src/cpu/sparc/vm/c1_Runtime1_sparc.cpp

changeset 0
f90c822e73f8
child 6876
710a3c8b516e
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp	Wed Apr 27 01:25:04 2016 +0800
     1.3 @@ -0,0 +1,1146 @@
     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 "c1/c1_Defs.hpp"
    1.30 +#include "c1/c1_MacroAssembler.hpp"
    1.31 +#include "c1/c1_Runtime1.hpp"
    1.32 +#include "interpreter/interpreter.hpp"
    1.33 +#include "nativeInst_sparc.hpp"
    1.34 +#include "oops/compiledICHolder.hpp"
    1.35 +#include "oops/oop.inline.hpp"
    1.36 +#include "prims/jvmtiExport.hpp"
    1.37 +#include "register_sparc.hpp"
    1.38 +#include "runtime/sharedRuntime.hpp"
    1.39 +#include "runtime/signature.hpp"
    1.40 +#include "runtime/vframeArray.hpp"
    1.41 +#include "utilities/macros.hpp"
    1.42 +#include "vmreg_sparc.inline.hpp"
    1.43 +#if INCLUDE_ALL_GCS
    1.44 +#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
    1.45 +#endif
    1.46 +
    1.47 +// Implementation of StubAssembler
    1.48 +
    1.49 +int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry_point, int number_of_arguments) {
    1.50 +  // for sparc changing the number of arguments doesn't change
    1.51 +  // anything about the frame size so we'll always lie and claim that
    1.52 +  // we are only passing 1 argument.
    1.53 +  set_num_rt_args(1);
    1.54 +
    1.55 +  assert_not_delayed();
    1.56 +  // bang stack before going to runtime
    1.57 +  set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
    1.58 +  st(G0, SP, G3_scratch);
    1.59 +
    1.60 +  // debugging support
    1.61 +  assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
    1.62 +
    1.63 +  set_last_Java_frame(SP, noreg);
    1.64 +  if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early
    1.65 +  save_thread(L7_thread_cache);
    1.66 +  // do the call
    1.67 +  call(entry_point, relocInfo::runtime_call_type);
    1.68 +  if (!VerifyThread) {
    1.69 +    delayed()->mov(G2_thread, O0);  // pass thread as first argument
    1.70 +  } else {
    1.71 +    delayed()->nop();             // (thread already passed)
    1.72 +  }
    1.73 +  int call_offset = offset();  // offset of return address
    1.74 +  restore_thread(L7_thread_cache);
    1.75 +  reset_last_Java_frame();
    1.76 +
    1.77 +  // check for pending exceptions
    1.78 +  { Label L;
    1.79 +    Address exception_addr(G2_thread, Thread::pending_exception_offset());
    1.80 +    ld_ptr(exception_addr, Gtemp);
    1.81 +    br_null_short(Gtemp, pt, L);
    1.82 +    Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
    1.83 +    st_ptr(G0, vm_result_addr);
    1.84 +    Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
    1.85 +    st_ptr(G0, vm_result_addr_2);
    1.86 +
    1.87 +    if (frame_size() == no_frame_size) {
    1.88 +      // we use O7 linkage so that forward_exception_entry has the issuing PC
    1.89 +      call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
    1.90 +      delayed()->restore();
    1.91 +    } else if (_stub_id == Runtime1::forward_exception_id) {
    1.92 +      should_not_reach_here();
    1.93 +    } else {
    1.94 +      AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));
    1.95 +      jump_to(exc, G4);
    1.96 +      delayed()->nop();
    1.97 +    }
    1.98 +    bind(L);
    1.99 +  }
   1.100 +
   1.101 +  // get oop result if there is one and reset the value in the thread
   1.102 +  if (oop_result1->is_valid()) {                    // get oop result if there is one and reset it in the thread
   1.103 +    get_vm_result  (oop_result1);
   1.104 +  } else {
   1.105 +    // be a little paranoid and clear the result
   1.106 +    Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
   1.107 +    st_ptr(G0, vm_result_addr);
   1.108 +  }
   1.109 +
   1.110 +  // get second result if there is one and reset the value in the thread
   1.111 +  if (metadata_result->is_valid()) {
   1.112 +    get_vm_result_2  (metadata_result);
   1.113 +  } else {
   1.114 +    // be a little paranoid and clear the result
   1.115 +    Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
   1.116 +    st_ptr(G0, vm_result_addr_2);
   1.117 +  }
   1.118 +
   1.119 +  return call_offset;
   1.120 +}
   1.121 +
   1.122 +
   1.123 +int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
   1.124 +  // O0 is reserved for the thread
   1.125 +  mov(arg1, O1);
   1.126 +  return call_RT(oop_result1, metadata_result, entry, 1);
   1.127 +}
   1.128 +
   1.129 +
   1.130 +int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
   1.131 +  // O0 is reserved for the thread
   1.132 +  mov(arg1, O1);
   1.133 +  mov(arg2, O2); assert(arg2 != O1, "smashed argument");
   1.134 +  return call_RT(oop_result1, metadata_result, entry, 2);
   1.135 +}
   1.136 +
   1.137 +
   1.138 +int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
   1.139 +  // O0 is reserved for the thread
   1.140 +  mov(arg1, O1);
   1.141 +  mov(arg2, O2); assert(arg2 != O1,               "smashed argument");
   1.142 +  mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
   1.143 +  return call_RT(oop_result1, metadata_result, entry, 3);
   1.144 +}
   1.145 +
   1.146 +
   1.147 +// Implementation of Runtime1
   1.148 +
   1.149 +#define __ sasm->
   1.150 +
   1.151 +static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
   1.152 +static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
   1.153 +static int reg_save_size_in_words;
   1.154 +static int frame_size_in_bytes = -1;
   1.155 +
   1.156 +static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
   1.157 +  assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
   1.158 +         "mismatch in calculation");
   1.159 +  sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
   1.160 +  int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
   1.161 +  OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
   1.162 +
   1.163 +  int i;
   1.164 +  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   1.165 +    Register r = as_Register(i);
   1.166 +    if (r == G1 || r == G3 || r == G4 || r == G5) {
   1.167 +      int sp_offset = cpu_reg_save_offsets[i];
   1.168 +      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
   1.169 +                                r->as_VMReg());
   1.170 +    }
   1.171 +  }
   1.172 +
   1.173 +  if (save_fpu_registers) {
   1.174 +    for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   1.175 +      FloatRegister r = as_FloatRegister(i);
   1.176 +      int sp_offset = fpu_reg_save_offsets[i];
   1.177 +      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
   1.178 +                                r->as_VMReg());
   1.179 +    }
   1.180 +  }
   1.181 +  return oop_map;
   1.182 +}
   1.183 +
   1.184 +static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
   1.185 +  assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
   1.186 +         "mismatch in calculation");
   1.187 +  __ save_frame_c1(frame_size_in_bytes);
   1.188 +
   1.189 +  // Record volatile registers as callee-save values in an OopMap so their save locations will be
   1.190 +  // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
   1.191 +  // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers
   1.192 +  // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
   1.193 +  // (as the stub's I's) when the runtime routine called by the stub creates its frame.
