1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp Sat Dec 01 00:00:00 2007 +0000
1.3 @@ -0,0 +1,898 @@
1.4 +/*
1.5 + * Copyright 1999-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or
1.24 + * have any questions.
1.25 + *
1.26 + */
1.27 +
1.28 +#include "incls/_precompiled.incl"
1.29 +#include "incls/_c1_Runtime1_sparc.cpp.incl"
1.30 +
1.31 +// Implementation of StubAssembler
1.32 +
1.33 +int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) {
1.34 + // for sparc changing the number of arguments doesn't change
1.35 + // anything about the frame size so we'll always lie and claim that
1.36 + // we are only passing 1 argument.
1.37 + set_num_rt_args(1);
1.38 +
1.39 + assert_not_delayed();
1.40 + // bang stack before going to runtime
1.41 + set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
1.42 + st(G0, SP, G3_scratch);
1.43 +
1.44 + // debugging support
1.45 + assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
1.46 +
1.47 + set_last_Java_frame(SP, noreg);
1.48 + if (VerifyThread) mov(G2_thread, O0); // about to be smashed; pass early
1.49 + save_thread(L7_thread_cache);
1.50 + // do the call
1.51 + call(entry_point, relocInfo::runtime_call_type);
1.52 + if (!VerifyThread) {
1.53 + delayed()->mov(G2_thread, O0); // pass thread as first argument
1.54 + } else {
1.55 + delayed()->nop(); // (thread already passed)
1.56 + }
1.57 + int call_offset = offset(); // offset of return address
1.58 + restore_thread(L7_thread_cache);
1.59 + reset_last_Java_frame();
1.60 +
1.61 + // check for pending exceptions
1.62 + { Label L;
1.63 + Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset()));
1.64 + ld_ptr(exception_addr, Gtemp);
1.65 + br_null(Gtemp, false, pt, L);
1.66 + delayed()->nop();
1.67 + Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset()));
1.68 + st_ptr(G0, vm_result_addr);
1.69 + Address vm_result_addr_2(G2_thread, 0, in_bytes(JavaThread::vm_result_2_offset()));
1.70 + st_ptr(G0, vm_result_addr_2);
1.71 +
1.72 + if (frame_size() == no_frame_size) {
1.73 + // we use O7 linkage so that forward_exception_entry has the issuing PC
1.74 + call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
1.75 + delayed()->restore();
1.76 + } else if (_stub_id == Runtime1::forward_exception_id) {
1.77 + should_not_reach_here();
1.78 + } else {
1.79 + Address exc(G4, Runtime1::entry_for(Runtime1::forward_exception_id));
1.80 + jump_to(exc, 0);
1.81 + delayed()->nop();
1.82 + }
1.83 + bind(L);
1.84 + }
1.85 +
1.86 + // get oop result if there is one and reset the value in the thread
1.87 + if (oop_result1->is_valid()) { // get oop result if there is one and reset it in the thread
1.88 + get_vm_result (oop_result1);
1.89 + } else {
1.90 + // be a little paranoid and clear the result
1.91 + Address vm_result_addr(G2_thread, 0, in_bytes(JavaThread::vm_result_offset()));
1.92 + st_ptr(G0, vm_result_addr);
1.93 + }
1.94 +
1.95 + if (oop_result2->is_valid()) {
1.96 + get_vm_result_2(oop_result2);
1.97 + } else {
1.98 + // be a little paranoid and clear the result
1.99 + Address vm_result_addr_2(G2_thread, 0, in_bytes(JavaThread::vm_result_2_offset()));
1.100 + st_ptr(G0, vm_result_addr_2);
1.101 + }
1.102 +
1.103 + return call_offset;
1.104 +}
1.105 +
1.106 +
1.107 +int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
1.108 + // O0 is reserved for the thread
1.109 + mov(arg1, O1);
1.110 + return call_RT(oop_result1, oop_result2, entry, 1);
1.111 +}
1.112 +
1.113 +
1.114 +int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
1.115 + // O0 is reserved for the thread
1.116 + mov(arg1, O1);
1.117 + mov(arg2, O2); assert(arg2 != O1, "smashed argument");
1.118 + return call_RT(oop_result1, oop_result2, entry, 2);
1.119 +}
1.120 +
1.121 +
1.122 +int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
1.123 + // O0 is reserved for the thread
1.124 + mov(arg1, O1);
1.125 + mov(arg2, O2); assert(arg2 != O1, "smashed argument");
1.126 + mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
