arg, to);
1.1974 + __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1.1975 + }
1.1976 +
1.1977 + return start;
1.1978 + }
1.1979 +
1.1980 + void generate_arraycopy_stubs() {
1.1981 + address entry;
1.1982 + address entry_jbyte_arraycopy;
1.1983 + address entry_jshort_arraycopy;
1.1984 + address entry_jint_arraycopy;
1.1985 + address entry_oop_arraycopy;
1.1986 + address entry_jlong_arraycopy;
1.1987 + address entry_checkcast_arraycopy;
1.1988 +
1.1989 + StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1.1990 + generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1.1991 + "arrayof_jbyte_disjoint_arraycopy");
1.1992 + StubRoutines::_arrayof_jbyte_arraycopy =
1.1993 + generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1.1994 + NULL, "arrayof_jbyte_arraycopy");
1.1995 + StubRoutines::_jbyte_disjoint_arraycopy =
1.1996 + generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1.1997 + "jbyte_disjoint_arraycopy");
1.1998 + StubRoutines::_jbyte_arraycopy =
1.1999 + generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1.2000 + &entry_jbyte_arraycopy, "jbyte_arraycopy");
1.2001 +
1.2002 + StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1.2003 + generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1.2004 + "arrayof_jshort_disjoint_arraycopy");
1.2005 + StubRoutines::_arrayof_jshort_arraycopy =
1.2006 + generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1.2007 + NULL, "arrayof_jshort_arraycopy");
1.2008 + StubRoutines::_jshort_disjoint_arraycopy =
1.2009 + generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1.2010 + "jshort_disjoint_arraycopy");
1.2011 + StubRoutines::_jshort_arraycopy =
1.2012 + generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1.2013 + &entry_jshort_arraycopy, "jshort_arraycopy");
1.2014 +
1.2015 + // Next arrays are always aligned on 4 bytes at least.
1.2016 + StubRoutines::_jint_disjoint_arraycopy =
1.2017 + generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1.2018 + "jint_disjoint_arraycopy");
1.2019 + StubRoutines::_jint_arraycopy =
1.2020 + generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1.2021 + &entry_jint_arraycopy, "jint_arraycopy");
1.2022 +
1.2023 + StubRoutines::_oop_disjoint_arraycopy =
1.2024 + generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1.2025 + "oop_disjoint_arraycopy");
1.2026 + StubRoutines::_oop_arraycopy =
1.2027 + generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
1.2028 + &entry_oop_arraycopy, "oop_arraycopy");
1.2029 +
1.2030 + StubRoutines::_oop_disjoint_arraycopy_uninit =
1.2031 + generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1.2032 + "oop_disjoint_arraycopy_uninit",
1.2033 + /*dest_uninitialized*/true);
1.2034 + StubRoutines::_oop_arraycopy_uninit =
1.2035 + generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
1.2036 + NULL, "oop_arraycopy_uninit",
1.2037 + /*dest_uninitialized*/true);
1.2038 +
1.2039 + StubRoutines::_jlong_disjoint_arraycopy =
1.2040 + generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
1.2041 + StubRoutines::_jlong_arraycopy =
1.2042 + generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
1.2043 + "jlong_arraycopy");
1.2044 +
1.2045 + StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
1.2046 + StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
1.2047 + StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
1.2048 + StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
1.2049 + StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
1.2050 + StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
1.2051 +
1.2052 + StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
1.2053 + StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
1.2054 + StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
1.2055 + StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
1.2056 +
1.2057 + StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
1.2058 + StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
1.2059 + StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
1.2060 + StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
1.2061 +
1.2062 + StubRoutines::_checkcast_arraycopy =
1.2063 + generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
1.2064 + StubRoutines::_checkcast_arraycopy_uninit =
1.2065 + generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
1.2066 +
1.2067 + StubRoutines::_unsafe_arraycopy =
1.2068 + generate_unsafe_copy("unsafe_arraycopy",
1.2069 + entry_jbyte_arraycopy,
1.2070 + entry_jshort_arraycopy,
1.2071 + entry_jint_arraycopy,
1.2072 + entry_jlong_arraycopy);
1.2073 +
1.2074 + StubRoutines::_generic_arraycopy =
1.2075 + generate_generic_copy("generic_arraycopy",
1.2076 + entry_jbyte_arraycopy,
1.2077 + entry_jshort_arraycopy,
1.2078 + entry_jint_arraycopy,
1.2079 + entry_oop_arraycopy,
1.2080 + entry_jlong_arraycopy,
1.2081 + entry_checkcast_arraycopy);
1.2082 + }
1.2083 +
1.2084 + void generate_math_stubs() {
1.2085 + {
1.2086 + StubCodeMark mark(this, "StubRoutines", "log");
1.2087 + StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
1.2088 +
1.2089 + __ fld_d(Address(rsp, 4));
1.2090 + __ flog();
1.2091 + __ ret(0);
1.2092 + }
1.2093 + {
1.2094 + StubCodeMark mark(this, "StubRoutines", "log10");
1.2095 + StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
1.2096 +
1.2097 + __ fld_d(Address(rsp, 4));
1.2098 + __ flog10();