   1.194 +  // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
   1.195 +
   1.196 +  int i;
   1.197 +  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   1.198 +    Register r = as_Register(i);
   1.199 +    if (r == G1 || r == G3 || r == G4 || r == G5) {
   1.200 +      int sp_offset = cpu_reg_save_offsets[i];
   1.201 +      __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
   1.202 +    }
   1.203 +  }
   1.204 +
   1.205 +  if (save_fpu_registers) {
   1.206 +    for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   1.207 +      FloatRegister r = as_FloatRegister(i);
   1.208 +      int sp_offset = fpu_reg_save_offsets[i];
   1.209 +      __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
   1.210 +    }
   1.211 +  }
   1.212 +
   1.213 +  return generate_oop_map(sasm, save_fpu_registers);
   1.214 +}
   1.215 +
   1.216 +static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
   1.217 +  for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
   1.218 +    Register r = as_Register(i);
   1.219 +    if (r == G1 || r == G3 || r == G4 || r == G5) {
   1.220 +      __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
   1.221 +    }
   1.222 +  }
   1.223 +
   1.224 +  if (restore_fpu_registers) {
   1.225 +    for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
   1.226 +      FloatRegister r = as_FloatRegister(i);
   1.227 +      __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
   1.228 +    }
   1.229 +  }
   1.230 +}
   1.231 +
   1.232 +
   1.233 +void Runtime1::initialize_pd() {
   1.234 +  // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
   1.235 +  //
   1.236 +  // A stub routine will have a frame that is at least large enough to hold
   1.237 +  // a register window save area (obviously) and the volatile g registers
   1.238 +  // and floating registers. A user of save_live_registers can have a frame
   1.239 +  // that has more scratch area in it (although typically they will use L-regs).
   1.240 +  // in that case the frame will look like this (stack growing down)
   1.241 +  //
   1.242 +  // FP -> |             |
   1.243 +  //       | scratch mem |
   1.244 +  //       |   "      "  |
   1.245 +  //       --------------
   1.246 +  //       | float regs  |
   1.247 +  //       |   "    "    |
   1.248 +  //       ---------------
   1.249 +  //       | G regs      |
   1.250 +  //       | "  "        |
   1.251 +  //       ---------------
   1.252 +  //       | abi reg.    |
   1.253 +  //       | window save |
   1.254 +  //       | area        |
   1.255 +  // SP -> ---------------
   1.256 +  //
   1.257 +  int i;
   1.258 +  int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned
   1.259 +
   1.260 +  // only G int registers are saved explicitly; others are found in register windows
   1.261 +  for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   1.262 +    Register r = as_Register(i);
   1.263 +    if (r == G1 || r == G3 || r == G4 || r == G5) {
   1.264 +      cpu_reg_save_offsets[i] = sp_offset;
   1.265 +      sp_offset++;
   1.266 +    }
   1.267 +  }
   1.268 +
   1.269 +  // all float registers are saved explicitly
   1.270 +  assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
   1.271 +  for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   1.272 +    fpu_reg_save_offsets[i] = sp_offset;
   1.273 +    sp_offset++;
   1.274 +  }
   1.275 +  reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
   1.276 +  // this should match assembler::total_frame_size_in_bytes, which
   1.277 +  // isn't callable from this context.  It's checked by an assert when
   1.278 +  // it's used though.
   1.279 +  frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
   1.280 +}
   1.281 +
   1.282 +
   1.283 +OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
   1.284 +  // make a frame and preserve the caller's caller-save registers
   1.285 +  OopMap* oop_map = save_live_registers(sasm);
   1.286 +  int call_offset;
   1.287 +  if (!has_argument) {
   1.288 +    call_offset = __ call_RT(noreg, noreg, target);
   1.289 +  } else {
   1.290 +    call_offset = __ call_RT(noreg, noreg, target, G4);
   1.291 +  }
   1.292 +  OopMapSet* oop_maps = new OopMapSet();
   1.293 +  oop_maps->add_gc_map(call_offset, oop_map);
   1.294 +
   1.295 +  __ should_not_reach_here();
   1.296 +  return oop_maps;
   1.297 +}
   1.298 +
   1.299 +
   1.300 +OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
   1.301 +                                        Register arg1, Register arg2, Register arg3) {
   1.302 +  // make a frame and preserve the caller's caller-save registers
   1.303 +  OopMap* oop_map = save_live_registers(sasm);
   1.304 +
   1.305 +  int call_offset;
   1.306 +  if (arg1 == noreg) {
   1.307 +    call_offset = __ call_RT(result, noreg, target);
   1.308 +  } else if (arg2 == noreg) {
   1.309 +    call_offset = __ call_RT(result, noreg, target, arg1);
   1.310 +  } else if (arg3 == noreg) {
   1.311 +    call_offset = __ call_RT(result, noreg, target, arg1, arg2);
   1.312 +  } else {
   1.313 +    call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
   1.314 +  }
   1.315 +  OopMapSet* oop_maps = NULL;
   1.316 +
   1.317 +  oop_maps = new OopMapSet();
   1.318 +  oop_maps->add_gc_map(call_offset, oop_map);
   1.319 +  restore_live_registers(sasm);
   1.320 +
   1.321 +  __ ret();
   1.322 +  __ delayed()->restore();
   1.323 +
   1.324 +  return oop_maps;
   1.325 +}
   1.326 +
   1.327 +
   1.328 +OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
   1.329 +  // make a frame and preserve the caller's caller-save registers
   1.330 +  OopMap* oop_map = save_live_registers(sasm);
   1.331 +
   1.332 +  // call the runtime patching routine, returns non-zero if nmethod got deopted.