1.127 + return call_RT(oop_result1, oop_result2, entry, 3);
1.128 +}
1.129 +
1.130 +
1.131 +// Implementation of Runtime1
1.132 +
1.133 +#define __ sasm->
1.134 +
1.135 +static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
1.136 +static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
1.137 +static int reg_save_size_in_words;
1.138 +static int frame_size_in_bytes = -1;
1.139 +
1.140 +static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
1.141 + assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
1.142 + " mismatch in calculation");
1.143 + sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
1.144 + int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
1.145 + OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
1.146 +
1.147 + int i;
1.148 + for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
1.149 + Register r = as_Register(i);
1.150 + if (r == G1 || r == G3 || r == G4 || r == G5) {
1.151 + int sp_offset = cpu_reg_save_offsets[i];
1.152 + oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
1.153 + r->as_VMReg());
1.154 + }
1.155 + }
1.156 +
1.157 + if (save_fpu_registers) {
1.158 + for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
1.159 + FloatRegister r = as_FloatRegister(i);
1.160 + int sp_offset = fpu_reg_save_offsets[i];
1.161 + oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
1.162 + r->as_VMReg());
1.163 + }
1.164 + }
1.165 + return oop_map;
1.166 +}
1.167 +
1.168 +static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
1.169 + assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
1.170 + " mismatch in calculation");
1.171 + __ save_frame_c1(frame_size_in_bytes);
1.172 + sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
1.173 +
1.174 + // Record volatile registers as callee-save values in an OopMap so their save locations will be
1.175 + // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
1.176 + // deoptimization; see compiledVFrame::create_stack_value). The caller's I, L and O registers
1.177 + // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
1.178 + // (as the stub's I's) when the runtime routine called by the stub creates its frame.
1.179 + // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
1.180 +
1.181 + int i;
1.182 + for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
1.183 + Register r = as_Register(i);
1.184 + if (r == G1 || r == G3 || r == G4 || r == G5) {
1.185 + int sp_offset = cpu_reg_save_offsets[i];
1.186 + __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
1.187 + }
1.188 + }
1.189 +
1.190 + if (save_fpu_registers) {
1.191 + for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
1.192 + FloatRegister r = as_FloatRegister(i);
1.193 + int sp_offset = fpu_reg_save_offsets[i];
1.194 + __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
1.195 + }
1.196 + }
1.197 +
1.198 + return generate_oop_map(sasm, save_fpu_registers);
1.199 +}
1.200 +
1.201 +static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
1.202 + for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
1.203 + Register r = as_Register(i);
1.204 + if (r == G1 || r == G3 || r == G4 || r == G5) {
1.205 + __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
1.206 + }
1.207 + }
1.208 +
1.209 + if (restore_fpu_registers) {
1.210 + for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
1.211 + FloatRegister r = as_FloatRegister(i);
1.212 + __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
1.213 + }
1.214 + }
1.215 +}
1.216 +
1.217 +
1.218 +void Runtime1::initialize_pd() {
1.219 + // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
1.220 + //
1.221 + // A stub routine will have a frame that is at least large enough to hold
1.222 + // a register window save area (obviously) and the volatile g registers
1.223 + // and floating registers. A user of save_live_registers can have a frame
1.224 + // that has more scratch area in it (although typically they will use L-regs).