1.2099 + __ ret(0);
1.2100 + }
1.2101 + {
1.2102 + StubCodeMark mark(this, "StubRoutines", "sin");
1.2103 + StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
1.2104 +
1.2105 + __ fld_d(Address(rsp, 4));
1.2106 + __ trigfunc('s');
1.2107 + __ ret(0);
1.2108 + }
1.2109 + {
1.2110 + StubCodeMark mark(this, "StubRoutines", "cos");
1.2111 + StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
1.2112 +
1.2113 + __ fld_d(Address(rsp, 4));
1.2114 + __ trigfunc('c');
1.2115 + __ ret(0);
1.2116 + }
1.2117 + {
1.2118 + StubCodeMark mark(this, "StubRoutines", "tan");
1.2119 + StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
1.2120 +
1.2121 + __ fld_d(Address(rsp, 4));
1.2122 + __ trigfunc('t');
1.2123 + __ ret(0);
1.2124 + }
1.2125 + {
1.2126 + StubCodeMark mark(this, "StubRoutines", "exp");
1.2127 + StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
1.2128 +
1.2129 + __ fld_d(Address(rsp, 4));
1.2130 + __ exp_with_fallback(0);
1.2131 + __ ret(0);
1.2132 + }
1.2133 + {
1.2134 + StubCodeMark mark(this, "StubRoutines", "pow");
1.2135 + StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
1.2136 +
1.2137 + __ fld_d(Address(rsp, 12));
1.2138 + __ fld_d(Address(rsp, 4));
1.2139 + __ pow_with_fallback(0);
1.2140 + __ ret(0);
1.2141 + }
1.2142 + }
1.2143 +
1.2144 + // AES intrinsic stubs
1.2145 + enum {AESBlockSize = 16};
1.2146 +
1.2147 + address generate_key_shuffle_mask() {
1.2148 + __ align(16);
1.2149 + StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
1.2150 + address start = __ pc();
1.2151 + __ emit_data(0x00010203, relocInfo::none, 0 );
1.2152 + __ emit_data(0x04050607, relocInfo::none, 0 );
1.2153 + __ emit_data(0x08090a0b, relocInfo::none, 0 );
1.2154 + __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
1.2155 + return start;
1.2156 + }
1.2157 +
1.2158 + // Utility routine for loading a 128-bit key word in little endian format
1.2159 + // can optionally specify that the shuffle mask is already in an xmmregister
1.2160 + void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
1.2161 + __ movdqu(xmmdst, Address(key, offset));
1.2162 + if (xmm_shuf_mask != NULL) {
1.2163 + __ pshufb(xmmdst, xmm_shuf_mask);
1.2164 + } else {
1.2165 + __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
1.2166 + }
1.2167 + }
1.2168 +
1.2169 + // aesenc using specified key+offset
1.2170 + // can optionally specify that the shuffle mask is already in an xmmregister
1.2171 + void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
1.2172 + load_key(xmmtmp, key, offset, xmm_shuf_mask);
1.2173 + __ aesenc(xmmdst, xmmtmp);
1.2174 + }
1.2175 +
1.2176 + // aesdec using specified key+offset
1.2177 + // can optionally specify that the shuffle mask is already in an xmmregister
1.2178 + void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
1.2179 + load_key(xmmtmp, key, offset, xmm_shuf_mask);
1.2180 + __ aesdec(xmmdst, xmmtmp);
1.2181 + }
1.2182 +
1.2183 +
1.2184 + // Arguments:
1.2185 + //
1.2186 + // Inputs:
1.2187 + // c_rarg0 - source byte array address
1.2188 + // c_rarg1 - destination byte array address
1.2189 + // c_rarg2 - K (key) in little endian int array
1.2190 + //
1.2191 + address generate_aescrypt_encryptBlock() {
1.2192 + assert(UseAES, "need AES instructions and misaligned SSE support");
1.2193 + __ align(CodeEntryAlignment);
1.2194 + StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
1.2195 + Label L_doLast;
1.2196 + address start = __ pc();
1.2197 +
1.2198 + const Register from = rdx; // source array address
1.2199 + const Register to = rdx; // destination array address
1.2200 + const Register key = rcx; // key array address
1.2201 + const Register keylen = rax;
1.2202 + const Address from_param(rbp, 8+0);
1.2203 + const Address to_param (rbp, 8+4);
1.2204 + const Address key_param (rbp, 8+8);
1.2205 +
1.2206 + const XMMRegister xmm_result = xmm0;
1.2207 + const XMMRegister xmm_key_shuf_mask = xmm1;
1.2208 + const XMMRegister xmm_temp1 = xmm2;
1.2209 + const XMMRegister xmm_temp2 = xmm3;
1.2210 + const XMMRegister xmm_temp3 = xmm4;
1.2211 + const XMMRegister xmm_temp4 = xmm5;
1.2212 +
1.2213 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2214 + __ movptr(from, from_param);
1.2215 + __ movptr(key, key_param);
1.2216 +
1.2217 + // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
1.2218 + __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1.2219 +
1.2220 + __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
1.2221 + __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
1.2222 + __ movptr(to, to_param);
1.2223 +
1.2224 + // For encryption, the java expanded key ordering is just what we need
1.2225 +
1.2226 + load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
1.2227 + __ pxor(xmm_result, xmm_temp1);
1.2228 +
1.2229 + load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
1.2230 + load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
1.2231 + load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
1.2232 + load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
1.2233 +
1.2234 + __ aesenc(xmm_result, xmm_temp1);
1.2235 + __ aesenc(xmm_result, xmm_temp2);
1.2236 + __ aesenc(xmm_result, xmm_temp3);
1.2237 + __ aesenc(xmm_result, xmm_temp4);
1.2238 +
1.2239 + load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
1.2240 + load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
