aoqi@1: /*
aoqi@1:  * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
huangxuguang@9251:  * Copyright (c) 2015, 2018, Loongson Technology. All rights reserved.
aoqi@1:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
aoqi@1:  *
aoqi@1:  * This code is free software; you can redistribute it and/or modify it
aoqi@1:  * under the terms of the GNU General Public License version 2 only, as
aoqi@1:  * published by the Free Software Foundation.
aoqi@1:  *
aoqi@1:  * This code is distributed in the hope that it will be useful, but WITHOUT
aoqi@1:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
aoqi@1:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
aoqi@1:  * version 2 for more details (a copy is included in the LICENSE file that
aoqi@1:  * accompanied this code).
aoqi@1:  *
aoqi@1:  * You should have received a copy of the GNU General Public License version
aoqi@1:  * 2 along with this work; if not, write to the Free Software Foundation,
aoqi@1:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
aoqi@1:  *
aoqi@1:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
aoqi@1:  * or visit www.oracle.com if you need additional information or have any
aoqi@1:  * questions.
aoqi@1:  *
aoqi@1:  */
aoqi@1: 
aoqi@1: #include "precompiled.hpp"
aoqi@1: #include "asm/assembler.hpp"
aoqi@1: #include "c1/c1_Defs.hpp"
aoqi@1: #include "c1/c1_MacroAssembler.hpp"
aoqi@1: #include "c1/c1_Runtime1.hpp"
aoqi@1: #include "interpreter/interpreter.hpp"
aoqi@1: #include "nativeInst_mips.hpp"
aoqi@1: #include "oops/compiledICHolder.hpp"
aoqi@1: #include "oops/oop.inline.hpp"
aoqi@1: #include "prims/jvmtiExport.hpp"
aoqi@1: #include "register_mips.hpp"
aoqi@1: #include "runtime/sharedRuntime.hpp"
aoqi@1: #include "runtime/signature.hpp"
aoqi@1: #include "runtime/vframeArray.hpp"
aoqi@6880: #include "utilities/macros.hpp"
aoqi@1: #include "vmreg_mips.inline.hpp"
aoqi@6880: #if INCLUDE_ALL_GCS
aoqi@6880: #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
aoqi@6880: #endif
aoqi@1: 
aoqi@1: 
aoqi@1: // Implementation of StubAssembler
aoqi@1: // this method will preserve the stack space for arguments as indicated by args_size
aoqi@1: // for stack alignment consideration, you cannot call this with argument in stack.
aoqi@1: // if you need >3 arguments, you must implement this method yourself.
fujie@9153: int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, int args_size) {
aoqi@6880:   // i use S7 for edi.
aoqi@6880:   // setup registers
aoqi@6880:   const Register thread = TREG; // is callee-saved register (Visual C++ calling conventions)
fujie@9153:   assert(!(oop_result1->is_valid() || metadata_result->is_valid()) || oop_result1 != metadata_result,                            "registers must be different");
fujie@9153:   assert(oop_result1 != thread && metadata_result != thread, "registers must be different");
aoqi@6880:   assert(args_size >= 0, "illegal args_size");
fujie@9153:   bool align_stack = false;
fujie@9153: #ifdef _LP64
fujie@9153:   // At a method handle call, the stack may not be properly aligned
fujie@9153:   // when returning with an exception.
fujie@9153:   align_stack = (stub_id() == Runtime1::handle_exception_from_callee_id);
fujie@9153: #endif
aoqi@1: 
aoqi@6880:   set_num_rt_args(1 + args_size);
aoqi@1: 
aoqi@1: 
aoqi@6880:   // push java thread (becomes first argument of C function)
aoqi@6880:   get_thread(thread);
aoqi@6880:   move(A0, thread);
aoqi@1: 
fujie@9153:   if(!align_stack) {
fujie@9153:     set_last_Java_frame(thread, NOREG, FP, NULL);
fujie@9153:   } else {
fujie@9153:     address the_pc = pc();
fujie@9153:     set_last_Java_frame(thread, NOREG, FP, the_pc);
fujie@9153:     move(AT, -(StackAlignmentInBytes));
fujie@9153:     andr(SP, SP, AT);
fujie@9153:   }
aoqi@1: 
aoqi@6880:   relocate(relocInfo::internal_pc_type);
aoqi@6880:   {
aoqi@1: #ifndef _LP64
aoqi@6880:     int save_pc = (int)pc() +  12 + NativeCall::return_address_offset;
aoqi@6880:     lui(AT, Assembler::split_high(save_pc));
aoqi@6880:     addiu(AT, AT, Assembler::split_low(save_pc));
aoqi@1: #else
aoqi@6880:     uintptr_t save_pc = (uintptr_t)pc() + NativeMovConstReg::instruction_size + 1 * BytesPerInstWord + NativeCall::return_address_offset_long;
aoqi@6880:     li48(AT, save_pc);
aoqi@1: #endif
aoqi@6880:   }
aoqi@6880:   st_ptr(AT, thread, in_bytes(JavaThread::last_Java_pc_offset()));
aoqi@1: 
aoqi@6880:   // do the call
aoqi@6880: #ifndef _LP64
aoqi@6880:   lui(T9, Assembler::split_high((int)entry));
aoqi@6880:   addiu(T9, T9, Assembler::split_low((int)entry));
aoqi@6880: #else
aoqi@6880:   li48(T9, (intptr_t)entry);
aoqi@1: #endif
aoqi@6880:   jalr(T9);
aoqi@6880:   delayed()->nop();
fujie@9161: 
fujie@9161:   int call_offset = offset();
aoqi@6880: 
aoqi@6880:   // verify callee-saved register
aoqi@6880: #ifdef ASSERT
aoqi@6880:   guarantee(thread != V0, "change this code");
aoqi@6880:   push(V0);
aoqi@6880:   {
aoqi@6880:     Label L;
aoqi@6880:     get_thread(V0);
aoqi@6880:     beq(thread, V0, L);
aoqi@6880:     delayed()->nop();
aoqi@6880:     int3();
aoqi@6880:     stop("StubAssembler::call_RT: edi not callee saved?");
aoqi@6880:     bind(L);
aoqi@6880:   }
aoqi@6880:   super_pop(V0);
aoqi@1: #endif
aoqi@6880:   // discard thread and arguments
aoqi@9228:   ld_ptr(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
fujie@9171:   reset_last_Java_frame(thread, true);
aoqi@6880:   // check for pending exceptions
aoqi@6880:   {
aoqi@6880:     Label L;
aoqi@6880:     ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset()));
aoqi@6880:     beq(AT, R0, L);
aoqi@6880:     delayed()->nop();
aoqi@6880:     // exception pending => remove activation and forward to exception handler
aoqi@6880:     // make sure that the vm_results are cleared
aoqi@6880:     if (oop_result1->is_valid()) {
aoqi@6880:       st_ptr(R0, thread, in_bytes(JavaThread::vm_result_offset()));
aoqi@6880:     }
fujie@9153:     if (metadata_result->is_valid()) {
aoqi@6880:       st_ptr(R0, thread, in_bytes(JavaThread::vm_result_2_offset()));
aoqi@6880:     }
aoqi@6880:     // the leave() in x86 just pops ebp and remains the return address on the top
aoqi@6880:     // of stack
aoqi@6880:     // the return address will be needed by forward_exception_entry()
aoqi@6880:     if (frame_size() == no_frame_size) {
aoqi@6880:       addiu(SP, FP, wordSize);
aoqi@6880:       ld_ptr(FP, SP, (-1) * wordSize);
aoqi@6880:       jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
aoqi@6880:       delayed()->nop();
aoqi@6880:     } else if (_stub_id == Runtime1::forward_exception_id) {
aoqi@6880:       should_not_reach_here();
aoqi@6880:     } else {
aoqi@8865:       jmp(Runtime1::entry_for(Runtime1::forward_exception_id), relocInfo::runtime_call_type);
aoqi@6880:       delayed()->nop();
aoqi@6880:     }
aoqi@6880:     bind(L);
aoqi@6880:   }
aoqi@6880:   // get oop results if there are any and reset the values in the thread
aoqi@6880:   if (oop_result1->is_valid()) {
aoqi@6880:     ld_ptr(oop_result1, thread, in_bytes(JavaThread::vm_result_offset()));
aoqi@6880:     st_ptr(R0, thread, in_bytes(JavaThread::vm_result_offset()));
aoqi@6880:     verify_oop(oop_result1);
aoqi@6880:   }
fujie@9153:   if (metadata_result->is_valid()) {
fujie@9153:     ld_ptr(metadata_result, thread, in_bytes(JavaThread::vm_result_2_offset()));
aoqi@6880:     st_ptr(R0, thread, in_bytes(JavaThread::vm_result_2_offset()));
fujie@9153:     verify_oop(metadata_result);
aoqi@6880:   }
aoqi@6880:   return call_offset;
aoqi@1: }
aoqi@1: 
aoqi@1: 
fujie@9153: int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
aoqi@6880:   if (arg1 != A1) move(A1, arg1);
fujie@9153:   return call_RT(oop_result1, metadata_result, entry, 1);
aoqi@1: }
aoqi@1: 
aoqi@1: 
fujie@9153: int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
aoqi@6880:   if (arg1!=A1) move(A1, arg1);
aoqi@6880:   if (arg2!=A2) move(A2, arg2); assert(arg2 != A1, "smashed argument");
fujie@9153:   return call_RT(oop_result1, metadata_result, entry, 2);
aoqi@1: }
aoqi@1: 
aoqi@1: 
fujie@9153: int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
aoqi@6880:   if (arg1!=A1) move(A1, arg1);
aoqi@6880:   if (arg2!=A2) move(A2, arg2); assert(arg2 != A1, "smashed argument");
aoqi@6880:   if (arg3!=A3) move(A3, arg3); assert(arg3 != A1 && arg3 != A2, "smashed argument");
fujie@9153:   return call_RT(oop_result1, metadata_result, entry, 3);
aoqi@1: }
aoqi@1: 
aoqi@1: 
aoqi@1: // Implementation of StubFrame
aoqi@1: 
aoqi@1: class StubFrame: public StackObj {
aoqi@6880:  private:
aoqi@6880:   StubAssembler* _sasm;
aoqi@1: 
aoqi@6880:  public:
aoqi@6880:   StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments);
aoqi@6880:   void load_argument(int offset_in_words, Register reg);
aoqi@6880: 
aoqi@6880:   ~StubFrame();
aoqi@1: };
aoqi@1: 
aoqi@1: 
aoqi@1: #define __ _sasm->
aoqi@1: 
aoqi@1: StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments) {
aoqi@6880:   _sasm = sasm;
aoqi@6880:   __ set_info(name, must_gc_arguments);
aoqi@6880:   __ enter();
aoqi@1: }
aoqi@1: 
aoqi@1: 
aoqi@1: //FIXME, I have no idea the frame architecture of mips
aoqi@1: // load parameters that were stored with LIR_Assembler::store_parameter
aoqi@6880: // Note: offsets for store_parameter and load_argument must match
aoqi@1: void StubFrame::load_argument(int offset_in_words, Register reg) {
aoqi@6880:   //ebp + 0: link
aoqi@6880:   //    + 1: return address
aoqi@6880:   //    + 2: argument with offset 0
aoqi@6880:   //    + 3: argument with offset 1
aoqi@6880:   //    + 4: ...