1.225 + // in that case the frame will look like this (stack growing down)
1.226 + //
1.227 + // FP -> | |
1.228 + // | scratch mem |
1.229 + // | " " |
1.230 + // --------------
1.231 + // | float regs |
1.232 + // | " " |
1.233 + // ---------------
1.234 + // | G regs |
1.235 + // | " " |
1.236 + // ---------------
1.237 + // | abi reg. |
1.238 + // | window save |
1.239 + // | area |
1.240 + // SP -> ---------------
1.241 + //
1.242 + int i;
1.243 + int sp_offset = round_to(frame::register_save_words, 2); // start doubleword aligned
1.244 +
1.245 + // only G int registers are saved explicitly; others are found in register windows
1.246 + for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
1.247 + Register r = as_Register(i);
1.248 + if (r == G1 || r == G3 || r == G4 || r == G5) {
1.249 + cpu_reg_save_offsets[i] = sp_offset;
1.250 + sp_offset++;
1.251 + }
1.252 + }
1.253 +
1.254 + // all float registers are saved explicitly
1.255 + assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
1.256 + for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
1.257 + fpu_reg_save_offsets[i] = sp_offset;
1.258 + sp_offset++;
1.259 + }
1.260 + reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
1.261 + // this should match assembler::total_frame_size_in_bytes, which
1.262 + // isn't callable from this context. It's checked by an assert when
1.263 + // it's used though.
1.264 + frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
1.265 +}
1.266 +
1.267 +
1.268 +OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
1.269 + // make a frame and preserve the caller's caller-save registers
1.270 + OopMap* oop_map = save_live_registers(sasm);
1.271 + int call_offset;
1.272 + if (!has_argument) {
1.273 + call_offset = __ call_RT(noreg, noreg, target);
1.274 + } else {
1.275 + call_offset = __ call_RT(noreg, noreg, target, G4);
1.276 + }
1.277 + OopMapSet* oop_maps = new OopMapSet();
1.278 + oop_maps->add_gc_map(call_offset, oop_map);
1.279 +
1.280 + __ should_not_reach_here();
1.281 + return oop_maps;
1.282 +}
1.283 +
1.284 +
1.285 +OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
1.286 + Register arg1, Register arg2, Register arg3) {
1.287 + // make a frame and preserve the caller's caller-save registers
1.288 + OopMap* oop_map = save_live_registers(sasm);
1.289 +
1.290 + int call_offset;
1.291 + if (arg1 == noreg) {
1.292 + call_offset = __ call_RT(result, noreg, target);
1.293 + } else if (arg2 == noreg) {
1.294 + call_offset = __ call_RT(result, noreg, target, arg1);
1.295 + } else if (arg3 == noreg) {
1.296 + call_offset = __ call_RT(result, noreg, target, arg1, arg2);
1.297 + } else {
1.298 + call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
1.299 + }
1.300 + OopMapSet* oop_maps = NULL;
1.301 +
1.302 + oop_maps = new OopMapSet();
1.303 + oop_maps->add_gc_map(call_offset, oop_map);
1.304 + restore_live_registers(sasm);
1.305 +
1.306 + __ ret();
1.307 + __ delayed()->restore();
1.308 +
1.309 + return oop_maps;
1.310 +}
1.311 +
1.312 +
1.313 +OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
1.314 + // make a frame and preserve the caller's caller-save registers
1.315 + OopMap* oop_map = save_live_registers(sasm);
1.316 +
1.317 + // call the runtime patching routine, returns non-zero if nmethod got deopted.
1.318 + int call_offset = __ call_RT(noreg, noreg, target);
1.319 + OopMapSet* oop_maps = new OopMapSet();
1.320 + oop_maps->add_gc_map(call_offset, oop_map);
1.321 +
1.322 + // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
1.323 + // deoptimization handler entry that will cause re-execution of the current bytecode
1.324 + DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1.325 + assert(deopt_blob != NULL, "deoptimization blob must have been created");
1.326 +
1.327 + Label no_deopt;
1.328 + __ tst(O0);
1.329 + __ brx(Assembler::equal, false, Assembler::pt, no_deopt);
1.330 + __ delayed()->nop();
1.331 +
1.332 + // return to the deoptimization handler entry for unpacking and rexecute
1.333 + // if we simply returned the we'd deopt as if any call we patched had just
1.334 + // returned.