1.2241 + load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
1.2242 + load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
1.2243 +
1.2244 + __ aesenc(xmm_result, xmm_temp1);
1.2245 + __ aesenc(xmm_result, xmm_temp2);
1.2246 + __ aesenc(xmm_result, xmm_temp3);
1.2247 + __ aesenc(xmm_result, xmm_temp4);
1.2248 +
1.2249 + load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
1.2250 + load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
1.2251 +
1.2252 + __ cmpl(keylen, 44);
1.2253 + __ jccb(Assembler::equal, L_doLast);
1.2254 +
1.2255 + __ aesenc(xmm_result, xmm_temp1);
1.2256 + __ aesenc(xmm_result, xmm_temp2);
1.2257 +
1.2258 + load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
1.2259 + load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
1.2260 +
1.2261 + __ cmpl(keylen, 52);
1.2262 + __ jccb(Assembler::equal, L_doLast);
1.2263 +
1.2264 + __ aesenc(xmm_result, xmm_temp1);
1.2265 + __ aesenc(xmm_result, xmm_temp2);
1.2266 +
1.2267 + load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
1.2268 + load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
1.2269 +
1.2270 + __ BIND(L_doLast);
1.2271 + __ aesenc(xmm_result, xmm_temp1);
1.2272 + __ aesenclast(xmm_result, xmm_temp2);
1.2273 + __ movdqu(Address(to, 0), xmm_result); // store the result
1.2274 + __ xorptr(rax, rax); // return 0
1.2275 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2276 + __ ret(0);
1.2277 +
1.2278 + return start;
1.2279 + }
1.2280 +
1.2281 +
1.2282 + // Arguments:
1.2283 + //
1.2284 + // Inputs:
1.2285 + // c_rarg0 - source byte array address
1.2286 + // c_rarg1 - destination byte array address
1.2287 + // c_rarg2 - K (key) in little endian int array
1.2288 + //
1.2289 + address generate_aescrypt_decryptBlock() {
1.2290 + assert(UseAES, "need AES instructions and misaligned SSE support");
1.2291 + __ align(CodeEntryAlignment);
1.2292 + StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
1.2293 + Label L_doLast;
1.2294 + address start = __ pc();
1.2295 +
1.2296 + const Register from = rdx; // source array address
1.2297 + const Register to = rdx; // destination array address
1.2298 + const Register key = rcx; // key array address
1.2299 + const Register keylen = rax;
1.2300 + const Address from_param(rbp, 8+0);
1.2301 + const Address to_param (rbp, 8+4);
1.2302 + const Address key_param (rbp, 8+8);
1.2303 +
1.2304 + const XMMRegister xmm_result = xmm0;
1.2305 + const XMMRegister xmm_key_shuf_mask = xmm1;
1.2306 + const XMMRegister xmm_temp1 = xmm2;
1.2307 + const XMMRegister xmm_temp2 = xmm3;
1.2308 + const XMMRegister xmm_temp3 = xmm4;
1.2309 + const XMMRegister xmm_temp4 = xmm5;
1.2310 +
1.2311 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2312 + __ movptr(from, from_param);
1.2313 + __ movptr(key, key_param);
1.2314 +
1.2315 + // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
1.2316 + __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1.2317 +
1.2318 + __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
1.2319 + __ movdqu(xmm_result, Address(from, 0));
1.2320 + __ movptr(to, to_param);
1.2321 +
1.2322 + // for decryption java expanded key ordering is rotated one position from what we want
1.2323 + // so we start from 0x10 here and hit 0x00 last
1.2324 + // we don't know if the key is aligned, hence not using load-execute form
1.2325 + load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
1.2326 + load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
1.2327 + load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
1.2328 + load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
1.2329 +
1.2330 + __ pxor (xmm_result, xmm_temp1);
1.2331 + __ aesdec(xmm_result, xmm_temp2);
1.2332 + __ aesdec(xmm_result, xmm_temp3);
1.2333 + __ aesdec(xmm_result, xmm_temp4);
1.2334 +
1.2335 + load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
1.2336 + load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
1.2337 + load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
1.2338 + load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
1.2339 +
1.2340 + __ aesdec(xmm_result, xmm_temp1);
1.2341 + __ aesdec(xmm_result, xmm_temp2);
1.2342 + __ aesdec(xmm_result, xmm_temp3);
1.2343 + __ aesdec(xmm_result, xmm_temp4);
1.2344 +
1.2345 + load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
1.2346 + load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
1.2347 + load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
1.2348 +
1.2349 + __ cmpl(keylen, 44);
1.2350 + __ jccb(Assembler::equal, L_doLast);
1.2351 +
1.2352 + __ aesdec(xmm_result, xmm_temp1);
1.2353 + __ aesdec(xmm_result, xmm_temp2);
1.2354 +
1.2355 + load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
1.2356 + load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
1.2357 +
1.2358 + __ cmpl(keylen, 52);
1.2359 + __ jccb(Assembler::equal, L_doLast);
1.2360 +
1.2361 + __ aesdec(xmm_result, xmm_temp1);
1.2362 + __ aesdec(xmm_result, xmm_temp2);
1.2363 +
1.2364 + load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
1.2365 + load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
1.2366 +
1.2367 + __ BIND(L_doLast);
1.2368 + __ aesdec(xmm_result, xmm_temp1);
1.2369 + __ aesdec(xmm_result, xmm_temp2);
1.2370 +
1.2371 + // for decryption the aesdeclast operation is always on key+0x00
1.2372 + __ aesdeclast(xmm_result, xmm_temp3);