   1.333 +  int call_offset = __ call_RT(noreg, noreg, target);
   1.334 +  OopMapSet* oop_maps = new OopMapSet();
   1.335 +  oop_maps->add_gc_map(call_offset, oop_map);
   1.336 +
   1.337 +  // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
   1.338 +  // deoptimization handler entry that will cause re-execution of the current bytecode
   1.339 +  DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
   1.340 +  assert(deopt_blob != NULL, "deoptimization blob must have been created");
   1.341 +
   1.342 +  Label no_deopt;
   1.343 +  __ br_null_short(O0, Assembler::pt, no_deopt);
   1.344 +
   1.345 +  // return to the deoptimization handler entry for unpacking and rexecute
   1.346 +  // if we simply returned the we'd deopt as if any call we patched had just
   1.347 +  // returned.
   1.348 +
   1.349 +  restore_live_registers(sasm);
   1.350 +
   1.351 +  AddressLiteral dest(deopt_blob->unpack_with_reexecution());
   1.352 +  __ jump_to(dest, O0);
   1.353 +  __ delayed()->restore();
   1.354 +
   1.355 +  __ bind(no_deopt);
   1.356 +  restore_live_registers(sasm);
   1.357 +  __ ret();
   1.358 +  __ delayed()->restore();
   1.359 +
   1.360 +  return oop_maps;
   1.361 +}
   1.362 +
   1.363 +OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
   1.364 +
   1.365 +  OopMapSet* oop_maps = NULL;
   1.366 +  // for better readability
   1.367 +  const bool must_gc_arguments = true;
   1.368 +  const bool dont_gc_arguments = false;
   1.369 +
   1.370 +  // stub code & info for the different stubs
   1.371 +  switch (id) {
   1.372 +    case forward_exception_id:
   1.373 +      {
   1.374 +        oop_maps = generate_handle_exception(id, sasm);
   1.375 +      }
   1.376 +      break;
   1.377 +
   1.378 +    case new_instance_id:
   1.379 +    case fast_new_instance_id:
   1.380 +    case fast_new_instance_init_check_id:
   1.381 +      {
   1.382 +        Register G5_klass = G5; // Incoming
   1.383 +        Register O0_obj   = O0; // Outgoing
   1.384 +
   1.385 +        if (id == new_instance_id) {
   1.386 +          __ set_info("new_instance", dont_gc_arguments);
   1.387 +        } else if (id == fast_new_instance_id) {
   1.388 +          __ set_info("fast new_instance", dont_gc_arguments);
   1.389 +        } else {
   1.390 +          assert(id == fast_new_instance_init_check_id, "bad StubID");
   1.391 +          __ set_info("fast new_instance init check", dont_gc_arguments);
   1.392 +        }
   1.393 +
   1.394 +        if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
   1.395 +            UseTLAB && FastTLABRefill) {
   1.396 +          Label slow_path;
   1.397 +          Register G1_obj_size = G1;
   1.398 +          Register G3_t1 = G3;
   1.399 +          Register G4_t2 = G4;
   1.400 +          assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
   1.401 +
   1.402 +          // Push a frame since we may do dtrace notification for the
   1.403 +          // allocation which requires calling out and we don't want
   1.404 +          // to stomp the real return address.
   1.405 +          __ save_frame(0);
   1.406 +
   1.407 +          if (id == fast_new_instance_init_check_id) {
   1.408 +            // make sure the klass is initialized
   1.409 +            __ ldub(G5_klass, in_bytes(InstanceKlass::init_state_offset()), G3_t1);
   1.410 +            __ cmp(G3_t1, InstanceKlass::fully_initialized);
   1.411 +            __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
   1.412 +            __ delayed()->nop();
   1.413 +          }
   1.414 +#ifdef ASSERT
   1.415 +          // assert object can be fast path allocated
   1.416 +          {
   1.417 +            Label ok, not_ok;
   1.418 +          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   1.419 +          // make sure it's an instance (LH > 0)
   1.420 +          __ cmp_and_br_short(G1_obj_size, 0, Assembler::lessEqual, Assembler::pn, not_ok);
   1.421 +          __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
   1.422 +          __ br(Assembler::zero, false, Assembler::pn, ok);
   1.423 +          __ delayed()->nop();
   1.424 +          __ bind(not_ok);
   1.425 +          __ stop("assert(can be fast path allocated)");
   1.426 +          __ should_not_reach_here();
   1.427 +          __ bind(ok);
   1.428 +          }
   1.429 +#endif // ASSERT
   1.430 +          // if we got here then the TLAB allocation failed, so try
   1.431 +          // refilling the TLAB or allocating directly from eden.