1.335 +
1.336 + restore_live_registers(sasm);
1.337 + __ restore();
1.338 + __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
1.339 + __ delayed()->nop();
1.340 +
1.341 + __ bind(no_deopt);
1.342 + restore_live_registers(sasm);
1.343 + __ ret();
1.344 + __ delayed()->restore();
1.345 +
1.346 + return oop_maps;
1.347 +}
1.348 +
1.349 +OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
1.350 +
1.351 + OopMapSet* oop_maps = NULL;
1.352 + // for better readability
1.353 + const bool must_gc_arguments = true;
1.354 + const bool dont_gc_arguments = false;
1.355 +
1.356 + // stub code & info for the different stubs
1.357 + switch (id) {
1.358 + case forward_exception_id:
1.359 + {
1.360 + // we're handling an exception in the context of a compiled
1.361 + // frame. The registers have been saved in the standard
1.362 + // places. Perform an exception lookup in the caller and
1.363 + // dispatch to the handler if found. Otherwise unwind and
1.364 + // dispatch to the callers exception handler.
1.365 +
1.366 + oop_maps = new OopMapSet();
1.367 + OopMap* oop_map = generate_oop_map(sasm, true);
1.368 +
1.369 + // transfer the pending exception to the exception_oop
1.370 + __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
1.371 + __ ld_ptr(Oexception, 0, G0);
1.372 + __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
1.373 + __ add(I7, frame::pc_return_offset, Oissuing_pc);
1.374 +
1.375 + generate_handle_exception(sasm, oop_maps, oop_map);
1.376 + __ should_not_reach_here();
1.377 + }
1.378 + break;
1.379 +
1.380 + case new_instance_id:
1.381 + case fast_new_instance_id:
1.382 + case fast_new_instance_init_check_id:
1.383 + {
1.384 + Register G5_klass = G5; // Incoming
1.385 + Register O0_obj = O0; // Outgoing
1.386 +
1.387 + if (id == new_instance_id) {
1.388 + __ set_info("new_instance", dont_gc_arguments);
1.389 + } else if (id == fast_new_instance_id) {
1.390 + __ set_info("fast new_instance", dont_gc_arguments);
1.391 + } else {
1.392 + assert(id == fast_new_instance_init_check_id, "bad StubID");
1.393 + __ set_info("fast new_instance init check", dont_gc_arguments);
1.394 + }
1.395 +
1.396 + if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
1.397 + UseTLAB && FastTLABRefill) {
1.398 + Label slow_path;
1.399 + Register G1_obj_size = G1;
1.400 + Register G3_t1 = G3;
1.401 + Register G4_t2 = G4;
1.402 + assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
1.403 +
1.404 + // Push a frame since we may do dtrace notification for the
1.405 + // allocation which requires calling out and we don't want
1.406 + // to stomp the real return address.
1.407 + __ save_frame(0);
1.408 +
1.409 + if (id == fast_new_instance_init_check_id) {
1.410 + // make sure the klass is initialized
1.411 + __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);
1.412 + __ cmp(G3_t1, instanceKlass::fully_initialized);
1.413 + __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
1.414 + __ delayed()->nop();
1.415 + }
1.416 +#ifdef ASSERT
1.417 + // assert object can be fast path allocated
1.418 + {
1.419 + Label ok, not_ok;
1.420 + __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);
1.421 + __ cmp(G1_obj_size, 0); // make sure it's an instance (LH > 0)
1.422 + __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);
1.423 + __ delayed()->nop();
1.424 + __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
1.425 + __ br(Assembler::zero, false, Assembler::pn, ok);
1.426 + __ delayed()->nop();
1.427 + __ bind(not_ok);
1.428 + __ stop("assert(can be fast path allocated)");
1.429 + __ should_not_reach_here();
1.430 + __ bind(ok);
1.431 + }
1.432 +#endif // ASSERT
1.433 + // if we got here then the TLAB allocation failed, so try
1.434 + // refilling the TLAB or allocating directly from eden.