1.2373 + __ movdqu(Address(to, 0), xmm_result); // store the result
1.2374 + __ xorptr(rax, rax); // return 0
1.2375 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2376 + __ ret(0);
1.2377 +
1.2378 + return start;
1.2379 + }
1.2380 +
1.2381 + void handleSOERegisters(bool saving) {
1.2382 + const int saveFrameSizeInBytes = 4 * wordSize;
1.2383 + const Address saved_rbx (rbp, -3 * wordSize);
1.2384 + const Address saved_rsi (rbp, -2 * wordSize);
1.2385 + const Address saved_rdi (rbp, -1 * wordSize);
1.2386 +
1.2387 + if (saving) {
1.2388 + __ subptr(rsp, saveFrameSizeInBytes);
1.2389 + __ movptr(saved_rsi, rsi);
1.2390 + __ movptr(saved_rdi, rdi);
1.2391 + __ movptr(saved_rbx, rbx);
1.2392 + } else {
1.2393 + // restoring
1.2394 + __ movptr(rsi, saved_rsi);
1.2395 + __ movptr(rdi, saved_rdi);
1.2396 + __ movptr(rbx, saved_rbx);
1.2397 + }
1.2398 + }
1.2399 +
1.2400 + // Arguments:
1.2401 + //
1.2402 + // Inputs:
1.2403 + // c_rarg0 - source byte array address
1.2404 + // c_rarg1 - destination byte array address
1.2405 + // c_rarg2 - K (key) in little endian int array
1.2406 + // c_rarg3 - r vector byte array address
1.2407 + // c_rarg4 - input length
1.2408 + //
1.2409 + // Output:
1.2410 + // rax - input length
1.2411 + //
1.2412 + address generate_cipherBlockChaining_encryptAESCrypt() {
1.2413 + assert(UseAES, "need AES instructions and misaligned SSE support");
1.2414 + __ align(CodeEntryAlignment);
1.2415 + StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
1.2416 + address start = __ pc();
1.2417 +
1.2418 + Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
1.2419 + const Register from = rsi; // source array address
1.2420 + const Register to = rdx; // destination array address
1.2421 + const Register key = rcx; // key array address
1.2422 + const Register rvec = rdi; // r byte array initialized from initvector array address
1.2423 + // and left with the results of the last encryption block
1.2424 + const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
1.2425 + const Register pos = rax;
1.2426 +
1.2427 + // xmm register assignments for the loops below
1.2428 + const XMMRegister xmm_result = xmm0;
1.2429 + const XMMRegister xmm_temp = xmm1;
1.2430 + // first 6 keys preloaded into xmm2-xmm7
1.2431 + const int XMM_REG_NUM_KEY_FIRST = 2;
1.2432 + const int XMM_REG_NUM_KEY_LAST = 7;
1.2433 + const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
1.2434 +
1.2435 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2436 + handleSOERegisters(true /*saving*/);
1.2437 +
1.2438 + // load registers from incoming parameters
1.2439 + const Address from_param(rbp, 8+0);
1.2440 + const Address to_param (rbp, 8+4);
1.2441 + const Address key_param (rbp, 8+8);
1.2442 + const Address rvec_param (rbp, 8+12);
1.2443 + const Address len_param (rbp, 8+16);
1.2444 + __ movptr(from , from_param);
1.2445 + __ movptr(to , to_param);
1.2446 + __ movptr(key , key_param);
1.2447 + __ movptr(rvec , rvec_param);
1.2448 + __ movptr(len_reg , len_param);
1.2449 +
1.2450 + const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
1.2451 + __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
1.2452 + // load up xmm regs 2 thru 7 with keys 0-5
1.2453 + for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2454 + load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
1.2455 + offset += 0x10;
1.2456 + }
1.2457 +
1.2458 + __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
1.2459 +
1.2460 + // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
1.2461 + __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1.2462 + __ cmpl(rax, 44);
1.2463 + __ jcc(Assembler::notEqual, L_key_192_256);
1.2464 +
1.2465 + // 128 bit code follows here
1.2466 + __ movl(pos, 0);
1.2467 + __ align(OptoLoopAlignment);
1.2468 + __ BIND(L_loopTop_128);
1.2469 + __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
1.2470 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2471 +
1.2472 + __ pxor (xmm_result, xmm_key0); // do the aes rounds
1.2473 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2474 + __ aesenc(xmm_result, as_XMMRegister(rnum));
1.2475 + }
1.2476 + for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
1.2477 + aes_enc_key(xmm_result, xmm_temp, key, key_offset);
1.2478 + }
1.2479 + load_key(xmm_temp, key, 0xa0);
1.2480 + __ aesenclast(xmm_result, xmm_temp);
1.2481 +
1.2482 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2483 + // no need to store r to memory until we exit
1.2484 + __ addptr(pos, AESBlockSize);
1.2485 + __ subptr(len_reg, AESBlockSize);
1.2486 + __ jcc(Assembler::notEqual, L_loopTop_128);
1.2487 +
1.2488 + __ BIND(L_exit);
1.2489 + __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
1.2490 +
1.2491 + handleSOERegisters(false /*restoring*/);
1.2492 + __ movptr(rax, len_param); // return length
1.2493 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2494 + __ ret(0);
1.2495 +
1.2496 + __ BIND(L_key_192_256);
1.2497 + // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
1.2498 + __ cmpl(rax, 52);
1.2499 + __ jcc(Assembler::notEqual, L_key_256);
1.2500 +
1.2501 + // 192-bit code follows here (could be changed to use more xmm registers)
1.2502 + __ movl(pos, 0);
1.2503 + __ align(OptoLoopAlignment);
1.2504 + __ BIND(L_loopTop_192);
1.2505 + __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
1.2506 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2507 +