aoqi@6880:   __ ld_ptr(reg, Address(FP, (offset_in_words + 2) * BytesPerWord));
aoqi@1: }
aoqi@6880: 
aoqi@6880: 
aoqi@1: StubFrame::~StubFrame() {
aoqi@6880:   __ leave();
aoqi@6880:   __ jr(RA);
aoqi@6880:   __ delayed()->nop();
aoqi@1: }
aoqi@1: 
aoqi@1: #undef __
aoqi@1: 
aoqi@1: 
aoqi@1: // Implementation of Runtime1
aoqi@1: 
aoqi@1: #define __ sasm->
aoqi@1: 
aoqi@1: //static OopMap* save_live_registers(MacroAssembler* sasm, int num_rt_args);
aoqi@1: //static void restore_live_registers(MacroAssembler* sasm);
aoqi@1: //DeoptimizationBlob* SharedRuntime::_deopt_blob = NULL;
aoqi@1: /*
aoqi@1: const int fpu_stack_as_doubles_size_in_words = 16;
aoqi@1: const int fpu_stack_as_doubles_size = 64;
aoqi@1: */
aoqi@1: const int float_regs_as_doubles_size_in_words = 16;
aoqi@1: 
aoqi@6880: //FIXME,
aoqi@1: // Stack layout for saving/restoring  all the registers needed during a runtime
aoqi@1: // call (this includes deoptimization)
aoqi@1: // Note: note that users of this frame may well have arguments to some runtime
aoqi@1: // while these values are on the stack. These positions neglect those arguments
aoqi@1: // but the code in save_live_registers will take the argument count into
aoqi@1: // account.
aoqi@1: //
aoqi@1: #ifdef _LP64
aoqi@1:   #define SLOT2(x) x,
aoqi@1:   #define SLOT_PER_WORD 2
aoqi@1: #else
aoqi@1:   #define SLOT2(x)
aoqi@1:   #define SLOT_PER_WORD 1
aoqi@1: #endif // _LP64
aoqi@1: 
aoqi@1: enum reg_save_layout {
aoqi@1: #ifndef _LP64
aoqi@1:   T0_off = 0,
aoqi@1:   S0_off = T0_off + SLOT_PER_WORD * 8,
aoqi@1: #else
aoqi@1:   A4_off = 0,
aoqi@1:   S0_off = A4_off + SLOT_PER_WORD * 8,
aoqi@1: #endif
aoqi@1:   FP_off = S0_off + SLOT_PER_WORD * 8, SLOT2(FPH_off)
aoqi@1:   T8_off, SLOT2(T8H_off)
aoqi@1:   T9_off, SLOT2(T9H_off)
aoqi@1:   SP_off, SLOT2(SPH_off)
aoqi@1:   V0_off, SLOT2(V0H_off)
aoqi@1:   V1_off, SLOT2(V1H_off)
aoqi@1:   A0_off, SLOT2(A0H_off)
aoqi@1:   A1_off, SLOT2(A1H_off)
aoqi@1:   A2_off, SLOT2(A2H_off)
aoqi@1:   A3_off, SLOT2(A3H_off)
aoqi@1: 
aoqi@1:   // Float registers
aoqi@1:   /* FIXME: Jin: In MIPS64, F0~23 are all caller-saved registers */
aoqi@1:   F0_off, SLOT2( F0H_off)
aoqi@1:   F1_off, SLOT2( F1H_off)
aoqi@1:   F2_off, SLOT2( F2H_off)
aoqi@1:   F3_off, SLOT2( F3H_off)
aoqi@1:   F4_off, SLOT2( F4H_off)
aoqi@1:   F5_off, SLOT2( F5H_off)
aoqi@1:   F6_off, SLOT2( F6H_off)
aoqi@1:   F7_off, SLOT2( F7H_off)
aoqi@1:   F8_off, SLOT2( F8H_off)
aoqi@1:   F9_off, SLOT2( F9H_off)
aoqi@1:   F10_off, SLOT2( F10H_off)
aoqi@1:   F11_off, SLOT2( F11H_off)
aoqi@1:   F12_off, SLOT2( F12H_off)
aoqi@1:   F13_off, SLOT2( F13H_off)
aoqi@1:   F14_off, SLOT2( F14H_off)
aoqi@1:   F15_off, SLOT2( F15H_off)
aoqi@1:   F16_off, SLOT2( F16H_off)
aoqi@1:   F17_off, SLOT2( F17H_off)
aoqi@1:   F18_off, SLOT2( F18H_off)
aoqi@1:   F19_off, SLOT2( F19H_off)
aoqi@1: 
aoqi@1:   GP_off, SLOT2( GPH_off)
aoqi@1:   //temp_2_off,
aoqi@1:   temp_1_off, SLOT2(temp_1H_off)
aoqi@1:   saved_fp_off, SLOT2(saved_fpH_off)
aoqi@1:   return_off, SLOT2(returnH_off)
aoqi@1: 
aoqi@1:   reg_save_frame_size,
aoqi@1: 
aoqi@1:   // illegal instruction handler
aoqi@1:   continue_dest_off = temp_1_off,
aoqi@1: 
aoqi@1:   // deoptimization equates
aoqi@1:   //deopt_type = temp_2_off,             // slot for type of deopt in progress
aoqi@1:   ret_type = temp_1_off                // slot for return type
aoqi@1: };
aoqi@1: 
aoqi@6880: 
aoqi@6880: 
aoqi@1: // Save off registers which might be killed by calls into the runtime.
aoqi@1: // Tries to smart of about FP registers.  In particular we separate
aoqi@1: // saving and describing the FPU registers for deoptimization since we
aoqi@1: // have to save the FPU registers twice if we describe them and on P4
aoqi@1: // saving FPU registers which don't contain anything appears
aoqi@1: // expensive.  The deopt blob is the only thing which needs to
aoqi@1: // describe FPU registers.  In all other cases it should be sufficient
aoqi@1: // to simply save their current value.