1.435 + Label retry_tlab, try_eden;
1.436 + __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
1.437 +
1.438 + __ bind(retry_tlab);
1.439 +
1.440 + // get the instance size
1.441 + __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
1.442 + __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
1.443 + __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
1.444 + __ verify_oop(O0_obj);
1.445 + __ mov(O0, I0);
1.446 + __ ret();
1.447 + __ delayed()->restore();
1.448 +
1.449 + __ bind(try_eden);
1.450 + // get the instance size
1.451 + __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
1.452 + __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
1.453 + __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
1.454 + __ verify_oop(O0_obj);
1.455 + __ mov(O0, I0);
1.456 + __ ret();
1.457 + __ delayed()->restore();
1.458 +
1.459 + __ bind(slow_path);
1.460 +
1.461 + // pop this frame so generate_stub_call can push it's own
1.462 + __ restore();
1.463 + }
1.464 +
1.465 + oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
1.466 + // I0->O0: new instance
1.467 + }
1.468 +
1.469 + break;
1.470 +
1.471 +#ifdef TIERED
1.472 + case counter_overflow_id:
1.473 + // G4 contains bci
1.474 + oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4);
1.475 + break;
1.476 +#endif // TIERED
1.477 +
1.478 + case new_type_array_id:
1.479 + case new_object_array_id:
1.480 + {
1.481 + Register G5_klass = G5; // Incoming
1.482 + Register G4_length = G4; // Incoming
1.483 + Register O0_obj = O0; // Outgoing
1.484 +
1.485 + Address klass_lh(G5_klass, 0, ((klassOopDesc::header_size() * HeapWordSize)
1.486 + + Klass::layout_helper_offset_in_bytes()));
1.487 + assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
1.488 + assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
1.489 + // Use this offset to pick out an individual byte of the layout_helper:
1.490 + const int klass_lh_header_size_offset = ((BytesPerInt - 1) // 3 - 2 selects byte {0,1,0,0}
1.491 + - Klass::_lh_header_size_shift / BitsPerByte);
1.492 +
1.493 + if (id == new_type_array_id) {
1.494 + __ set_info("new_type_array", dont_gc_arguments);
1.495 + } else {
1.496 + __ set_info("new_object_array", dont_gc_arguments);
1.497 + }
1.498 +
1.499 +#ifdef ASSERT
1.500 + // assert object type is really an array of the proper kind
1.501 + {
1.502 + Label ok;
1.503 + Register G3_t1 = G3;
1.504 + __ ld(klass_lh, G3_t1);
1.505 + __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
1.506 + int tag = ((id == new_type_array_id)
1.507 + ? Klass::_lh_array_tag_type_value
1.508 + : Klass::_lh_array_tag_obj_value);
1.509 + __ cmp(G3_t1, tag);
1.510 + __ brx(Assembler::equal, false, Assembler::pt, ok);
1.511 + __ delayed()->nop();
1.512 + __ stop("assert(is an array klass)");
1.513 + __ should_not_reach_here();
1.514 + __ bind(ok);
1.515 + }
1.516 +#endif // ASSERT
1.517 +
1.518 + if (UseTLAB && FastTLABRefill) {
1.519 + Label slow_path;
1.520 + Register G1_arr_size = G1;
1.521 + Register G3_t1 = G3;
1.522 + Register O1_t2 = O1;
1.523 + assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
1.524 +
1.525 + // check that array length is small enough for fast path
1.526 + __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
1.527 + __ cmp(G4_length, G3_t1);
1.528 + __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
1.529 + __ delayed()->nop();
1.530 +
1.531 + // if we got here then the TLAB allocation failed, so try
1.532 + // refilling the TLAB or allocating directly from eden.