1.2508 + __ pxor (xmm_result, xmm_key0); // do the aes rounds
1.2509 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2510 + __ aesenc(xmm_result, as_XMMRegister(rnum));
1.2511 + }
1.2512 + for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
1.2513 + aes_enc_key(xmm_result, xmm_temp, key, key_offset);
1.2514 + }
1.2515 + load_key(xmm_temp, key, 0xc0);
1.2516 + __ aesenclast(xmm_result, xmm_temp);
1.2517 +
1.2518 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2519 + // no need to store r to memory until we exit
1.2520 + __ addptr(pos, AESBlockSize);
1.2521 + __ subptr(len_reg, AESBlockSize);
1.2522 + __ jcc(Assembler::notEqual, L_loopTop_192);
1.2523 + __ jmp(L_exit);
1.2524 +
1.2525 + __ BIND(L_key_256);
1.2526 + // 256-bit code follows here (could be changed to use more xmm registers)
1.2527 + __ movl(pos, 0);
1.2528 + __ align(OptoLoopAlignment);
1.2529 + __ BIND(L_loopTop_256);
1.2530 + __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
1.2531 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2532 +
1.2533 + __ pxor (xmm_result, xmm_key0); // do the aes rounds
1.2534 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2535 + __ aesenc(xmm_result, as_XMMRegister(rnum));
1.2536 + }
1.2537 + for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
1.2538 + aes_enc_key(xmm_result, xmm_temp, key, key_offset);
1.2539 + }
1.2540 + load_key(xmm_temp, key, 0xe0);
1.2541 + __ aesenclast(xmm_result, xmm_temp);
1.2542 +
1.2543 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2544 + // no need to store r to memory until we exit
1.2545 + __ addptr(pos, AESBlockSize);
1.2546 + __ subptr(len_reg, AESBlockSize);
1.2547 + __ jcc(Assembler::notEqual, L_loopTop_256);
1.2548 + __ jmp(L_exit);
1.2549 +
1.2550 + return start;
1.2551 + }
1.2552 +
1.2553 +
1.2554 + // CBC AES Decryption.
1.2555 + // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
1.2556 + //
1.2557 + // Arguments:
1.2558 + //
1.2559 + // Inputs:
1.2560 + // c_rarg0 - source byte array address
1.2561 + // c_rarg1 - destination byte array address
1.2562 + // c_rarg2 - K (key) in little endian int array
1.2563 + // c_rarg3 - r vector byte array address
1.2564 + // c_rarg4 - input length
1.2565 + //
1.2566 + // Output:
1.2567 + // rax - input length
1.2568 + //
1.2569 +
1.2570 + address generate_cipherBlockChaining_decryptAESCrypt() {
1.2571 + assert(UseAES, "need AES instructions and misaligned SSE support");
1.2572 + __ align(CodeEntryAlignment);
1.2573 + StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
1.2574 + address start = __ pc();
1.2575 +
1.2576 + Label L_exit, L_key_192_256, L_key_256;
1.2577 + Label L_singleBlock_loopTop_128;
1.2578 + Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
1.2579 + const Register from = rsi; // source array address
1.2580 + const Register to = rdx; // destination array address
1.2581 + const Register key = rcx; // key array address
1.2582 + const Register rvec = rdi; // r byte array initialized from initvector array address
1.2583 + // and left with the results of the last encryption block
1.2584 + const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
1.2585 + const Register pos = rax;
1.2586 +
1.2587 + // xmm register assignments for the loops below
1.2588 + const XMMRegister xmm_result = xmm0;
1.2589 + const XMMRegister xmm_temp = xmm1;
1.2590 + // first 6 keys preloaded into xmm2-xmm7
1.2591 + const int XMM_REG_NUM_KEY_FIRST = 2;
1.2592 + const int XMM_REG_NUM_KEY_LAST = 7;
1.2593 + const int FIRST_NON_REG_KEY_offset = 0x70;
1.2594 + const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
1.2595 +
1.2596 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2597 + handleSOERegisters(true /*saving*/);
1.2598 +
1.2599 + // load registers from incoming parameters
1.2600 + const Address from_param(rbp, 8+0);
1.2601 + const Address to_param (rbp, 8+4);
1.2602 + const Address key_param (rbp, 8+8);
1.2603 + const Address rvec_param (rbp, 8+12);
1.2604 + const Address len_param (rbp, 8+16);
1.2605 + __ movptr(from , from_param);
1.2606 + __ movptr(to , to_param);
1.2607 + __ movptr(key , key_param);
1.2608 + __ movptr(rvec , rvec_param);
1.2609 + __ movptr(len_reg , len_param);
1.2610 +
1.2611 + // the java expanded key ordering is rotated one position from what we want
1.2612 + // so we start from 0x10 here and hit 0x00 last
1.2613 + const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
1.2614 + __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
1.2615 + // load up xmm regs 2 thru 6 with first 5 keys
1.2616 + for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2617 + load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
1.2618 + offset += 0x10;
1.2619 + }
1.2620 +
1.2621 + // inside here, use the rvec register to point to previous block cipher
1.2622 + // with which we xor at the end of each newly decrypted block
1.2623 + const Register prev_block_cipher_ptr = rvec;
1.2624 +
1.2625 + // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
1.2626 + __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
1.2627 + __ cmpl(rax, 44);
1.2628 + __ jcc(Assembler::notEqual, L_key_192_256);
1.2629 +
1.2630 +
1.2631 + // 128-bit code follows here, parallelized
1.2632 + __ movl(pos, 0);
1.2633 + __ align(OptoLoopAlignment);
1.2634 + __ BIND(L_singleBlock_loopTop_128);
1.2635 + __ cmpptr(len_reg, 0); // any blocks left??