aoqi@1: static OopMap* generate_oop_map(StubAssembler* sasm, int num_rt_args,
aoqi@6880:                                 bool save_fpu_registers = true, bool describe_fpu_registers = false) {
aoqi@1: 
aoqi@8865:   LP64_ONLY(num_rt_args = 0);
aoqi@8865:   LP64_ONLY(assert((reg_save_frame_size * VMRegImpl::stack_slot_size) % 16 == 0, "must be 16 byte aligned");)
aoqi@8865:   int frame_size_in_slots = reg_save_frame_size + num_rt_args * wordSize / VMRegImpl::slots_per_word;   // args + thread
aoqi@8865:   sasm->set_frame_size(frame_size_in_slots / VMRegImpl::slots_per_word);
aoqi@1: 
aoqi@1:   // record saved value locations in an OopMap
aoqi@6880:   // locations are offsets from sp after runtime call; num_rt_args is number of arguments
aoqi@1:   // in call, including thread
aoqi@1:   OopMap* map = new OopMap(reg_save_frame_size, 0);
aoqi@6880: 
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(V0_off + num_rt_args), V0->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(V1_off + num_rt_args), V1->as_VMReg());
aoqi@1: #ifdef _LP64
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(V0H_off + num_rt_args), V0->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(V1H_off + num_rt_args), V1->as_VMReg()->next());
aoqi@1: #endif
aoqi@1: 
aoqi@1:   int i = 0;
aoqi@1: #ifndef _LP64
aoqi@1:   for (Register r = T0; r != T7->successor(); r = r->successor() ) {
aoqi@1:     map->set_callee_saved(VMRegImpl::stack2reg(T0_off + num_rt_args + i++), r->as_VMReg());
aoqi@1:   }
aoqi@1: #else
aoqi@1:   for (Register r = A4; r != T3->successor(); r = r->successor() ) {
aoqi@1:     map->set_callee_saved(VMRegImpl::stack2reg(A4_off + num_rt_args + i++), r->as_VMReg());
aoqi@1:     map->set_callee_saved(VMRegImpl::stack2reg(A4_off + num_rt_args + i++), r->as_VMReg()->next());
aoqi@1:   }
aoqi@1: #endif
aoqi@1: 
aoqi@1:   i = 0;
aoqi@1:   for (Register r = S0; r != S7->successor(); r = r->successor() ) {
aoqi@1:     map->set_callee_saved(VMRegImpl::stack2reg(S0_off + num_rt_args + i++), r->as_VMReg());
aoqi@1: #ifdef _LP64
aoqi@1:     map->set_callee_saved(VMRegImpl::stack2reg(S0_off + num_rt_args + i++), r->as_VMReg()->next());
aoqi@1: #endif
aoqi@1:   }
aoqi@1: 
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(FP_off + num_rt_args), FP->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(GP_off + num_rt_args), GP->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(T8_off + num_rt_args), T8->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(T9_off + num_rt_args), T9->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A0_off + num_rt_args), A0->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A1_off + num_rt_args), A1->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A2_off + num_rt_args), A2->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A3_off + num_rt_args), A3->as_VMReg());
aoqi@1: 
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F0_off + num_rt_args), F0->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F1_off + num_rt_args), F1->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F2_off + num_rt_args), F2->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F3_off + num_rt_args), F1->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F4_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F5_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F6_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F7_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F8_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F9_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F10_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F11_off + num_rt_args), F4->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F12_off + num_rt_args), F12->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F13_off + num_rt_args), F13->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F14_off + num_rt_args), F14->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F15_off + num_rt_args), F15->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F16_off + num_rt_args), F16->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F17_off + num_rt_args), F17->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F18_off + num_rt_args), F18->as_VMReg());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(F19_off + num_rt_args), F19->as_VMReg());
aoqi@1: 
aoqi@1: #ifdef _LP64
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(FPH_off + num_rt_args), FP->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(GPH_off + num_rt_args), GP->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(T8H_off + num_rt_args), T8->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(T9H_off + num_rt_args), T9->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A0H_off + num_rt_args), A0->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A1H_off + num_rt_args), A1->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A2H_off + num_rt_args), A2->as_VMReg()->next());
aoqi@1:   map->set_callee_saved(VMRegImpl::stack2reg(A3H_off + num_rt_args), A3->as_VMReg()->next());
aoqi@1: #endif
aoqi@1:   return map;
aoqi@1: }
aoqi@1: 
aoqi@8865: static OopMap* save_live_registers(StubAssembler* sasm, int num_rt_args,
aoqi@8865:                                    bool save_fpu_registers = true,
aoqi@1:                                    bool describe_fpu_registers = false) {
aoqi@1:   //const int reg_save_frame_size = return_off + 1 + num_rt_args;
aoqi@1:   __ block_comment("save_live_registers");
aoqi@1: 
aoqi@6880:   // save all register state - int, fpu
aoqi@1:   __ addi(SP, SP, -(reg_save_frame_size / SLOT_PER_WORD - 2)* wordSize);
aoqi@6880: 
aoqi@1: #ifndef _LP64
aoqi@1:   for (Register r = T0; r != T7->successor(); r = r->successor() ) {
aoqi@1:     __ sw(r, SP, (r->encoding() - T0->encoding() + T0_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #else
aoqi@1:   for (Register r = A4; r != T3->successor(); r = r->successor() ) {
aoqi@1:     __ sd(r, SP, (r->encoding() - A4->encoding() + A4_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #endif
aoqi@1:   }
aoqi@1:   for (Register r = S0; r != S7->successor(); r = r->successor() ) {
aoqi@1:     __ st_ptr(r, SP, (r->encoding() - S0->encoding() + S0_off / SLOT_PER_WORD) * wordSize);
aoqi@1:   }
aoqi@1:   __ st_ptr(FP, SP, FP_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(GP, SP, GP_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(T8, SP, T8_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(T9, SP, T9_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(A0, SP, A0_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(A1, SP, A1_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(A2, SP, A2_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(A3, SP, A3_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ st_ptr(V0, SP, V0_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ st_ptr(V1, SP, V1_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@6880:   __ sdc1(F0, SP, F0_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F1, SP, F1_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F2, SP, F2_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F3, SP, F3_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F4, SP, F4_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F5, SP, F5_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F6, SP, F6_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F7, SP, F7_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F8, SP, F8_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F9, SP, F9_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F10, SP, F10_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F11, SP, F11_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F12, SP, F12_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F13, SP, F13_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F14, SP, F14_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F15, SP, F15_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F16, SP, F16_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F17, SP, F17_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F18, SP, F18_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ sdc1(F19, SP, F19_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   return generate_oop_map(sasm, num_rt_args, save_fpu_registers, describe_fpu_registers);
aoqi@1: }
aoqi@1: 
aoqi@1: static void restore_fpu(StubAssembler* sasm, bool restore_fpu_registers = true) {
aoqi@1:   //static void restore_live_registers(MacroAssembler* sasm) {
aoqi@1: #ifndef _LP64
aoqi@1:   for (Register r = T0; r != T7->successor(); r = r->successor() ) {
aoqi@1:     __ lw(r, SP, (r->encoding() - T0->encoding() + T0_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #else
aoqi@1:   for (Register r = A4; r != T3->successor(); r = r->successor() ) {
aoqi@1:     __ ld(r, SP, (r->encoding() - A4->encoding() + A4_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #endif
aoqi@1:   }
aoqi@1:   for (Register r = S0; r != S7->successor(); r = r->successor() ) {
aoqi@1:     __ ld_ptr(r, SP, (r->encoding() - S0->encoding() + S0_off / SLOT_PER_WORD) * wordSize);
aoqi@1:   }
aoqi@1:   __ ld_ptr(FP, SP, FP_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(GP, SP, GP_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ ld_ptr(T8, SP, T8_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(T9, SP, T9_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A0, SP, A0_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A1, SP, A1_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A2, SP, A2_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A3, SP, A3_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ ld_ptr(V0, SP, V0_off * wordSize / SLOT_PER_WORD);
aoqi@6880:   __ ld_ptr(V1, SP, V1_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ ldc1(F0, SP, F0_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F1, SP, F1_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F2, SP, F2_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F3, SP, F3_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F4, SP, F4_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F5, SP, F5_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F6, SP, F6_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F7, SP, F7_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F8, SP, F8_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F9, SP, F9_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F10, SP, F10_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F11, SP, F11_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F12, SP, F12_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F13, SP, F13_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F14, SP, F14_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F15, SP, F15_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F16, SP, F16_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F17, SP, F17_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F18, SP, F18_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F19, SP, F19_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ addiu(SP, SP, (reg_save_frame_size / SLOT_PER_WORD - 2) * wordSize);
aoqi@1: }
aoqi@1: 
aoqi@1: static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
aoqi@1:   __ block_comment("restore_live_registers");
aoqi@1:   restore_fpu(sasm, restore_fpu_registers);
aoqi@1: }
aoqi@1: 
aoqi@6880: static void restore_live_registers_except_V0(StubAssembler* sasm, bool restore_fpu_registers = true) {
aoqi@1:   //static void restore_live_registers(MacroAssembler* sasm) {
aoqi@1:   //FIXME , maybe V1 need to be saved too
aoqi@1:   __ block_comment("restore_live_registers except V0");
aoqi@1: #ifndef _LP64
aoqi@1:   for (Register r = T0; r != T7->successor(); r = r->successor() ) {
aoqi@6880:     __ lw(r, SP, (r->encoding() - T0->encoding() + T0_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #else
aoqi@1:   for (Register r = A4; r != T3->successor(); r = r->successor() ) {
aoqi@6880:     __ ld(r, SP, (r->encoding() - A4->encoding() + A4_off / SLOT_PER_WORD) * wordSize);
aoqi@1: #endif
aoqi@1:   }
aoqi@1:   for (Register r = S0; r != S7->successor(); r = r->successor() ) {
aoqi@6880:     __ ld_ptr(r, SP, (r->encoding() - S0->encoding() + S0_off / SLOT_PER_WORD) * wordSize);
aoqi@1:   }
aoqi@1:   __ ld_ptr(FP, SP, FP_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(GP, SP, GP_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ ld_ptr(T8, SP, T8_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(T9, SP, T9_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A0, SP, A0_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A1, SP, A1_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A2, SP, A2_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ld_ptr(A3, SP, A3_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1: #if 1
aoqi@1:   __ ldc1(F0, SP, F0_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F1, SP, F1_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F2, SP, F2_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F3, SP, F3_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F4, SP, F4_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F5, SP, F5_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F6, SP, F6_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F7, SP, F7_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F8, SP, F8_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F9, SP, F9_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F10, SP, F10_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F11, SP, F11_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F12, SP, F12_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F13, SP, F13_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F14, SP, F14_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F15, SP, F15_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F16, SP, F16_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F17, SP, F17_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F18, SP, F18_off * wordSize / SLOT_PER_WORD);
aoqi@1:   __ ldc1(F19, SP, F19_off * wordSize / SLOT_PER_WORD);
aoqi@1: #endif
aoqi@1: 
aoqi@6880:   __ ld_ptr(V1, SP, V1_off * wordSize / SLOT_PER_WORD);
aoqi@1: 
aoqi@1:   __ addiu(SP, SP, (reg_save_frame_size / SLOT_PER_WORD - 2) * wordSize);
aoqi@1: }
aoqi@1: 
aoqi@1: void Runtime1::initialize_pd() {
aoqi@1:   // nothing to do
aoqi@1: }
aoqi@1: 
aoqi@1: // target: the entry point of the method that creates and posts the exception oop
aoqi@1: // has_argument: true if the exception needs an argument (passed on stack because registers must be preserved)
aoqi@1: OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
aoqi@6880:   // preserve all registers
aoqi@6880:   OopMap* oop_map = save_live_registers(sasm, 0);
aoqi@1: 
aoqi@6880:   // now all registers are saved and can be used freely
aoqi@6880:   // verify that no old value is used accidentally
aoqi@6880:   //all reigster are saved , I think mips do not need this
aoqi@1: 
aoqi@6880:   // registers used by this stub
aoqi@6880:   const Register temp_reg = T3;
aoqi@6880:   // load argument for exception that is passed as an argument into the stub
aoqi@6880:   if (has_argument) {
aoqi@6880:     __ ld_ptr(temp_reg, Address(FP, 2*BytesPerWord));
aoqi@6880:   }
aoqi@6880:   int call_offset;
aoqi@6880:   if (has_argument)
aoqi@6880:      call_offset = __ call_RT(noreg, noreg, target, temp_reg);
aoqi@1:   else
aoqi@6880:      call_offset = __ call_RT(noreg, noreg, target);
aoqi@1: 
aoqi@6880:   OopMapSet* oop_maps = new OopMapSet();
aoqi@6880:   oop_maps->add_gc_map(call_offset, oop_map);
aoqi@1: 
aoqi@6880:   __ stop("should not reach here");
aoqi@6880: 
aoqi@6880:   return oop_maps;
aoqi@1: }
aoqi@1: 
aoqi@1: OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
aoqi@8865:   __ block_comment("generate_handle_exception");
aoqi@8865: 
aoqi@8865:   // incoming parameters
aoqi@6880:   const Register exception_oop = V0;
aoqi@6880:   const Register exception_pc = V1;
aoqi@6880:   // other registers used in this stub
aoqi@8865:   const Register thread = TREG;
aoqi@8865: #ifndef OPT_THREAD
aoqi@8865:   __ get_thread(thread);
aoqi@8865: #endif
aoqi@1:   // Save registers, if required.