1.533 + Label retry_tlab, try_eden;
1.534 + __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
1.535 +
1.536 + __ bind(retry_tlab);
1.537 +
1.538 + // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
1.539 + __ ld(klass_lh, G3_t1);
1.540 + __ sll(G4_length, G3_t1, G1_arr_size);
1.541 + __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
1.542 + __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
1.543 + __ add(G1_arr_size, G3_t1, G1_arr_size);
1.544 + __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size); // align up
1.545 + __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
1.546 +
1.547 + __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path); // preserves G1_arr_size
1.548 +
1.549 + __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
1.550 + __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
1.551 + __ sub(G1_arr_size, G3_t1, O1_t2); // body length
1.552 + __ add(O0_obj, G3_t1, G3_t1); // body start
1.553 + __ initialize_body(G3_t1, O1_t2);
1.554 + __ verify_oop(O0_obj);
1.555 + __ retl();
1.556 + __ delayed()->nop();
1.557 +
1.558 + __ bind(try_eden);
1.559 + // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
1.560 + __ ld(klass_lh, G3_t1);
1.561 + __ sll(G4_length, G3_t1, G1_arr_size);
1.562 + __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
1.563 + __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
1.564 + __ add(G1_arr_size, G3_t1, G1_arr_size);
1.565 + __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
1.566 + __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
1.567 +
1.568 + __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path); // preserves G1_arr_size
1.569 +
1.570 + __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
1.571 + __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
1.572 + __ sub(G1_arr_size, G3_t1, O1_t2); // body length
1.573 + __ add(O0_obj, G3_t1, G3_t1); // body start
1.574 + __ initialize_body(G3_t1, O1_t2);
1.575 + __ verify_oop(O0_obj);
1.576 + __ retl();
1.577 + __ delayed()->nop();
1.578 +
1.579 + __ bind(slow_path);
1.580 + }
1.581 +
1.582 + if (id == new_type_array_id) {
1.583 + oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
1.584 + } else {
1.585 + oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
1.586 + }
1.587 + // I0 -> O0: new array
1.588 + }
1.589 + break;
1.590 +
1.591 + case new_multi_array_id:
1.592 + { // O0: klass
1.593 + // O1: rank
1.594 + // O2: address of 1st dimension
1.595 + __ set_info("new_multi_array", dont_gc_arguments);
1.596 + oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
1.597 + // I0 -> O0: new multi array
1.598 + }
1.599 + break;
1.600 +
1.601 + case register_finalizer_id:
1.602 + {
1.603 + __ set_info("register_finalizer", dont_gc_arguments);
1.604 +
1.605 + // load the klass and check the has finalizer flag
1.606 + Label register_finalizer;
1.607 + Register t = O1;
1.608 + __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), t);
1.609 + __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t);
1.610 + __ set(JVM_ACC_HAS_FINALIZER, G3);
1.611 + __ andcc(G3, t, G0);
1.612 + __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
1.613 + __ delayed()->nop();
1.614 +
1.615 + // do a leaf return
1.616 + __ retl();
1.617 + __ delayed()->nop();
1.618 +
1.619 + __ bind(register_finalizer);
1.620 + OopMap* oop_map = save_live_registers(sasm);
1.621 + int call_offset = __ call_RT(noreg, noreg,
1.622 + CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
1.623 + oop_maps = new OopMapSet();
1.624 + oop_maps->add_gc_map(call_offset, oop_map);
1.625 +
1.626 + // Now restore all the live registers
1.627 + restore_live_registers(sasm);
1.628 +
1.629 + __ ret();
1.630 + __ delayed()->restore();
1.631 + }
1.632 + break;
1.633 +
1.634 + case throw_range_check_failed_id:
1.635 + { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
1.636 + // G4: index
1.637 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1.638 + }
1.639 + break;
1.640 +
1.641 + case throw_index_exception_id:
1.642 + { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
1.643 + // G4: index
1.644 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1.645 + }
1.646 + break;
1.647 +
1.648 + case throw_div0_exception_id:
1.649 + { __ set_info("throw_div0_exception", dont_gc_arguments);