1.2636 + __ jcc(Assembler::equal, L_exit);
1.2637 + __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
1.2638 + __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
1.2639 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2640 + __ aesdec(xmm_result, as_XMMRegister(rnum));
1.2641 + }
1.2642 + for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
1.2643 + aes_dec_key(xmm_result, xmm_temp, key, key_offset);
1.2644 + }
1.2645 + load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
1.2646 + __ aesdeclast(xmm_result, xmm_temp);
1.2647 + __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
1.2648 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2649 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2650 + // no need to store r to memory until we exit
1.2651 + __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
1.2652 + __ addptr(pos, AESBlockSize);
1.2653 + __ subptr(len_reg, AESBlockSize);
1.2654 + __ jmp(L_singleBlock_loopTop_128);
1.2655 +
1.2656 +
1.2657 + __ BIND(L_exit);
1.2658 + __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
1.2659 + __ movptr(rvec , rvec_param); // restore this since used in loop
1.2660 + __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
1.2661 + handleSOERegisters(false /*restoring*/);
1.2662 + __ movptr(rax, len_param); // return length
1.2663 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2664 + __ ret(0);
1.2665 +
1.2666 +
1.2667 + __ BIND(L_key_192_256);
1.2668 + // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
1.2669 + __ cmpl(rax, 52);
1.2670 + __ jcc(Assembler::notEqual, L_key_256);
1.2671 +
1.2672 + // 192-bit code follows here (could be optimized to use parallelism)
1.2673 + __ movl(pos, 0);
1.2674 + __ align(OptoLoopAlignment);
1.2675 + __ BIND(L_singleBlock_loopTop_192);
1.2676 + __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
1.2677 + __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
1.2678 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2679 + __ aesdec(xmm_result, as_XMMRegister(rnum));
1.2680 + }
1.2681 + for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
1.2682 + aes_dec_key(xmm_result, xmm_temp, key, key_offset);
1.2683 + }
1.2684 + load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
1.2685 + __ aesdeclast(xmm_result, xmm_temp);
1.2686 + __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
1.2687 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2688 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2689 + // no need to store r to memory until we exit
1.2690 + __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
1.2691 + __ addptr(pos, AESBlockSize);
1.2692 + __ subptr(len_reg, AESBlockSize);
1.2693 + __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
1.2694 + __ jmp(L_exit);
1.2695 +
1.2696 + __ BIND(L_key_256);
1.2697 + // 256-bit code follows here (could be optimized to use parallelism)
1.2698 + __ movl(pos, 0);
1.2699 + __ align(OptoLoopAlignment);
1.2700 + __ BIND(L_singleBlock_loopTop_256);
1.2701 + __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
1.2702 + __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
1.2703 + for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
1.2704 + __ aesdec(xmm_result, as_XMMRegister(rnum));
1.2705 + }
1.2706 + for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
1.2707 + aes_dec_key(xmm_result, xmm_temp, key, key_offset);
1.2708 + }
1.2709 + load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
1.2710 + __ aesdeclast(xmm_result, xmm_temp);
1.2711 + __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
1.2712 + __ pxor (xmm_result, xmm_temp); // xor with the current r vector
1.2713 + __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
1.2714 + // no need to store r to memory until we exit
1.2715 + __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
1.2716 + __ addptr(pos, AESBlockSize);
1.2717 + __ subptr(len_reg, AESBlockSize);
1.2718 + __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
1.2719 + __ jmp(L_exit);
1.2720 +
1.2721 + return start;
1.2722 + }
1.2723 +
1.2724 + /**
1.2725 + * Arguments:
1.2726 + *
1.2727 + * Inputs:
1.2728 + * rsp(4) - int crc
1.2729 + * rsp(8) - byte* buf
1.2730 + * rsp(12) - int length
1.2731 + *
1.2732 + * Ouput:
1.2733 + * rax - int crc result
1.2734 + */
1.2735 + address generate_updateBytesCRC32() {
1.2736 + assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
1.2737 +
1.2738 + __ align(CodeEntryAlignment);
1.2739 + StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
1.2740 +
1.2741 + address start = __ pc();
1.2742 +
1.2743 + const Register crc = rdx; // crc
1.2744 + const Register buf = rsi; // source java byte array address
1.2745 + const Register len = rcx; // length
1.2746 + const Register table = rdi; // crc_table address (reuse register)
1.2747 + const Register tmp = rbx;
1.2748 + assert_different_registers(crc, buf, len, table, tmp, rax);
1.2749 +
1.2750 + BLOCK_COMMENT("Entry:");
1.2751 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2752 + __ push(rsi);
1.2753 + __ push(rdi);
1.2754 + __ push(rbx);
1.2755 +
1.2756 + Address crc_arg(rbp, 8 + 0);
1.2757 + Address buf_arg(rbp, 8 + 4);
1.2758 + Address len_arg(rbp, 8 + 8);
1.2759 +
1.2760 + // Load up:
1.2761 + __ movl(crc, crc_arg);
1.2762 + __ movptr(buf, buf_arg);
1.2763 + __ movl(len, len_arg);
1.2764 +
1.2765 + __ kernel_crc32(crc, buf, len, table, tmp);
1.2766 +
1.2767 + __ movl(rax, crc);
1.2768 + __ pop(rbx);
1.2769 + __ pop(rdi);
1.2770 + __ pop(rsi);
1.2771 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2772 + __ ret(0);
1.2773 +
1.2774 + return start;
1.2775 + }
1.2776 +
1.2777 + // Safefetch stubs.