   1.432 +          Label retry_tlab, try_eden;
   1.433 +          __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
   1.434 +
   1.435 +          __ bind(retry_tlab);
   1.436 +
   1.437 +          // get the instance size
   1.438 +          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   1.439 +
   1.440 +          __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
   1.441 +
   1.442 +          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
   1.443 +          __ verify_oop(O0_obj);
   1.444 +          __ mov(O0, I0);
   1.445 +          __ ret();
   1.446 +          __ delayed()->restore();
   1.447 +
   1.448 +          __ bind(try_eden);
   1.449 +          // get the instance size
   1.450 +          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   1.451 +          __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
   1.452 +          __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
   1.453 +
   1.454 +          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
   1.455 +          __ verify_oop(O0_obj);
   1.456 +          __ mov(O0, I0);
   1.457 +          __ ret();
   1.458 +          __ delayed()->restore();
   1.459 +
   1.460 +          __ bind(slow_path);
   1.461 +
   1.462 +          // pop this frame so generate_stub_call can push it's own
   1.463 +          __ restore();
   1.464 +        }
   1.465 +
   1.466 +        oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
   1.467 +        // I0->O0: new instance
   1.468 +      }
   1.469 +
   1.470 +      break;
   1.471 +
   1.472 +    case counter_overflow_id:
   1.473 +        // G4 contains bci, G5 contains method
   1.474 +      oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
   1.475 +      break;
   1.476 +
   1.477 +    case new_type_array_id:
   1.478 +    case new_object_array_id:
   1.479 +      {
   1.480 +        Register G5_klass = G5; // Incoming
   1.481 +        Register G4_length = G4; // Incoming
   1.482 +        Register O0_obj   = O0; // Outgoing
   1.483 +
   1.484 +        Address klass_lh(G5_klass, Klass::layout_helper_offset());
   1.485 +        assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
   1.486 +        assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
   1.487 +        // Use this offset to pick out an individual byte of the layout_helper:
   1.488 +        const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
   1.489 +                                                 - Klass::_lh_header_size_shift / BitsPerByte);
   1.490 +
   1.491 +        if (id == new_type_array_id) {
   1.492 +          __ set_info("new_type_array", dont_gc_arguments);
   1.493 +        } else {
   1.494 +          __ set_info("new_object_array", dont_gc_arguments);
   1.495 +        }
   1.496 +
   1.497 +#ifdef ASSERT
   1.498 +        // assert object type is really an array of the proper kind
   1.499 +        {
   1.500 +          Label ok;
   1.501 +          Register G3_t1 = G3;
   1.502 +          __ ld(klass_lh, G3_t1);
   1.503 +          __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
   1.504 +          int tag = ((id == new_type_array_id)
   1.505 +                     ? Klass::_lh_array_tag_type_value
   1.506 +                     : Klass::_lh_array_tag_obj_value);
   1.507 +          __ cmp_and_brx_short(G3_t1, tag, Assembler::equal, Assembler::pt, ok);
   1.508 +          __ stop("assert(is an array klass)");
   1.509 +          __ should_not_reach_here();
   1.510 +          __ bind(ok);
   1.511 +        }
   1.512 +#endif // ASSERT
   1.513 +
   1.514 +        if (UseTLAB && FastTLABRefill) {
   1.515 +          Label slow_path;
   1.516 +          Register G1_arr_size = G1;
   1.517 +          Register G3_t1 = G3;
   1.518 +          Register O1_t2 = O1;
   1.519 +          assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
   1.520 +
   1.521 +          // check that array length is small enough for fast path
   1.522 +          __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
   1.523 +          __ cmp(G4_length, G3_t1);
   1.524 +          __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
   1.525 +          __ delayed()->nop();
   1.526 +
   1.527 +          // if we got here then the TLAB allocation failed, so try
   1.528 +          // refilling the TLAB or allocating directly from eden.
   1.529 +          Label retry_tlab, try_eden;
   1.530 +          __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
   1.531 +
   1.532 +          __ bind(retry_tlab);
   1.533 +
   1.534 +          // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
   1.535 +          __ ld(klass_lh, G3_t1);
   1.536 +          __ sll(G4_length, G3_t1, G1_arr_size);
   1.537 +          __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
   1.538 +          __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
   1.539 +          __ add(G1_arr_size, G3_t1, G1_arr_size);
   1.540 +          __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
   1.541 +          __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
   1.542 +
   1.543 +          __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
   1.544 +
   1.545 +          __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
   1.546 +          __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
   1.547 +          __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
   1.548 +          __ add(O0_obj, G3_t1, G3_t1);       // body start
   1.549 +          __ initialize_body(G3_t1, O1_t2);
   1.550 +          __ verify_oop(O0_obj);
   1.551 +          __ retl();
   1.552 +          __ delayed()->nop();
   1.553 +
   1.554 +          __ bind(try_eden);
   1.555 +          // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
   1.556 +          __ ld(klass_lh, G3_t1);
   1.557 +          __ sll(G4_length, G3_t1, G1_arr_size);
   1.558 +          __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
   1.559 +          __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
   1.560 +          __ add(G1_arr_size, G3_t1, G1_arr_size);
   1.561 +          __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
   1.562 +          __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
   1.563 +
   1.564 +          __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
   1.565 +          __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
   1.566 +
   1.567 +          __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