1.650 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
1.651 + }
1.652 + break;
1.653 +
1.654 + case throw_null_pointer_exception_id:
1.655 + { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
1.656 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
1.657 + }
1.658 + break;
1.659 +
1.660 + case handle_exception_id:
1.661 + {
1.662 + __ set_info("handle_exception", dont_gc_arguments);
1.663 + // make a frame and preserve the caller's caller-save registers
1.664 +
1.665 + oop_maps = new OopMapSet();
1.666 + OopMap* oop_map = save_live_registers(sasm);
1.667 + __ mov(Oexception->after_save(), Oexception);
1.668 + __ mov(Oissuing_pc->after_save(), Oissuing_pc);
1.669 + generate_handle_exception(sasm, oop_maps, oop_map);
1.670 + }
1.671 + break;
1.672 +
1.673 + case unwind_exception_id:
1.674 + {
1.675 + // O0: exception
1.676 + // I7: address of call to this method
1.677 +
1.678 + __ set_info("unwind_exception", dont_gc_arguments);
1.679 + __ mov(Oexception, Oexception->after_save());
1.680 + __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
1.681 +
1.682 + __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1.683 + Oissuing_pc->after_save());
1.684 + __ verify_not_null_oop(Oexception->after_save());
1.685 + __ jmp(O0, 0);
1.686 + __ delayed()->restore();
1.687 + }
1.688 + break;
1.689 +
1.690 + case throw_array_store_exception_id:
1.691 + {
1.692 + __ set_info("throw_array_store_exception", dont_gc_arguments);
1.693 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false);
1.694 + }
1.695 + break;
1.696 +
1.697 + case throw_class_cast_exception_id:
1.698 + {
1.699 + // G4: object
1.700 + __ set_info("throw_class_cast_exception", dont_gc_arguments);
1.701 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
1.702 + }
1.703 + break;
1.704 +
1.705 + case throw_incompatible_class_change_error_id:
1.706 + {
1.707 + __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
1.708 + oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
1.709 + }
1.710 + break;
1.711 +
1.712 + case slow_subtype_check_id:
1.713 + { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
1.714 + // Arguments :
1.715 + //
1.716 + // ret : G3
1.717 + // sub : G3, argument, destroyed
1.718 + // super: G1, argument, not changed
1.719 + // raddr: O7, blown by call
1.720 + Label loop, miss;
1.721 +
1.722 + __ save_frame(0); // Blow no registers!
1.723 +
1.724 + __ ld_ptr( G3, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 );
1.725 + __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0); // length in l0
1.726 + __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1); // ptr into array
1.727 + __ clr(L4); // Index
1.728 + // Load a little early; will load 1 off the end of the array.
1.729 + // Ok for now; revisit if we have other uses of this routine.
1.730 + __ ld_ptr(L1,0,L2); // Will load a little early
1.731 +
1.732 + // The scan loop
1.733 + __ bind(loop);
1.734 + __ add(L1,wordSize,L1); // Bump by OOP size
1.735 + __ cmp(L4,L0);
1.736 + __ br(Assembler::equal,false,Assembler::pn,miss);
1.737 + __ delayed()->inc(L4); // Bump index
1.738 + __ subcc(L2,G1,L3); // Check for match; zero in L3 for a hit
1.739 + __ brx( Assembler::notEqual, false, Assembler::pt, loop );
1.740 + __ delayed()->ld_ptr(L1,0,L2); // Will load a little early
1.741 +
1.742 + // Got a hit; report success; set cache
1.743 + __ st_ptr( G1, G3, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() );
1.744 +
1.745 + __ mov(1, G3);
1.746 + __ ret(); // Result in G5 is ok; flags set
1.747 + __ delayed()->restore(); // free copy or add can go here
1.748 +
1.749 + __ bind(miss);
1.750 + __ mov(0, G3);
1.751 + __ ret(); // Result in G5 is ok; flags set
1.752 + __ delayed()->restore(); // free copy or add can go here
1.753 + }
1.754 +
1.755 + case monitorenter_nofpu_id:
1.756 + case monitorenter_id:
1.757 + { // G4: object
1.758 + // G5: lock address
1.759 + __ set_info("monitorenter", dont_gc_arguments);
1.760 +
1.761 + int save_fpu_registers = (id == monitorenter_id);
1.762 + // make a frame and preserve the caller's caller-save registers
1.763 + OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
1.764 +
1.765 + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
1.766 +
1.767 + oop_maps = new OopMapSet();
1.768 + oop_maps->add_gc_map(call_offset, oop_map);