1.2778 + void generate_safefetch(const char* name, int size, address* entry,
1.2779 + address* fault_pc, address* continuation_pc) {
1.2780 + // safefetch signatures:
1.2781 + // int SafeFetch32(int* adr, int errValue);
1.2782 + // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
1.2783 +
1.2784 + StubCodeMark mark(this, "StubRoutines", name);
1.2785 +
1.2786 + // Entry point, pc or function descriptor.
1.2787 + *entry = __ pc();
1.2788 +
1.2789 + __ movl(rax, Address(rsp, 0x8));
1.2790 + __ movl(rcx, Address(rsp, 0x4));
1.2791 + // Load *adr into eax, may fault.
1.2792 + *fault_pc = __ pc();
1.2793 + switch (size) {
1.2794 + case 4:
1.2795 + // int32_t
1.2796 + __ movl(rax, Address(rcx, 0));
1.2797 + break;
1.2798 + case 8:
1.2799 + // int64_t
1.2800 + Unimplemented();
1.2801 + break;
1.2802 + default:
1.2803 + ShouldNotReachHere();
1.2804 + }
1.2805 +
1.2806 + // Return errValue or *adr.
1.2807 + *continuation_pc = __ pc();
1.2808 + __ ret(0);
1.2809 + }
1.2810 +
1.2811 + public:
1.2812 + // Information about frame layout at time of blocking runtime call.
1.2813 + // Note that we only have to preserve callee-saved registers since
1.2814 + // the compilers are responsible for supplying a continuation point
1.2815 + // if they expect all registers to be preserved.
1.2816 + enum layout {
1.2817 + thread_off, // last_java_sp
1.2818 + arg1_off,
1.2819 + arg2_off,
1.2820 + rbp_off, // callee saved register
1.2821 + ret_pc,
1.2822 + framesize
1.2823 + };
1.2824 +
1.2825 + private:
1.2826 +
1.2827 +#undef __
1.2828 +#define __ masm->
1.2829 +
1.2830 + //------------------------------------------------------------------------------------------------------------------------
1.2831 + // Continuation point for throwing of implicit exceptions that are not handled in
1.2832 + // the current activation. Fabricates an exception oop and initiates normal
1.2833 + // exception dispatching in this frame.
1.2834 + //
1.2835 + // Previously the compiler (c2) allowed for callee save registers on Java calls.
1.2836 + // This is no longer true after adapter frames were removed but could possibly
1.2837 + // be brought back in the future if the interpreter code was reworked and it
1.2838 + // was deemed worthwhile. The comment below was left to describe what must
1.2839 + // happen here if callee saves were resurrected. As it stands now this stub
1.2840 + // could actually be a vanilla BufferBlob and have now oopMap at all.
1.2841 + // Since it doesn't make much difference we've chosen to leave it the
1.2842 + // way it was in the callee save days and keep the comment.
1.2843 +
1.2844 + // If we need to preserve callee-saved values we need a callee-saved oop map and
1.2845 + // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
1.2846 + // If the compiler needs all registers to be preserved between the fault
1.2847 + // point and the exception handler then it must assume responsibility for that in
1.2848 + // AbstractCompiler::continuation_for_implicit_null_exception or
1.2849 + // continuation_for_implicit_division_by_zero_exception. All other implicit
1.2850 + // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
1.2851 + // either at call sites or otherwise assume that stack unwinding will be initiated,
1.2852 + // so caller saved registers were assumed volatile in the compiler.
1.2853 + address generate_throw_exception(const char* name, address runtime_entry,
1.2854 + Register arg1 = noreg, Register arg2 = noreg) {
1.2855 +
1.2856 + int insts_size = 256;
1.2857 + int locs_size = 32;
1.2858 +
1.2859 + CodeBuffer code(name, insts_size, locs_size);
1.2860 + OopMapSet* oop_maps = new OopMapSet();
1.2861 + MacroAssembler* masm = new MacroAssembler(&code);
1.2862 +
1.2863 + address start = __ pc();
1.2864 +
1.2865 + // This is an inlined and slightly modified version of call_VM
1.2866 + // which has the ability to fetch the return PC out of
1.2867 + // thread-local storage and also sets up last_Java_sp slightly
1.2868 + // differently than the real call_VM
1.2869 + Register java_thread = rbx;
1.2870 + __ get_thread(java_thread);
1.2871 +
1.2872 + __ enter(); // required for proper stackwalking of RuntimeStub frame
1.2873 +
1.2874 + // pc and rbp, already pushed
1.2875 + __ subptr(rsp, (framesize-2) * wordSize); // prolog
1.2876 +
1.2877 + // Frame is now completed as far as size and linkage.
1.2878 +
1.2879 + int frame_complete = __ pc() - start;
1.2880 +
1.2881 + // push java thread (becomes first argument of C function)
1.2882 + __ movptr(Address(rsp, thread_off * wordSize), java_thread);
1.2883 + if (arg1 != noreg) {
1.2884 + __ movptr(Address(rsp, arg1_off * wordSize), arg1);
1.2885 + }
1.2886 + if (arg2 != noreg) {
1.2887 + assert(arg1 != noreg, "missing reg arg");
1.2888 + __ movptr(Address(rsp, arg2_off * wordSize), arg2);
1.2889 + }
1.2890 +
1.2891 + // Set up last_Java_sp and last_Java_fp
1.2892 + __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
1.2893 +
1.2894 + // Call runtime
1.2895 + BLOCK_COMMENT("call runtime_entry");
1.2896 + __ call(RuntimeAddress(runtime_entry));
1.2897 + // Generate oop map
1.2898 + OopMap* map = new OopMap(framesize, 0);
1.2899 + oop_maps->add_gc_map(__ pc() - start, map);
1.2900 +
1.2901 + // restore the thread (cannot use the pushed argument since arguments
1.2902 + // may be overwritten by C code generated by an optimizing compiler);
1.2903 + // however can use the register value directly if it is callee saved.