   1.568 +          __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
   1.569 +          __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
   1.570 +          __ add(O0_obj, G3_t1, G3_t1);       // body start
   1.571 +          __ initialize_body(G3_t1, O1_t2);
   1.572 +          __ verify_oop(O0_obj);
   1.573 +          __ retl();
   1.574 +          __ delayed()->nop();
   1.575 +
   1.576 +          __ bind(slow_path);
   1.577 +        }
   1.578 +
   1.579 +        if (id == new_type_array_id) {
   1.580 +          oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
   1.581 +        } else {
   1.582 +          oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
   1.583 +        }
   1.584 +        // I0 -> O0: new array
   1.585 +      }
   1.586 +      break;
   1.587 +
   1.588 +    case new_multi_array_id:
   1.589 +      { // O0: klass
   1.590 +        // O1: rank
   1.591 +        // O2: address of 1st dimension
   1.592 +        __ set_info("new_multi_array", dont_gc_arguments);
   1.593 +        oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
   1.594 +        // I0 -> O0: new multi array
   1.595 +      }
   1.596 +      break;
   1.597 +
   1.598 +    case register_finalizer_id:
   1.599 +      {
   1.600 +        __ set_info("register_finalizer", dont_gc_arguments);
   1.601 +
   1.602 +        // load the klass and check the has finalizer flag
   1.603 +        Label register_finalizer;
   1.604 +        Register t = O1;
   1.605 +        __ load_klass(O0, t);
   1.606 +        __ ld(t, in_bytes(Klass::access_flags_offset()), t);
   1.607 +        __ set(JVM_ACC_HAS_FINALIZER, G3);
   1.608 +        __ andcc(G3, t, G0);
   1.609 +        __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
   1.610 +        __ delayed()->nop();
   1.611 +
   1.612 +        // do a leaf return
   1.613 +        __ retl();
   1.614 +        __ delayed()->nop();
   1.615 +
   1.616 +        __ bind(register_finalizer);
   1.617 +        OopMap* oop_map = save_live_registers(sasm);
   1.618 +        int call_offset = __ call_RT(noreg, noreg,
   1.619 +                                     CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
   1.620 +        oop_maps = new OopMapSet();
   1.621 +        oop_maps->add_gc_map(call_offset, oop_map);
   1.622 +
   1.623 +        // Now restore all the live registers
   1.624 +        restore_live_registers(sasm);
   1.625 +
   1.626 +        __ ret();
   1.627 +        __ delayed()->restore();
   1.628 +      }
   1.629 +      break;
   1.630 +
   1.631 +    case throw_range_check_failed_id:
   1.632 +      { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
   1.633 +        // G4: index
   1.634 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
   1.635 +      }
   1.636 +      break;
   1.637 +
   1.638 +    case throw_index_exception_id:
   1.639 +      { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
   1.640 +        // G4: index
   1.641 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
   1.642 +      }
   1.643 +      break;
   1.644 +
   1.645 +    case throw_div0_exception_id:
   1.646 +      { __ set_info("throw_div0_exception", dont_gc_arguments);
   1.647 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
   1.648 +      }
   1.649 +      break;
   1.650 +
   1.651 +    case throw_null_pointer_exception_id:
   1.652 +      { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
   1.653 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
   1.654 +      }
   1.655 +      break;
   1.656 +
   1.657 +    case handle_exception_id:
   1.658 +      { __ set_info("handle_exception", dont_gc_arguments);
   1.659 +        oop_maps = generate_handle_exception(id, sasm);
   1.660 +      }
   1.661 +      break;
   1.662 +
   1.663 +    case handle_exception_from_callee_id:
   1.664 +      { __ set_info("handle_exception_from_callee", dont_gc_arguments);
   1.665 +        oop_maps = generate_handle_exception(id, sasm);
   1.666 +      }
   1.667 +      break;
   1.668 +
   1.669 +    case unwind_exception_id:
   1.670 +      {
   1.671 +        // O0: exception
   1.672 +        // I7: address of call to this method
   1.673 +
   1.674 +        __ set_info("unwind_exception", dont_gc_arguments);
   1.675 +        __ mov(Oexception, Oexception->after_save());
   1.676 +        __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
   1.677 +
   1.678 +        __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
   1.679 +                        G2_thread, Oissuing_pc->after_save());
   1.680 +        __ verify_not_null_oop(Oexception->after_save());
   1.681 +
   1.682 +        // Restore SP from L7 if the exception PC is a method handle call site.
   1.683 +        __ mov(O0, G5);  // Save the target address.
   1.684 +        __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
   1.685 +        __ tst(L0);  // Condition codes are preserved over the restore.
   1.686 +        __ restore();
   1.687 +
   1.688 +        __ jmp(G5, 0);
   1.689 +        __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
   1.690 +      }
   1.691 +      break;
   1.692 +
   1.693 +    case throw_array_store_exception_id:
   1.694 +      {
   1.695 +        __ set_info("throw_array_store_exception", dont_gc_arguments);
   1.696 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
   1.697 +      }
   1.698 +      break;
   1.699 +
   1.700 +    case throw_class_cast_exception_id:
   1.701 +      {
   1.702 +        // G4: object
   1.703 +        __ set_info("throw_class_cast_exception", dont_gc_arguments);
   1.704 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
   1.705 +      }
   1.706 +      break;
   1.707 +
   1.708 +    case throw_incompatible_class_change_error_id:
   1.709 +      {
   1.710 +        __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
   1.711 +        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
   1.712 +      }
   1.713 +      break;
   1.714 +
   1.715 +    case slow_subtype_check_id:
   1.716 +      { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
   1.717 +        // Arguments :
   1.718 +        //
   1.719 +        //      ret  : G3
   1.720 +        //      sub  : G3, argument, destroyed
   1.721 +        //      super: G1, argument, not changed
   1.722 +        //      raddr: O7, blown by call
   1.723 +        Label miss;
   1.724 +
   1.725 +        __ save_frame(0);               // Blow no registers!