aoqi@8865:   OopMapSet* oop_maps = new OopMapSet();
aoqi@8865:   OopMap* oop_map = NULL;
aoqi@8865:   switch (id) {
aoqi@8865:   case forward_exception_id:
aoqi@8865:     // We're handling an exception in the context of a compiled frame.
aoqi@8865:     // The registers have been saved in the standard places.  Perform
aoqi@8865:     // an exception lookup in the caller and dispatch to the handler
aoqi@8865:     // if found.  Otherwise unwind and dispatch to the callers
aoqi@8865:     // exception handler.
aoqi@8865:     oop_map = generate_oop_map(sasm, 1 /*thread*/);
aoqi@6880: 
aoqi@8865:     // load and clear pending exception oop into RAX
aoqi@8865:     __ ld_ptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
aoqi@8865:     __ st_ptr(R0, Address(thread, Thread::pending_exception_offset()));
aoqi@6880: 
aoqi@8865:     // load issuing PC (the return address for this stub) into rdx
aoqi@8865:     __ ld_ptr(exception_pc, Address(FP, 1*BytesPerWord));
aoqi@6880: 
aoqi@8865:     // make sure that the vm_results are cleared (may be unnecessary)
aoqi@8865:     __ st_ptr(R0, Address(thread, JavaThread::vm_result_offset()));
aoqi@8865:     __ st_ptr(R0, Address(thread, JavaThread::vm_result_2_offset()));
aoqi@8865:     break;
aoqi@8865:   case handle_exception_nofpu_id:
aoqi@8865:   case handle_exception_id:
aoqi@8865:     // At this point all registers MAY be live.
aoqi@8865:     oop_map = save_live_registers(sasm, 1 /*thread*/, id != handle_exception_nofpu_id);
aoqi@8865:     break;
aoqi@8865:   case handle_exception_from_callee_id: {
aoqi@8865:     // At this point all registers except exception oop (RAX) and
aoqi@8865:     // exception pc (RDX) are dead.
aoqi@8865:     const int frame_size = 2 /*BP, return address*/ NOT_LP64(+ 1 /*thread*/);
aoqi@8865:     oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
aoqi@8865:     sasm->set_frame_size(frame_size);
aoqi@8865:     break;
aoqi@8865:   }
aoqi@8865:   default:  ShouldNotReachHere();
aoqi@8865:   }
aoqi@1: 
aoqi@1: #ifdef TIERED
aoqi@6880:   // C2 can leave the fpu stack dirty
aoqi@6880:   __ empty_FPU_stack();
aoqi@1: #endif // TIERED
aoqi@1: 
aoqi@6880:   // verify that only V0 and V1 is valid at this time
aoqi@6880:   // verify that V0 contains a valid exception
aoqi@6880:   __ verify_not_null_oop(exception_oop);
aoqi@1: 
aoqi@6880:   // load address of JavaThread object for thread-local data
aoqi@6880:   __ get_thread(thread);
aoqi@1: 
aoqi@1: #ifdef ASSERT
aoqi@6880:   // check that fields in JavaThread for exception oop and issuing pc are
aoqi@6880:   // empty before writing to them
aoqi@6880:   Label oop_empty;
aoqi@6880:   __ ld_ptr(AT, Address(thread, in_bytes(JavaThread::exception_oop_offset())));
aoqi@6880:   __ beq(AT, R0, oop_empty);
aoqi@6880:   __ delayed()->nop();
aoqi@6880:   __ stop("exception oop already set");
aoqi@6880:   __ bind(oop_empty);
fujie@9207: 
aoqi@6880:   Label pc_empty;
aoqi@6880:   __ ld_ptr(AT, Address(thread, in_bytes(JavaThread::exception_pc_offset())));
aoqi@6880:   __ beq(AT, R0, pc_empty);
aoqi@6880:   __ delayed()->nop();
aoqi@6880:   __ stop("exception pc already set");
aoqi@6880:   __ bind(pc_empty);
aoqi@1: #endif
aoqi@1: 
aoqi@6880:   // save exception oop and issuing pc into JavaThread
aoqi@6880:   // (exception handler will load it from here)
aoqi@6880:   __ st_ptr(exception_oop, Address(thread, in_bytes(JavaThread::exception_oop_offset())));
aoqi@6880:   __ st_ptr(exception_pc, Address(thread, in_bytes(JavaThread::exception_pc_offset())));
aoqi@1: 
aoqi@8865:   // patch throwing pc into return address (has bci & oop map)
aoqi@8865:   __ st_ptr(exception_pc, Address(FP, 1*BytesPerWord));
aoqi@1: 
aoqi@6880:   // compute the exception handler.
aoqi@6880:   // the exception oop and the throwing pc are read from the fields in JavaThread
fujie@9161:   __ block_comment(";; will call_RT exception_handler_for_pc");
aoqi@8865:   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
aoqi@6880:   oop_maps->add_gc_map(call_offset, oop_map);
fujie@9161:   __ block_comment(";; end of call_RT exception_handler_for_pc");
fujie@9207: 
aoqi@6880:   // V0:  handler address or NULL if no handler exists
aoqi@6880:   //      will be the deopt blob if nmethod was deoptimized while we looked up
aoqi@6880:   //      handler regardless of whether handler existed in the nmethod.
aoqi@1: 
aoqi@6880:   // only V0 is valid at this time, all other registers have been destroyed by the
aoqi@6880:   // runtime call
aoqi@1: 
aoqi@6880:   // patch the return address -> the stub will directly return to the exception handler
aoqi@8865:   __ st_ptr(V0, Address(FP, 1 * BytesPerWord));
aoqi@1: 
aoqi@8865:   switch (id) {
aoqi@8865:   case forward_exception_id:
aoqi@8865:   case handle_exception_nofpu_id:
aoqi@8865:   case handle_exception_id:
aoqi@8865:     // Restore the registers that were saved at the beginning.
aoqi@8865:     restore_live_registers(sasm, id != handle_exception_nofpu_id);
aoqi@8865:     break;
aoqi@8865:   case handle_exception_from_callee_id:
aoqi@8865:     // WIN64_ONLY: No need to add frame::arg_reg_save_area_bytes to SP
aoqi@8865:     // since we do a leave anyway.
aoqi@1: 
aoqi@8865:     // Pop the return address since we are possibly changing SP (restoring from BP).