1.2904 + __ get_thread(java_thread);
1.2905 +
1.2906 + __ reset_last_Java_frame(java_thread, true, false);
1.2907 +
1.2908 + __ leave(); // required for proper stackwalking of RuntimeStub frame
1.2909 +
1.2910 + // check for pending exceptions
1.2911 +#ifdef ASSERT
1.2912 + Label L;
1.2913 + __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1.2914 + __ jcc(Assembler::notEqual, L);
1.2915 + __ should_not_reach_here();
1.2916 + __ bind(L);
1.2917 +#endif /* ASSERT */
1.2918 + __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1.2919 +
1.2920 +
1.2921 + RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
1.2922 + return stub->entry_point();
1.2923 + }
1.2924 +
1.2925 +
1.2926 + void create_control_words() {
1.2927 + // Round to nearest, 53-bit mode, exceptions masked
1.2928 + StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
1.2929 + // Round to zero, 53-bit mode, exception mased
1.2930 + StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
1.2931 + // Round to nearest, 24-bit mode, exceptions masked
1.2932 + StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
1.2933 + // Round to nearest, 64-bit mode, exceptions masked
1.2934 + StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
1.2935 + // Round to nearest, 64-bit mode, exceptions masked
1.2936 + StubRoutines::_mxcsr_std = 0x1F80;
1.2937 + // Note: the following two constants are 80-bit values
1.2938 + // layout is critical for correct loading by FPU.
1.2939 + // Bias for strict fp multiply/divide
1.2940 + StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
1.2941 + StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
1.2942 + StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
1.2943 + // Un-Bias for strict fp multiply/divide
1.2944 + StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
1.2945 + StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
1.2946 + StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
1.2947 + }
1.2948 +
1.2949 + //---------------------------------------------------------------------------
1.2950 + // Initialization
1.2951 +
1.2952 + void generate_initial() {
1.2953 + // Generates all stubs and initializes the entry points
1.2954 +
1.2955 + //------------------------------------------------------------------------------------------------------------------------
1.2956 + // entry points that exist in all platforms
1.2957 + // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
1.2958 + // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
1.2959 + StubRoutines::_forward_exception_entry = generate_forward_exception();
1.2960 +
1.2961 + StubRoutines::_call_stub_entry =
1.2962 + generate_call_stub(StubRoutines::_call_stub_return_address);
1.2963 + // is referenced by megamorphic call
1.2964 + StubRoutines::_catch_exception_entry = generate_catch_exception();
1.2965 +
1.2966 + // These are currently used by Solaris/Intel
1.2967 + StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
1.2968 +
1.2969 + StubRoutines::_handler_for_unsafe_access_entry =
1.2970 + generate_handler_for_unsafe_access();
1.2971 +
1.2972 + // platform dependent
1.2973 + create_control_words();
1.2974 +
1.2975 + StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
1.2976 + StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
1.2977 + StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
1.2978 + CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1.2979 + StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
1.2980 + CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1.2981 +
1.2982 + // Build this early so it's available for the interpreter
1.2983 + StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
1.2984 +
1.2985 + if (UseCRC32Intrinsics) {
1.2986 + // set table address before stub generation which use it
1.2987 + StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
1.2988 + StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
1.2989 + }
1.2990 + }
1.2991 +
1.2992 +
1.2993 + void generate_all() {
1.2994 + // Generates all stubs and initializes the entry points
1.2995 +
1.2996 + // These entry points require SharedInfo::stack0 to be set up in non-core builds
1.2997 + // and need to be relocatable, so they each fabricate a RuntimeStub internally.
1.2998 + StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
1.2999 + StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
1.3000 + StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
1.3001 +
1.3002 + //------------------------------------------------------------------------------------------------------------------------
1.3003 + // entry points that are platform specific
1.3004 +
1.3005 + // support for verify_oop (must happen after universe_init)
1.3006 + StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
1.3007 +
1.3008 + // arraycopy stubs used by compilers
1.3009 + generate_arraycopy_stubs();
1.3010 +
1.3011 + generate_math_stubs();
1.3012 +
1.3013 + // don't bother generating these AES intrinsic stubs unless global flag is set
1.3014 + if (UseAESIntrinsics) {
1.3015 + StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
1.3016 +
1.3017 + StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
1.3018 + StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
1.3019 + StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
1.3020 + StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
1.3021 + }
1.3022 +
1.3023 + // Safefetch stubs.
1.3024 + generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
1.3025 + &StubRoutines::_safefetch32_fault_pc,
1.3026 + &StubRoutines::_safefetch32_continuation_pc);
1.3027 + StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
1.3028 + StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
1.3029 + StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
1.3030 + }
1.3031 +
1.3032 +
1.3033 + public:
1.3034 + StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
1.3035 + if (all) {
1.3036 + generate_all();
1.3037 + } else {
1.3038 + generate_initial();
1.3039 + }
1.3040 + }
1.3041 +}; // end class declaration
1.3042 +
1.3043 +
1.3044 +void StubGenerator_generate(CodeBuffer* code, bool all) {
1.3045 + StubGenerator g(code, all);
1.3046 +}