   1.726 +
   1.727 +        __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
   1.728 +
   1.729 +        __ mov(1, G3);
   1.730 +        __ ret();                       // Result in G5 is 'true'
   1.731 +        __ delayed()->restore();        // free copy or add can go here
   1.732 +
   1.733 +        __ bind(miss);
   1.734 +        __ mov(0, G3);
   1.735 +        __ ret();                       // Result in G5 is 'false'
   1.736 +        __ delayed()->restore();        // free copy or add can go here
   1.737 +      }
   1.738 +
   1.739 +    case monitorenter_nofpu_id:
   1.740 +    case monitorenter_id:
   1.741 +      { // G4: object
   1.742 +        // G5: lock address
   1.743 +        __ set_info("monitorenter", dont_gc_arguments);
   1.744 +
   1.745 +        int save_fpu_registers = (id == monitorenter_id);
   1.746 +        // make a frame and preserve the caller's caller-save registers
   1.747 +        OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
   1.748 +
   1.749 +        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
   1.750 +
   1.751 +        oop_maps = new OopMapSet();
   1.752 +        oop_maps->add_gc_map(call_offset, oop_map);
   1.753 +        restore_live_registers(sasm, save_fpu_registers);
   1.754 +
   1.755 +        __ ret();
   1.756 +        __ delayed()->restore();
   1.757 +      }
   1.758 +      break;
   1.759 +
   1.760 +    case monitorexit_nofpu_id:
   1.761 +    case monitorexit_id:
   1.762 +      { // G4: lock address
   1.763 +        // note: really a leaf routine but must setup last java sp
   1.764 +        //       => use call_RT for now (speed can be improved by
   1.765 +        //       doing last java sp setup manually)
   1.766 +        __ set_info("monitorexit", dont_gc_arguments);
   1.767 +
   1.768 +        int save_fpu_registers = (id == monitorexit_id);
   1.769 +        // make a frame and preserve the caller's caller-save registers
   1.770 +        OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
   1.771 +
   1.772 +        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
   1.773 +
   1.774 +        oop_maps = new OopMapSet();
   1.775 +        oop_maps->add_gc_map(call_offset, oop_map);
   1.776 +        restore_live_registers(sasm, save_fpu_registers);
   1.777 +
   1.778 +        __ ret();
   1.779 +        __ delayed()->restore();
   1.780 +      }
   1.781 +      break;
   1.782 +
   1.783 +    case deoptimize_id:
   1.784 +      {
   1.785 +        __ set_info("deoptimize", dont_gc_arguments);
   1.786 +        OopMap* oop_map = save_live_registers(sasm);
   1.787 +        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
   1.788 +        oop_maps = new OopMapSet();
   1.789 +        oop_maps->add_gc_map(call_offset, oop_map);
   1.790 +        restore_live_registers(sasm);
   1.791 +        DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
   1.792 +        assert(deopt_blob != NULL, "deoptimization blob must have been created");
   1.793 +        AddressLiteral dest(deopt_blob->unpack_with_reexecution());
   1.794 +        __ jump_to(dest, O0);
   1.795 +        __ delayed()->restore();
   1.796 +      }
   1.797 +      break;
   1.798 +
   1.799 +    case access_field_patching_id:
   1.800 +      { __ set_info("access_field_patching", dont_gc_arguments);
   1.801 +        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
   1.802 +      }
   1.803 +      break;
   1.804 +
   1.805 +    case load_klass_patching_id:
   1.806 +      { __ set_info("load_klass_patching", dont_gc_arguments);
   1.807 +        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
   1.808 +      }
   1.809 +      break;
   1.810 +
   1.811 +    case load_mirror_patching_id:
   1.812 +      { __ set_info("load_mirror_patching", dont_gc_arguments);
   1.813 +        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
   1.814 +      }
   1.815 +      break;
   1.816 +
   1.817 +    case load_appendix_patching_id:
   1.818 +      { __ set_info("load_appendix_patching", dont_gc_arguments);
   1.819 +        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
   1.820 +      }
   1.821 +      break;
   1.822 +
   1.823 +    case dtrace_object_alloc_id:
   1.824 +      { // O0: object
   1.825 +        __ set_info("dtrace_object_alloc", dont_gc_arguments);
   1.826 +        // we can't gc here so skip the oopmap but make sure that all
   1.827 +        // the live registers get saved.
   1.828 +        save_live_registers(sasm);
   1.829 +
   1.830 +        __ save_thread(L7_thread_cache);
   1.831 +        __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
   1.832 +                relocInfo::runtime_call_type);
   1.833 +        __ delayed()->mov(I0, O0);
   1.834 +        __ restore_thread(L7_thread_cache);
   1.835 +
   1.836 +        restore_live_registers(sasm);
   1.837 +        __ ret();
   1.838 +        __ delayed()->restore();
   1.839 +      }
   1.840 +      break;
   1.841 +
   1.842 +#if INCLUDE_ALL_GCS
   1.843 +    case g1_pre_barrier_slow_id:
   1.844 +      { // G4: previous value of memory
   1.845 +        BarrierSet* bs = Universe::heap()->barrier_set();
   1.846 +        if (bs->kind() != BarrierSet::G1SATBCTLogging) {
   1.847 +          __ save_frame(0);
   1.848 +          __ set((int)id, O1);
   1.849 +          __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
   1.850 +          __ should_not_reach_here();
   1.851 +          break;
   1.852 +        }
   1.853 +
   1.854 +        __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
   1.855 +
   1.856 +        Register pre_val = G4;
   1.857 +        Register tmp  = G1_scratch;
   1.858 +        Register tmp2 = G3_scratch;
   1.859 +
   1.860 +        Label refill, restart;
   1.861 +        bool with_frame = false; // I don't know if we can do with-frame.
   1.862 +        int satb_q_index_byte_offset =
   1.863 +          in_bytes(JavaThread::satb_mark_queue_offset() +
   1.864 +                   PtrQueue::byte_offset_of_index());
   1.865 +        int satb_q_buf_byte_offset =
   1.866 +          in_bytes(JavaThread::satb_mark_queue_offset() +
   1.867 +                   PtrQueue::byte_offset_of_buf());
   1.868 +
   1.869 +        __ bind(restart);
   1.870 +        // Load the index into the SATB buffer. PtrQueue::_index is a
   1.871 +        // size_t so ld_ptr is appropriate
   1.872 +        __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
   1.873 +
   1.874 +        // index == 0?