fujie@9207:     __ move(SP, FP);
fujie@9207:     __ pop(FP);
fujie@9207:     __ pop(RA);
aoqi@8865:     __ jr(RA);  // jump to exception handler
aoqi@8865:     __ delayed()->nop();
aoqi@8865:     break;
aoqi@8865:    default:  ShouldNotReachHere();
aoqi@8865:   }
aoqi@1: 
aoqi@8865:   return oop_maps;
aoqi@8865: }
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1: void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
aoqi@6880:   // incoming parameters
aoqi@6880:   const Register exception_oop = V0;
aoqi@8865:   // callee-saved copy of exception_oop during runtime call
aoqi@8865:   const Register exception_oop_callee_saved = S0;
aoqi@6880:   // other registers used in this stub
aoqi@6880:   const Register exception_pc = V1;
aoqi@6880:   const Register handler_addr = T3;
aoqi@8865:   const Register thread = TREG;
aoqi@1: 
aoqi@6880:   // verify that only eax is valid at this time
aoqi@6880:   //  __ invalidate_registers(false, true, true, true, true, true);
aoqi@1: 
aoqi@1: #ifdef ASSERT
aoqi@6880:   // check that fields in JavaThread for exception oop and issuing pc are empty
aoqi@6880:   __ get_thread(thread);
aoqi@6880:   Label oop_empty;
aoqi@6880:   __ ld_ptr(AT, thread, in_bytes(JavaThread::exception_oop_offset()));
aoqi@6880:   __ beq(AT, R0, oop_empty);
aoqi@6880:   __ delayed()->nop();
aoqi@6880:   __ stop("exception oop must be empty");
aoqi@6880:   __ bind(oop_empty);
aoqi@1: 
aoqi@6880:   Label pc_empty;
aoqi@6880:   __ ld_ptr(AT, thread, in_bytes(JavaThread::exception_pc_offset()));
aoqi@8865:   __ beq(AT, R0, pc_empty);
aoqi@6880:   __ delayed()->nop();
aoqi@6880:   __ stop("exception pc must be empty");
aoqi@6880:   __ bind(pc_empty);
aoqi@1: #endif
aoqi@6880:   // clear the FPU stack in case any FPU results are left behind
aoqi@6880:   __ empty_FPU_stack();
aoqi@1: 
aoqi@8865:   // save exception_oop in callee-saved register to preserve it during runtime calls
aoqi@8865:   __ verify_not_null_oop(exception_oop);
aoqi@8865:   __ move(exception_oop_callee_saved, exception_oop);
aoqi@8865: 
aoqi@8865: #ifndef OPT_THREAD
aoqi@8865:   __ get_thread(thread);
aoqi@8865: #endif
fujie@9207:   // Get return address (is in RA after leave).
aoqi@8865: 
fujie@9207:   __ move(exception_pc, RA);
fujie@9207:   __ push(RA);
aoqi@1: 
aoqi@6880:   // search the exception handler address of the caller (using the return address)
aoqi@8865:   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
aoqi@8865:   // V0: exception handler address of the caller
aoqi@1: 
aoqi@6880:   // only eax is valid at this time, all other registers have been destroyed by the call
aoqi@1: 
aoqi@6880:   // move result of call into correct register
aoqi@6880:   __ move(handler_addr, V0);
aoqi@1: 
aoqi@8865:   // Restore exception oop to V0 (required convention of exception handler).
aoqi@8865:   __ move(exception_oop, exception_oop_callee_saved);
aoqi@8865: 
aoqi@8865:   // verify that there is really a valid exception in V0
aoqi@6880:   __ verify_oop(exception_oop);
aoqi@1: 
aoqi@6880:   // get throwing pc (= return address).
aoqi@8865:   // V1 has been destroyed by the call, so it must be set again
aoqi@6880:   // the pop is also necessary to simulate the effect of a ret(0)
aoqi@6880:   __ super_pop(exception_pc);
aoqi@1: 
aoqi@6880:   // continue at exception handler (return address removed)
aoqi@6880:   // note: do *not* remove arguments when unwinding the
aoqi@6880:   //       activation since the caller assumes having
aoqi@6880:   //       all arguments on the stack when entering the
aoqi@6880:   //       runtime to determine the exception handler
aoqi@6880:   //       (GC happens at call site with arguments!)
aoqi@8865:   // V0: exception oop
aoqi@8865:   // V1: throwing pc
aoqi@8865:   // T3: exception handler
aoqi@6880:   __ jr(handler_addr);
aoqi@6880:   __ delayed()->nop();
aoqi@1: }
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1: //static address deopt_with_exception_entry_for_patch = NULL;
aoqi@1: 
aoqi@1: OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
aoqi@1: 
aoqi@6880:   // use the maximum number of runtime-arguments here because it is difficult to
aoqi@6880:   // distinguish each RT-Call.
aoqi@6880:   // Note: This number affects also the RT-Call in generate_handle_exception because
aoqi@6880:   //       the oop-map is shared for all calls.
aoqi@1: 
aoqi@6880:   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
aoqi@6880:   assert(deopt_blob != NULL, "deoptimization blob must have been created");
aoqi@6880:   // assert(deopt_with_exception_entry_for_patch != NULL,
aoqi@6880:   // "deoptimization blob must have been created");
aoqi@6880: 
aoqi@6880:   //OopMap* oop_map = save_live_registers(sasm, num_rt_args);
aoqi@6880:   OopMap* oop_map = save_live_registers(sasm, 0);
aoqi@6880:   const Register thread = T8;
aoqi@6880:   // push java thread (becomes first argument of C function)
aoqi@6880:   __ get_thread(thread);
aoqi@6880:   __ move(A0, thread);
aoqi@1: 
aoqi@1: 
aoqi@6880: /*
aoqi@6880:  *  NOTE: this frame should be compiled frame, but at this point, the pc in frame-anchor
aoqi@6880:  *  is contained in interpreter. It should be wrong, and should be cleared but is not.
aoqi@6880:  *   even if we cleared the wrong pc in anchor, the default way to get caller pc in class frame
aoqi@6880:  *   is not right. It depends on that the caller pc is stored in *(sp - 1) but it's not the case
aoqi@1:  */
aoqi@6880:   __ set_last_Java_frame(thread, NOREG, FP, NULL);
aoqi@6880:   NOT_LP64(__ addiu(SP, SP, (-1) * wordSize));
aoqi@6880:   __ move(AT, -(StackAlignmentInBytes));
aoqi@6880:   __ andr(SP, SP, AT);
aoqi@6880:   __ relocate(relocInfo::internal_pc_type);
aoqi@6880:   {
aoqi@1: #ifndef _LP64
aoqi@6880:     int save_pc = (int)__ pc() +  12 + NativeCall::return_address_offset;
aoqi@6880:     __ lui(AT, Assembler::split_high(save_pc));
aoqi@6880:     __ addiu(AT, AT, Assembler::split_low(save_pc));
aoqi@1: #else
aoqi@6880:     uintptr_t save_pc = (uintptr_t)__ pc() + NativeMovConstReg::instruction_size + 1 * BytesPerInstWord + NativeCall::return_address_offset_long;
aoqi@6880:     __ li48(AT, save_pc);
aoqi@1: #endif
aoqi@6880:   }
aoqi@6880:   __ st_ptr(AT, thread, in_bytes(JavaThread::last_Java_pc_offset()));
aoqi@1: 
aoqi@6880:   // do the call
aoqi@1: #ifndef _LP64
aoqi@6880:   __ lui(T9, Assembler::split_high((int)target));
aoqi@6880:   __ addiu(T9, T9, Assembler::split_low((int)target));
aoqi@1: #else
aoqi@6880:   __ li48(T9, (intptr_t)target);
aoqi@1: #endif
aoqi@6880:   __ jalr(T9);
aoqi@6880:   __ delayed()->nop();
aoqi@6880:   OopMapSet*  oop_maps = new OopMapSet();
aoqi@6880:   oop_maps->add_gc_map(__ offset(),  oop_map);
aoqi@1: 
aoqi@6880:   __ get_thread(thread);
aoqi@1: 
aoqi@6880:   __ ld_ptr (SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
fujie@9171:   __ reset_last_Java_frame(thread, true);
aoqi@6880:   // discard thread arg
aoqi@6880:   // check for pending exceptions
aoqi@6880:   {
aoqi@6880:     Label L, skip;
aoqi@6880:     //Label no_deopt;
aoqi@6880:     __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset()));
aoqi@6880:     __ beq(AT, R0, L);
aoqi@6880:     __ delayed()->nop();
aoqi@6880:     // exception pending => remove activation and forward to exception handler
aoqi@1: 
aoqi@8865:     __ bne(V0, R0, skip);
aoqi@6880:     __ delayed()->nop();
aoqi@6880:     __ jmp(Runtime1::entry_for(Runtime1::forward_exception_id),
aoqi@6880:         relocInfo::runtime_call_type);
aoqi@6880:     __ delayed()->nop();
aoqi@6880:     __ bind(skip);
aoqi@1: 
aoqi@6880:     // the deopt blob expects exceptions in the special fields of
aoqi@6880:     // JavaThread, so copy and clear pending exception.