1.769 + restore_live_registers(sasm, save_fpu_registers);
1.770 +
1.771 + __ ret();
1.772 + __ delayed()->restore();
1.773 + }
1.774 + break;
1.775 +
1.776 + case monitorexit_nofpu_id:
1.777 + case monitorexit_id:
1.778 + { // G4: lock address
1.779 + // note: really a leaf routine but must setup last java sp
1.780 + // => use call_RT for now (speed can be improved by
1.781 + // doing last java sp setup manually)
1.782 + __ set_info("monitorexit", dont_gc_arguments);
1.783 +
1.784 + int save_fpu_registers = (id == monitorexit_id);
1.785 + // make a frame and preserve the caller's caller-save registers
1.786 + OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
1.787 +
1.788 + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
1.789 +
1.790 + oop_maps = new OopMapSet();
1.791 + oop_maps->add_gc_map(call_offset, oop_map);
1.792 + restore_live_registers(sasm, save_fpu_registers);
1.793 +
1.794 + __ ret();
1.795 + __ delayed()->restore();
1.796 +
1.797 + }
1.798 + break;
1.799 +
1.800 + case access_field_patching_id:
1.801 + { __ set_info("access_field_patching", dont_gc_arguments);
1.802 + oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1.803 + }
1.804 + break;
1.805 +
1.806 + case load_klass_patching_id:
1.807 + { __ set_info("load_klass_patching", dont_gc_arguments);
1.808 + oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1.809 + }
1.810 + break;
1.811 +
1.812 + case jvmti_exception_throw_id:
1.813 + { // Oexception : exception
1.814 + __ set_info("jvmti_exception_throw", dont_gc_arguments);
1.815 + oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0);
1.816 + }
1.817 + break;
1.818 +
1.819 + case dtrace_object_alloc_id:
1.820 + { // O0: object
1.821 + __ set_info("dtrace_object_alloc", dont_gc_arguments);
1.822 + // we can't gc here so skip the oopmap but make sure that all
1.823 + // the live registers get saved.
1.824 + save_live_registers(sasm);
1.825 +
1.826 + __ save_thread(L7_thread_cache);
1.827 + __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
1.828 + relocInfo::runtime_call_type);
1.829 + __ delayed()->mov(I0, O0);
1.830 + __ restore_thread(L7_thread_cache);
1.831 +
1.832 + restore_live_registers(sasm);
1.833 + __ ret();
1.834 + __ delayed()->restore();
1.835 + }
1.836 + break;
1.837 +
1.838 + default:
1.839 + { __ set_info("unimplemented entry", dont_gc_arguments);
1.840 + __ save_frame(0);
1.841 + __ set((int)id, O1);
1.842 + __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
1.843 + __ should_not_reach_here();
1.844 + }
1.845 + break;
1.846 + }
1.847 + return oop_maps;
1.848 +}
1.849 +
1.850 +
1.851 +void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) {
1.852 + Label no_deopt;
1.853 + Label no_handler;
1.854 +
1.855 + __ verify_not_null_oop(Oexception);
1.856 +
1.857 + // save the exception and issuing pc in the thread
1.858 + __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1.859 + __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1.860 +
1.861 + // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map)
1.862 + __ mov(I7, L0);
1.863 + __ mov(Oissuing_pc, I7);
1.864 + __ sub(I7, frame::pc_return_offset, I7);
1.865 + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1.866 +
1.867 + // Note: if nmethod has been deoptimized then regardless of
1.868 + // whether it had a handler or not we will deoptimize
1.869 + // by entering the deopt blob with a pending exception.
1.870 +
1.871 + __ tst(O0);
1.872 + __ br(Assembler::zero, false, Assembler::pn, no_handler);
1.873 + __ delayed()->nop();
1.874 +
1.875 + // restore the registers that were saved at the beginning and jump to the exception handler.
1.876 + restore_live_registers(sasm);
1.877 +
1.878 + __ jmp(O0, 0);
1.879 + __ delayed()->restore();
1.880 +
1.881 + __ bind(no_handler);
1.882 + __ mov(L0, I7); // restore return address
1.883 +
1.884 + // restore exception oop
1.885 + __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), Oexception->after_save());
1.886 + __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1.887 +
1.888 + __ restore();
1.889 +
1.890 + Address exc(G4, Runtime1::entry_for(Runtime1::unwind_exception_id));
1.891 + __ jump_to(exc, 0);
1.892 + __ delayed()->nop();
1.893 +
1.894 +
1.895 + oop_maps->add_gc_map(call_offset, oop_map);
1.896 +}
1.897 +
1.898 +
1.899 +#undef __
1.900 +
1.901 +#define __ masm->