   1.875 +        __ cmp_and_brx_short(tmp, G0, Assembler::equal, Assembler::pn, refill);
   1.876 +
   1.877 +        __ ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
   1.878 +        __ sub(tmp, oopSize, tmp);
   1.879 +
   1.880 +        __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
   1.881 +        // Use return-from-leaf
   1.882 +        __ retl();
   1.883 +        __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
   1.884 +
   1.885 +        __ bind(refill);
   1.886 +        __ save_frame(0);
   1.887 +
   1.888 +        __ mov(pre_val, L0);
   1.889 +        __ mov(tmp,     L1);
   1.890 +        __ mov(tmp2,    L2);
   1.891 +
   1.892 +        __ call_VM_leaf(L7_thread_cache,
   1.893 +                        CAST_FROM_FN_PTR(address,
   1.894 +                                         SATBMarkQueueSet::handle_zero_index_for_thread),
   1.895 +                                         G2_thread);
   1.896 +
   1.897 +        __ mov(L0, pre_val);
   1.898 +        __ mov(L1, tmp);
   1.899 +        __ mov(L2, tmp2);
   1.900 +
   1.901 +        __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
   1.902 +        __ delayed()->restore();
   1.903 +      }
   1.904 +      break;
   1.905 +
   1.906 +    case g1_post_barrier_slow_id:
   1.907 +      {
   1.908 +        BarrierSet* bs = Universe::heap()->barrier_set();
   1.909 +        if (bs->kind() != BarrierSet::G1SATBCTLogging) {
   1.910 +          __ save_frame(0);
   1.911 +          __ set((int)id, O1);
   1.912 +          __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
   1.913 +          __ should_not_reach_here();
   1.914 +          break;
   1.915 +        }
   1.916 +
   1.917 +        __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
   1.918 +
   1.919 +        Register addr = G4;
   1.920 +        Register cardtable = G5;
   1.921 +        Register tmp  = G1_scratch;
   1.922 +        Register tmp2 = G3_scratch;
   1.923 +        jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
   1.924 +
   1.925 +        Label not_already_dirty, restart, refill, young_card;
   1.926 +
   1.927 +#ifdef _LP64
   1.928 +        __ srlx(addr, CardTableModRefBS::card_shift, addr);
   1.929 +#else
   1.930 +        __ srl(addr, CardTableModRefBS::card_shift, addr);
   1.931 +#endif
   1.932 +
   1.933 +        AddressLiteral rs(byte_map_base);
   1.934 +        __ set(rs, cardtable);         // cardtable := <card table base>
   1.935 +        __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
   1.936 +
   1.937 +        __ cmp_and_br_short(tmp, G1SATBCardTableModRefBS::g1_young_card_val(), Assembler::equal, Assembler::pt, young_card);
   1.938 +
   1.939 +        __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
   1.940 +        __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
   1.941 +
   1.942 +        assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
   1.943 +        __ cmp_and_br_short(tmp, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);
   1.944 +
   1.945 +        __ bind(young_card);
   1.946 +        // We didn't take the branch, so we're already dirty: return.
   1.947 +        // Use return-from-leaf
   1.948 +        __ retl();
   1.949 +        __ delayed()->nop();
   1.950 +
   1.951 +        // Not dirty.
   1.952 +        __ bind(not_already_dirty);
   1.953 +
   1.954 +        // Get cardtable + tmp into a reg by itself
   1.955 +        __ add(addr, cardtable, tmp2);
   1.956 +
   1.957 +        // First, dirty it.
   1.958 +        __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
   1.959 +
   1.960 +        Register tmp3 = cardtable;
   1.961 +        Register tmp4 = tmp;
   1.962 +
   1.963 +        // these registers are now dead
   1.964 +        addr = cardtable = tmp = noreg;
   1.965 +
   1.966 +        int dirty_card_q_index_byte_offset =
   1.967 +          in_bytes(JavaThread::dirty_card_queue_offset() +
   1.968 +                   PtrQueue::byte_offset_of_index());
   1.969 +        int dirty_card_q_buf_byte_offset =
   1.970 +          in_bytes(JavaThread::dirty_card_queue_offset() +
   1.971 +                   PtrQueue::byte_offset_of_buf());
   1.972 +
   1.973 +        __ bind(restart);
   1.974 +
   1.975 +        // Get the index into the update buffer. PtrQueue::_index is
   1.976 +        // a size_t so ld_ptr is appropriate here.
   1.977 +        __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
   1.978 +
   1.979 +        // index == 0?
   1.980 +        __ cmp_and_brx_short(tmp3, G0, Assembler::equal,  Assembler::pn, refill);
   1.981 +
   1.982 +        __ ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
   1.983 +        __ sub(tmp3, oopSize, tmp3);
   1.984 +
   1.985 +        __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
   1.986 +        // Use return-from-leaf
   1.987 +        __ retl();
   1.988 +        __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
   1.989 +
   1.990 +        __ bind(refill);
   1.991 +        __ save_frame(0);
   1.992 +
   1.993 +        __ mov(tmp2, L0);
   1.994 +        __ mov(tmp3, L1);
   1.995 +        __ mov(tmp4, L2);
   1.996 +
   1.997 +        __ call_VM_leaf(L7_thread_cache,
   1.998 +                        CAST_FROM_FN_PTR(address,
   1.999 +                                         DirtyCardQueueSet::handle_zero_index_for_thread),
  1.1000 +                                         G2_thread);
  1.1001 +
  1.1002 +        __ mov(L0, tmp2);
  1.1003 +        __ mov(L1, tmp3);
  1.1004 +        __ mov(L2, tmp4);
  1.1005 +
  1.1006 +        __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
  1.1007 +        __ delayed()->restore();
  1.1008 +      }
  1.1009 +      break;
  1.1010 +#endif // INCLUDE_ALL_GCS
  1.1011 +
  1.1012 +    case predicate_failed_trap_id:
  1.1013 +      {
  1.1014 +        __ set_info("predicate_failed_trap", dont_gc_arguments);
  1.1015 +        OopMap* oop_map = save_live_registers(sasm);
  1.1016 +
  1.1017 +        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
  1.1018 +
  1.1019 +        oop_maps = new OopMapSet();
  1.1020 +        oop_maps->add_gc_map(call_offset, oop_map);
  1.1021 +
  1.1022 +        DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
  1.1023 +        assert(deopt_blob != NULL, "deoptimization blob must have been created");
  1.1024 +        restore_live_registers(sasm);
  1.1025 +
  1.1026 +        AddressLiteral dest(deopt_blob->unpack_with_reexecution());
  1.1027 +        __ jump_to(dest, O0);
  1.1028 +        __ delayed()->restore();
  1.1029 +      }
  1.1030 +      break;
  1.1031 +
  1.1032 +    default:
  1.1033 +      { __ set_info("unimplemented entry", dont_gc_arguments);
  1.1034 +        __ save_frame(0);
  1.1035 +        __ set((int)id, O1);
  1.1036 +        __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
  1.1037 +        __ should_not_reach_here();
  1.1038 +      }
  1.1039 +      break;
  1.1040 +  }
  1.1041 +  return oop_maps;
  1.1042 +}
  1.1043 +
  1.1044 +
  1.1045 +OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler* sasm) {
  1.1046 +  __ block_comment("generate_handle_exception");
  1.1047 +
  1.1048 +  // Save registers, if required.