aoqi@1: 
aoqi@6880:     // load and clear pending exception
aoqi@6880:     __ ld_ptr(V0, Address(thread,in_bytes(Thread::pending_exception_offset())));
aoqi@6880:     __ st_ptr(R0, Address(thread, in_bytes(Thread::pending_exception_offset())));
aoqi@1: 
aoqi@6880:     // check that there is really a valid exception
aoqi@6880:     __ verify_not_null_oop(V0);
aoqi@6880: 
aoqi@6880:     // load throwing pc: this is the return address of the stub
aoqi@6880:     __ ld_ptr(V1, Address(SP, return_off * BytesPerWord));
aoqi@1: 
aoqi@1: 
aoqi@1: #ifdef ASSERT
aoqi@6880:     // check that fields in JavaThread for exception oop and issuing pc are empty
aoqi@6880:     Label oop_empty;
aoqi@6880:     __ ld_ptr(AT, Address(thread, in_bytes(JavaThread::exception_oop_offset())));
aoqi@6880:     __ beq(AT,R0,oop_empty);
aoqi@6880:     __ delayed()->nop();
aoqi@6880:     __ stop("exception oop must be empty");
aoqi@6880:     __ bind(oop_empty);
aoqi@1: 
aoqi@6880:     Label pc_empty;
aoqi@6880:     __ ld_ptr(AT, Address(thread, in_bytes(JavaThread::exception_pc_offset())));
aoqi@6880:     __ beq(AT,R0,pc_empty);
aoqi@6880:     __ delayed()->nop();
aoqi@6880:     __ stop("exception pc must be empty");
aoqi@6880:     __ bind(pc_empty);
aoqi@1: #endif
aoqi@1: 
aoqi@6880:     // store exception oop and throwing pc to JavaThread
aoqi@6880:     __ st_ptr(V0,Address(thread, in_bytes(JavaThread::exception_oop_offset())));
aoqi@6880:     __ st_ptr(V1,Address(thread, in_bytes(JavaThread::exception_pc_offset())));
aoqi@1: 
aoqi@6880:     restore_live_registers(sasm);
aoqi@1: 
aoqi@6880:     __ leave();
aoqi@1: 
aoqi@6880:     // Forward the exception directly to deopt blob. We can blow no
aoqi@6880:     // registers and must leave throwing pc on the stack.  A patch may
aoqi@6880:     // have values live in registers so the entry point with the
aoqi@6880:     // exception in tls.
aoqi@6880:     __ jmp(deopt_blob->unpack_with_exception_in_tls(), relocInfo::runtime_call_type);
aoqi@6880:     __ delayed()->nop();
aoqi@1: 
aoqi@6880:     __ bind(L);
aoqi@6880:   }
aoqi@1: 
aoqi@6880:   // Runtime will return true if the nmethod has been deoptimized during
aoqi@6880:   // the patching process. In that case we must do a deopt reexecute instead.
aoqi@1: 
aoqi@6880:   Label reexecuteEntry, cont;
aoqi@1: 
aoqi@6880:   __ beq(V0, R0, cont);                              // have we deoptimized?
aoqi@6880:   __ delayed()->nop();
aoqi@1: 
aoqi@6880:   // Will reexecute. Proper return address is already on the stack we just restore
aoqi@6880:   // registers, pop all of our frame but the return address and jump to the deopt blob
aoqi@6880:   restore_live_registers(sasm);
aoqi@1: 
aoqi@6880:   __ leave();
aoqi@6880:   __ jmp(deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
aoqi@6880:   __ delayed()->nop();
aoqi@1: 
aoqi@6880:   __ bind(cont);
aoqi@6880:   restore_live_registers(sasm);
aoqi@1: 
aoqi@6880:   __ leave();
aoqi@6880:   __ jr(RA);
aoqi@6880:   __ delayed()->nop();
aoqi@6880: 
aoqi@6880:   return oop_maps;
aoqi@1: }
aoqi@1: 
aoqi@1: 
aoqi@1: OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
aoqi@6880:   // for better readability
aoqi@6880:   const bool must_gc_arguments = true;
aoqi@6880:   const bool dont_gc_arguments = false;
aoqi@1: 
aoqi@1: 
aoqi@6880:   // default value; overwritten for some optimized stubs that are called
aoqi@6880:   // from methods that do not use the fpu
aoqi@6880:   bool save_fpu_registers = true;
aoqi@1: 
aoqi@1: 
aoqi@6880:   // stub code & info for the different stubs
aoqi@6880:   OopMapSet* oop_maps = NULL;
aoqi@1: 
aoqi@1:   switch (id) {
aoqi@1:     case forward_exception_id:
aoqi@1:       {
aoqi@8865:         oop_maps = generate_handle_exception(id, sasm);
aoqi@8865:         __ leave();
aoqi@8865:         __ jr(RA);
aoqi@8865:         __ delayed()->nop();
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@1:     case new_instance_id:
aoqi@1:     case fast_new_instance_id:
aoqi@1:     case fast_new_instance_init_check_id:
aoqi@1:       {
aoqi@1:         Register klass = A4; // Incoming
aoqi@1:         Register obj   = V0; // Result
aoqi@1: 
aoqi@1:         if (id == new_instance_id) {
aoqi@1:           __ set_info("new_instance", dont_gc_arguments);
aoqi@1:         } else if (id == fast_new_instance_id) {
aoqi@1:           __ set_info("fast new_instance", dont_gc_arguments);
aoqi@1:         } else {
aoqi@1:           assert(id == fast_new_instance_init_check_id, "bad StubID");
aoqi@1:           __ set_info("fast new_instance init check", dont_gc_arguments);
aoqi@1:         }
aoqi@1: 
aoqi@6880:         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id)
aoqi@1:              && UseTLAB && FastTLABRefill) {
aoqi@1:           Label slow_path;
aoqi@1:           Register obj_size = T0;
aoqi@1:           Register t1       = T2;
aoqi@1:           Register t2       = T3;
aoqi@1:           assert_different_registers(klass, obj, obj_size, t1, t2);
aoqi@1:           if (id == fast_new_instance_init_check_id) {
aoqi@1:             // make sure the klass is initialized
aoqi@8865:             __ ld_ptr(AT, Address(klass, in_bytes(InstanceKlass::init_state_offset())));
aoqi@1:             __ move(t1, InstanceKlass::fully_initialized);
aoqi@1:             __ bne(AT, t1, slow_path);
aoqi@1:             __ delayed()->nop();
aoqi@1:           }
aoqi@1: #ifdef ASSERT
aoqi@1:           // assert object can be fast path allocated
aoqi@1:           {
aoqi@1:             Label ok, not_ok;
aoqi@1:             __ lw(obj_size, klass, in_bytes(Klass::layout_helper_offset()));
aoqi@1:             __ blez(obj_size, not_ok);
aoqi@1:             __ delayed()->nop();
aoqi@1:             __ andi(t1 , obj_size, Klass::_lh_instance_slow_path_bit);
aoqi@1:             __ beq(t1, R0, ok);
aoqi@1:             __ delayed()->nop();
aoqi@1:             __ bind(not_ok);
aoqi@1:             __ stop("assert(can be fast path allocated)");
aoqi@1:             __ should_not_reach_here();
aoqi@1:             __ bind(ok);
aoqi@1:           }
aoqi@1: #endif // ASSERT
aoqi@1:           // if we got here then the TLAB allocation failed, so try
aoqi@1:           // refilling the TLAB or allocating directly from eden.
aoqi@6880: 
aoqi@1:           Label retry_tlab, try_eden;
aoqi@1:           __ tlab_refill(retry_tlab, try_eden, slow_path); // does not destroy edx (klass)
aoqi@6880: 
aoqi@1:           __ bind(retry_tlab);
aoqi@6880: 
aoqi@1:           // get the instance size
aoqi@1:           __ lw(obj_size, klass, in_bytes(Klass::layout_helper_offset()));
aoqi@1:           __ tlab_allocate(obj, obj_size, 0, t1, t2, slow_path);
aoqi@1:           __ initialize_object(obj, klass, obj_size, 0, t1, t2);
aoqi@1:           __ verify_oop(obj);
aoqi@1:           __ jr(RA);
aoqi@1:           __ delayed()->nop();
aoqi@6880: 
wangxue@9205: #ifndef OPT_THREAD
wangxue@9205:           const Register thread = T8;
wangxue@9205:           __ get_thread(thread);
wangxue@9205: #else
wangxue@9205:           const Register thread = TREG;
wangxue@9205: #endif
wangxue@9205: 
aoqi@1:           __ bind(try_eden);
aoqi@1: 
aoqi@6880:           // get the instance size
aoqi@1:           __ lw(obj_size, klass, in_bytes(Klass::layout_helper_offset()));
aoqi@1:           __ eden_allocate(obj, obj_size, 0, t1, t2, slow_path);
wangxue@9205:           __ incr_allocated_bytes(thread, obj_size, 0);
wangxue@9205: 
aoqi@1:           __ initialize_object(obj, klass, obj_size, 0, t1, t2);
aoqi@1:           __ verify_oop(obj);
aoqi@1:           __ jr(RA);
aoqi@1:           __ delayed()->nop();
aoqi@6880: 
aoqi@1:           __ bind(slow_path);
aoqi@1:         }
aoqi@1:         __ enter();
aoqi@1:         OopMap* map = save_live_registers(sasm, 0);
aoqi@1:         int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
aoqi@1:         oop_maps = new OopMapSet();
aoqi@1:         oop_maps->add_gc_map(call_offset, map);
aoqi@1:         restore_live_registers_except_V0(sasm);
aoqi@1:         __ verify_oop(obj);
aoqi@1:         __ leave();
aoqi@1:         __ jr(RA);
aoqi@1:         __ delayed()->nop();
aoqi@6880: 
aoqi@1:         // V0: new instance
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@1: 
aoqi@1: #ifdef TIERED
aoqi@1: //FIXME, I hava no idea which register to use
aoqi@1:     case counter_overflow_id:
aoqi@1:       {
aoqi@1: #ifndef _LP64
aoqi@1:         Register bci = T5;
aoqi@1: #else
aoqi@1:         Register bci = A5;
aoqi@1: #endif
aoqi@8865:         Register method = AT;
aoqi@1:         __ enter();
aoqi@1:         OopMap* map = save_live_registers(sasm, 0);
aoqi@1:         // Retrieve bci
aoqi@1:         __ lw(bci, Address(FP, 2*BytesPerWord));// FIXME:wuhui.ebp==??
aoqi@8865:         __ ld(method, Address(FP, 3*BytesPerWord));
aoqi@8865:         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
aoqi@1:         oop_maps = new OopMapSet();
aoqi@1:         oop_maps->add_gc_map(call_offset, map);
aoqi@1:         restore_live_registers(sasm);
aoqi@1:         __ leave();
aoqi@6880:         __ jr(RA);
aoqi@6880:         __ delayed()->nop();
aoqi@1:       }
aoqi@1:       break;
aoqi@1: #endif // TIERED
aoqi@1: 
aoqi@1: 
aoqi@1: 
aoqi@1:     case new_type_array_id:
aoqi@1:     case new_object_array_id:
aoqi@6880:       {
aoqi@6880:         // i use T2 as length register, T4 as klass register, V0 as result register.
aoqi@1:         // MUST accord with NewTypeArrayStub::emit_code, NewObjectArrayStub::emit_code
aoqi@1:         Register length   = T2; // Incoming
aoqi@1: #ifndef _LP64
aoqi@1:         Register klass    = T4; // Incoming
aoqi@1: #else
aoqi@1:         Register klass    = A4; // Incoming
aoqi@1: #endif
aoqi@1:         Register obj      = V0; // Result
aoqi@6880: 
aoqi@1:         if (id == new_type_array_id) {
aoqi@1:           __ set_info("new_type_array", dont_gc_arguments);
aoqi@1:         } else {
aoqi@1:           __ set_info("new_object_array", dont_gc_arguments);
aoqi@1:         }
aoqi@6880: 
aoqi@1:         if (UseTLAB && FastTLABRefill) {
aoqi@1:           Register arr_size = T0;
aoqi@6880:           Register t1       = T1;
aoqi@1:           Register t2       = T3;
aoqi@1:           Label slow_path;
aoqi@1:           assert_different_registers(length, klass, obj, arr_size, t1, t2);
aoqi@6880: 
aoqi@1:           // check that array length is small enough for fast path
aoqi@1:           __ move(AT, C1_MacroAssembler::max_array_allocation_length);
aoqi@1:           __ sltu(AT, AT, length);
aoqi@1:           __ bne(AT, R0, slow_path);
aoqi@1:           __ delayed()->nop();
aoqi@1: 
aoqi@1:           // if we got here then the TLAB allocation failed, so try
aoqi@1:           // refilling the TLAB or allocating directly from eden.
aoqi@1:           Label retry_tlab, try_eden;
aoqi@6880:           //T0,T1,T5,T8 have changed!
aoqi@1:           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves T2 & T4
aoqi@6880: 
aoqi@1:           __ bind(retry_tlab);
aoqi@6880: 
aoqi@1:           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
aoqi@6880:           __ lw(t1, klass, in_bytes(Klass::layout_helper_offset()));
aoqi@1:           __ andi(AT, t1, 0x1f);
aoqi@1:           __ sllv(arr_size, length, AT);
aoqi@1:           __ srl(t1, t1, Klass::_lh_header_size_shift);
aoqi@1:           __ andi(t1, t1, Klass::_lh_header_size_mask);
aoqi@1:           __ add(arr_size, t1, arr_size);
aoqi@1:           __ addi(arr_size, arr_size, MinObjAlignmentInBytesMask);  // align up
aoqi@1:           __ move(AT, ~MinObjAlignmentInBytesMask);
aoqi@1:           __ andr(arr_size, arr_size, AT);
aoqi@6880: 
aoqi@6880: 
aoqi@1:           __ tlab_allocate(obj, arr_size, 0, t1, t2, slow_path);  // preserves arr_size
aoqi@1:           __ initialize_header(obj, klass, length,t1,t2);
aoqi@6880:           __ lbu(t1, Address(klass, in_bytes(Klass::layout_helper_offset())
aoqi@1:                                     + (Klass::_lh_header_size_shift / BitsPerByte)));
aoqi@1:           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
aoqi@1:           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
aoqi@1:           __ andi(t1, t1, Klass::_lh_header_size_mask);
aoqi@1:           __ sub(arr_size, arr_size, t1);  // body length
aoqi@1:           __ add(t1, t1, obj);             // body start
aoqi@1:           __ initialize_body(t1, arr_size, 0, t2);
aoqi@1:           __ verify_oop(obj);
aoqi@1:           __ jr(RA);
aoqi@1:           __ delayed()->nop();
aoqi@6880: 
wangxue@9205: #ifndef OPT_THREAD
wangxue@9205:           const Register thread = T8;
wangxue@9205:           __ get_thread(thread);
wangxue@9205: #else
wangxue@9205:           const Register thread = TREG;
wangxue@9205: #endif
wangxue@9205: 
aoqi@1:           __ bind(try_eden);
aoqi@1:           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
aoqi@6880:           __ lw(t1, klass, in_bytes(Klass::layout_helper_offset()));
aoqi@1:           __ andi(AT, t1, 0x1f);
aoqi@1:           __ sllv(arr_size, length, AT);
aoqi@1:           __ srl(t1, t1, Klass::_lh_header_size_shift);
aoqi@1:           __ andi(t1, t1, Klass::_lh_header_size_mask);
aoqi@1:           __ add(arr_size, t1, arr_size);
aoqi@1:           __ addi(arr_size, arr_size, MinObjAlignmentInBytesMask);  // align up
aoqi@1:           __ move(AT, ~MinObjAlignmentInBytesMask);
aoqi@1:           __ andr(arr_size, arr_size, AT);
aoqi@1:           __ eden_allocate(obj, arr_size, 0, t1, t2, slow_path);  // preserves arr_size
wangxue@9205:           __ incr_allocated_bytes(thread, arr_size, 0);
wangxue@9205: 
aoqi@1:           __ initialize_header(obj, klass, length,t1,t2);
aoqi@1:           __ lbu(t1, Address(klass, in_bytes(Klass::layout_helper_offset())
aoqi@1:                                     + (Klass::_lh_header_size_shift / BitsPerByte)));
aoqi@1:           __ andi(t1, t1, Klass::_lh_header_size_mask);
aoqi@1:           __ sub(arr_size, arr_size, t1);  // body length
aoqi@1:           __ add(t1, t1, obj);             // body start
aoqi@6880: 
aoqi@1:           __ initialize_body(t1, arr_size, 0, t2);
aoqi@1:           __ verify_oop(obj);
aoqi@1:           __ jr(RA);
aoqi@1:           __ delayed()->nop();
aoqi@1:           __ bind(slow_path);
aoqi@1:         }
aoqi@6880: 
aoqi@6880: 
aoqi@1:         __ enter();
aoqi@1:         OopMap* map = save_live_registers(sasm, 0);
aoqi@1:         int call_offset;
aoqi@1:         if (id == new_type_array_id) {
aoqi@6880:           call_offset = __ call_RT(obj, noreg,
aoqi@1:                                     CAST_FROM_FN_PTR(address, new_type_array), klass, length);
aoqi@1:         } else {
aoqi@6880:           call_offset = __ call_RT(obj, noreg,
aoqi@1:                                    CAST_FROM_FN_PTR(address, new_object_array), klass, length);
aoqi@1:         }
aoqi@6880: 
aoqi@1:         oop_maps = new OopMapSet();
aoqi@1:         oop_maps->add_gc_map(call_offset, map);
aoqi@1:         restore_live_registers_except_V0(sasm);
aoqi@1:         __ verify_oop(obj);
aoqi@6880:         __ leave();
aoqi@1:         __ jr(RA);
aoqi@1:         __ delayed()->nop();
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@1:     case new_multi_array_id:
aoqi@6880:       {
aoqi@6880:         StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
aoqi@6880:        //refer to c1_LIRGenerate_mips.cpp:do_NewmultiArray
aoqi@6880:         // V0: klass
aoqi@6880:         // T2: rank
aoqi@6880:         // T0: address of 1st dimension
aoqi@6880:         //__ call_RT(V0, noreg, CAST_FROM_FN_PTR(address, new_multi_array), A1, A2, A3);
aoqi@6880:         //OopMap* map = save_live_registers(sasm, 4);
aoqi@6880:         OopMap* map = save_live_registers(sasm, 0);
aoqi@6880:         int call_offset = __ call_RT(V0, noreg, CAST_FROM_FN_PTR(address, new_multi_array),
aoqi@6880:             V0,T2,T0);
aoqi@6880:         oop_maps = new OopMapSet();
aoqi@6880:         oop_maps->add_gc_map(call_offset, map);
aoqi@6880:         //FIXME
aoqi@6880:         restore_live_registers_except_V0(sasm);
aoqi@6880:         // V0: new multi array
aoqi@6880:         __ verify_oop(V0);
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@6880: 
aoqi@1:     case register_finalizer_id:
aoqi@1:       {
aoqi@6880:         __ set_info("register_finalizer", dont_gc_arguments);
aoqi@1: 
aoqi@6880:         // The object is passed on the stack and we haven't pushed a
aoqi@6880:         // frame yet so it's one work away from top of stack.