  1.1049 +  OopMapSet* oop_maps = new OopMapSet();
  1.1050 +  OopMap* oop_map = NULL;
  1.1051 +  switch (id) {
  1.1052 +  case forward_exception_id:
  1.1053 +    // We're handling an exception in the context of a compiled frame.
  1.1054 +    // The registers have been saved in the standard places.  Perform
  1.1055 +    // an exception lookup in the caller and dispatch to the handler
  1.1056 +    // if found.  Otherwise unwind and dispatch to the callers
  1.1057 +    // exception handler.
  1.1058 +     oop_map = generate_oop_map(sasm, true);
  1.1059 +
  1.1060 +     // transfer the pending exception to the exception_oop
  1.1061 +     __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
  1.1062 +     __ ld_ptr(Oexception, 0, G0);
  1.1063 +     __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
  1.1064 +     __ add(I7, frame::pc_return_offset, Oissuing_pc);
  1.1065 +    break;
  1.1066 +  case handle_exception_id:
  1.1067 +    // At this point all registers MAY be live.
  1.1068 +    oop_map = save_live_registers(sasm);
  1.1069 +    __ mov(Oexception->after_save(),  Oexception);
  1.1070 +    __ mov(Oissuing_pc->after_save(), Oissuing_pc);
  1.1071 +    break;
  1.1072 +  case handle_exception_from_callee_id:
  1.1073 +    // At this point all registers except exception oop (Oexception)
  1.1074 +    // and exception pc (Oissuing_pc) are dead.
  1.1075 +    oop_map = new OopMap(frame_size_in_bytes / sizeof(jint), 0);
  1.1076 +    sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
  1.1077 +    __ save_frame_c1(frame_size_in_bytes);
  1.1078 +    __ mov(Oexception->after_save(),  Oexception);
  1.1079 +    __ mov(Oissuing_pc->after_save(), Oissuing_pc);
  1.1080 +    break;
  1.1081 +  default:  ShouldNotReachHere();
  1.1082 +  }
  1.1083 +
  1.1084 +  __ verify_not_null_oop(Oexception);
  1.1085 +
  1.1086 +#ifdef ASSERT
  1.1087 +  // check that fields in JavaThread for exception oop and issuing pc are
  1.1088 +  // empty before writing to them
  1.1089 +  Label oop_empty;
  1.1090 +  Register scratch = I7;  // We can use I7 here because it's overwritten later anyway.
  1.1091 +  __ ld_ptr(Address(G2_thread, JavaThread::exception_oop_offset()), scratch);
  1.1092 +  __ br_null(scratch, false, Assembler::pt, oop_empty);
  1.1093 +  __ delayed()->nop();
  1.1094 +  __ stop("exception oop already set");
  1.1095 +  __ bind(oop_empty);
  1.1096 +
  1.1097 +  Label pc_empty;
  1.1098 +  __ ld_ptr(Address(G2_thread, JavaThread::exception_pc_offset()), scratch);
  1.1099 +  __ br_null(scratch, false, Assembler::pt, pc_empty);
  1.1100 +  __ delayed()->nop();
  1.1101 +  __ stop("exception pc already set");
  1.1102 +  __ bind(pc_empty);
  1.1103 +#endif
  1.1104 +
  1.1105 +  // save the exception and issuing pc in the thread
  1.1106 +  __ st_ptr(Oexception,  G2_thread, in_bytes(JavaThread::exception_oop_offset()));
  1.1107 +  __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
  1.1108 +
  1.1109 +  // use the throwing pc as the return address to lookup (has bci & oop map)
  1.1110 +  __ mov(Oissuing_pc, I7);
  1.1111 +  __ sub(I7, frame::pc_return_offset, I7);
  1.1112 +  int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
  1.1113 +  oop_maps->add_gc_map(call_offset, oop_map);
  1.1114 +
  1.1115 +  // Note: if nmethod has been deoptimized then regardless of
  1.1116 +  // whether it had a handler or not we will deoptimize
  1.1117 +  // by entering the deopt blob with a pending exception.
  1.1118 +
  1.1119 +  // Restore the registers that were saved at the beginning, remove
  1.1120 +  // the frame and jump to the exception handler.
  1.1121 +  switch (id) {
  1.1122 +  case forward_exception_id:
  1.1123 +  case handle_exception_id:
  1.1124 +    restore_live_registers(sasm);
  1.1125 +    __ jmp(O0, 0);
  1.1126 +    __ delayed()->restore();
  1.1127 +    break;
  1.1128 +  case handle_exception_from_callee_id:
  1.1129 +    // Restore SP from L7 if the exception PC is a method handle call site.
  1.1130 +    __ mov(O0, G5);  // Save the target address.
  1.1131 +    __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
  1.1132 +    __ tst(L0);  // Condition codes are preserved over the restore.
  1.1133 +    __ restore();
  1.1134 +
  1.1135 +    __ jmp(G5, 0);  // jump to the exception handler
  1.1136 +    __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
  1.1137 +    break;
  1.1138 +  default:  ShouldNotReachHere();
  1.1139 +  }
  1.1140 +
  1.1141 +  return oop_maps;
  1.1142 +}
  1.1143 +
  1.1144 +
  1.1145 +#undef __
  1.1146 +
  1.1147 +const char *Runtime1::pd_name_for_address(address entry) {
  1.1148 +  return "<unknown function>";
  1.1149 +}

mercurial