aoqi@1:         //reference to LIRGenerator::do_RegisterFinalizer, call_runtime
aoqi@6880:         __ move(V0, A0);
aoqi@6880:         __ verify_oop(V0);
aoqi@6880:         // load the klass and check the has finalizer flag
aoqi@6880:         Label register_finalizer;
aoqi@1: #ifndef _LP64
aoqi@6880:         Register t = T5;
aoqi@1: #else
aoqi@6880:         Register t = A5;
aoqi@1: #endif
aoqi@6880:         //__ ld_ptr(t, Address(V0, oopDesc::klass_offset_in_bytes()));
aoqi@6880:         __ load_klass(t, V0);
aoqi@6880:         __ lw(t, Address(t, Klass::access_flags_offset()));
aoqi@6880:         __ move(AT, JVM_ACC_HAS_FINALIZER);
aoqi@6880:         __ andr(AT, AT, t);
aoqi@1: 
aoqi@6880:         __ bne(AT, R0, register_finalizer);
aoqi@6880:         __ delayed()->nop();
aoqi@6880:         __ jr(RA);
aoqi@6880:         __ delayed()->nop();
aoqi@6880:         __ bind(register_finalizer);
aoqi@6880:         __ enter();
aoqi@6880:         OopMap* map = save_live_registers(sasm, 0 /*num_rt_args */);
aoqi@6880: 
aoqi@6880:         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address,
aoqi@6880:               SharedRuntime::register_finalizer), V0);
aoqi@6880:         oop_maps = new OopMapSet();
aoqi@1:         oop_maps->add_gc_map(call_offset, map);
aoqi@1: 
aoqi@6880:         // Now restore all the live registers
aoqi@6880:         restore_live_registers(sasm);
aoqi@1: 
aoqi@6880:         __ leave();
aoqi@6880:         __ jr(RA);
aoqi@6880:         __ delayed()->nop();
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@6880: //  case range_check_failed_id:
aoqi@6880:   case throw_range_check_failed_id:
aoqi@8865:       {
aoqi@8865:         StubFrame f(sasm, "range_check_failed", dont_gc_arguments);
aoqi@6880:         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
aoqi@1:               throw_range_check_exception),true);
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@1:       case throw_index_exception_id:
aoqi@6880:       {
aoqi@6880:         // i use A1 as the index register, for this will be the first argument, see call_RT
aoqi@6880:         StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments);
aoqi@6880:         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
aoqi@6880:               throw_index_exception), true);
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@6880:   case throw_div0_exception_id:
aoqi@1:       { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments);
aoqi@6880:         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
aoqi@6880:               throw_div0_exception), false);
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@6880:   case throw_null_pointer_exception_id:
aoqi@8865:       {
aoqi@8865:         StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments);
aoqi@6880:         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
aoqi@6880:               throw_null_pointer_exception),false);
aoqi@1:       }
aoqi@1:       break;
aoqi@1: 
aoqi@8865:   case handle_exception_nofpu_id:
aoqi@6880:     save_fpu_registers = false;
aoqi@6880:      // fall through
aoqi@6880:   case handle_exception_id:
aoqi@6880:     {
aoqi@6880:       StubFrame f(sasm, "handle_exception", dont_gc_arguments);
aoqi@6880:       //OopMap* oop_map = save_live_registers(sasm, 1, save_fpu_registers);
aoqi@6880:       oop_maps = generate_handle_exception(id, sasm);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@8865:   case handle_exception_from_callee_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
aoqi@8865:       oop_maps = generate_handle_exception(id, sasm);
aoqi@8865:     }
aoqi@8865:     break;
aoqi@6880:   case unwind_exception_id:
aoqi@6880:     {
aoqi@6880:       __ set_info("unwind_exception", dont_gc_arguments);
aoqi@6880:       generate_unwind_exception(sasm);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@1: 
aoqi@6880:   case throw_array_store_exception_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments);
aoqi@6880:       // tos + 0: link
aoqi@6880:       //     + 1: return address
aoqi@6880:       oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
wangxue@9248:             throw_array_store_exception), true);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@6880:   case throw_class_cast_exception_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments);
aoqi@6880:       oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address,
aoqi@8865:             throw_class_cast_exception), true);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@6880:   case throw_incompatible_class_change_error_id:
aoqi@6880:     {
aoqi@8865:       StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments);
aoqi@8865:       oop_maps = generate_exception_throw(sasm,
aoqi@8865:             CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@6880:   case slow_subtype_check_id:
aoqi@6880:     {
aoqi@6880:     //actually , We do not use it
aoqi@6880:       // A0:klass_RInfo    sub
aoqi@6880:       // A1:k->encoding() super
aoqi@6880:       __ set_info("slow_subtype_check", dont_gc_arguments);
aoqi@6880:       __ st_ptr(T0, SP, (-1) * wordSize);
aoqi@6880:       __ st_ptr(T1, SP, (-2) * wordSize);
aoqi@6880:       __ addiu(SP, SP, (-2) * wordSize);
aoqi@1: 
fujie@9132:       Label miss;
fujie@9132:       __ check_klass_subtype_slow_path(A0, A1, T0, T1, NULL, &miss);
aoqi@1: 
aoqi@6880:       __ addiu(V0, R0, 1);
aoqi@6880:       __ addiu(SP, SP, 2 * wordSize);
aoqi@6880:       __ ld_ptr(T0, SP, (-1) * wordSize);
aoqi@6880:       __ ld_ptr(T1, SP, (-2) * wordSize);
aoqi@6880:       __ jr(RA);
aoqi@6880:       __ delayed()->nop();
aoqi@1: 
aoqi@1: 
aoqi@6880:       __ bind(miss);
aoqi@6880:       __ move(V0, R0);
aoqi@6880:       __ addiu(SP, SP, 2 * wordSize);
aoqi@6880:       __ ld_ptr(T0, SP, (-1) * wordSize);
aoqi@6880:       __ ld_ptr(T1, SP, (-2) * wordSize);
aoqi@6880:       __ jr(RA);
aoqi@6880:       __ delayed()->nop();
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@1:   case monitorenter_nofpu_id:
aoqi@1:     save_fpu_registers = false;// fall through
aoqi@1: 
aoqi@6880:   case monitorenter_id:
aoqi@1:     {
aoqi@6880:       StubFrame f(sasm, "monitorenter", dont_gc_arguments);
aoqi@6880:       OopMap* map = save_live_registers(sasm, 0, save_fpu_registers);
aoqi@1: 
aoqi@6880:       f.load_argument(1, V0); // V0: object
aoqi@1: #ifndef _LP64
aoqi@6880:       f.load_argument(0, T6); // T6: lock address
aoqi@6880:       int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address,
aoqi@6880:            monitorenter), V0, T6);
aoqi@1: #else
aoqi@6880:       f.load_argument(0, A6); // A6: lock address
aoqi@6880:       int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address,
aoqi@6880:            monitorenter), V0, A6);
aoqi@1: #endif
aoqi@1: 
aoqi@6880:       oop_maps = new OopMapSet();
aoqi@6880:       oop_maps->add_gc_map(call_offset, map);
aoqi@6880:       restore_live_registers(sasm, save_fpu_registers);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@1: 
aoqi@6880:   case monitorexit_nofpu_id:
aoqi@6880:     save_fpu_registers = false;
aoqi@6880:         // fall through
aoqi@6880:   case monitorexit_id:
aoqi@6880:     {
aoqi@1:       StubFrame f(sasm, "monitorexit", dont_gc_arguments);
aoqi@1:       OopMap* map = save_live_registers(sasm, 0, save_fpu_registers);
aoqi@6880: 
aoqi@1: #ifndef _LP64
aoqi@1:       f.load_argument(0, T6); // eax: lock address
aoqi@1: #else
aoqi@1:       f.load_argument(0, A6); // A6: lock address
aoqi@1: #endif
aoqi@1:       // note: really a leaf routine but must setup last java sp
aoqi@1:       //       => use call_RT for now (speed can be improved by
aoqi@1:       //       doing last java sp setup manually)
aoqi@1: #ifndef _LP64
aoqi@6880:       int call_offset = __ call_RT(noreg, noreg,
aoqi@6880:                                     CAST_FROM_FN_PTR(address, monitorexit), T6);
aoqi@1: #else
aoqi@6880:       int call_offset = __ call_RT(noreg, noreg,
aoqi@6880:                                     CAST_FROM_FN_PTR(address, monitorexit), A6);
aoqi@1: #endif
aoqi@1:       oop_maps = new OopMapSet();
aoqi@1:       oop_maps->add_gc_map(call_offset, map);
aoqi@1:       restore_live_registers(sasm, save_fpu_registers);
aoqi@6880: 
aoqi@1:     }
aoqi@1:     break;
aoqi@6880:         //  case init_check_patching_id:
aoqi@6880:   case access_field_patching_id:
aoqi@6880:     {
aoqi@1:       StubFrame f(sasm, "access_field_patching", dont_gc_arguments);
aoqi@1:       // we should set up register map
aoqi@1:       oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
aoqi@1: 
aoqi@1:     }
aoqi@1:     break;
aoqi@1: 
aoqi@6880:   case load_klass_patching_id:
aoqi@6880:     {
aoqi@6880:       StubFrame f(sasm, "load_klass_patching", dont_gc_arguments);
aoqi@6880:       // we should set up register map
aoqi@6880:       oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address,
aoqi@6880:             move_klass_patching));
aoqi@6880:     }
aoqi@6880:     break;
aoqi@6880: /*  case jvmti_exception_throw_id:
aoqi@6880:     {
aoqi@6880:       // V0: exception oop
aoqi@6880:       // V1: exception pc
aoqi@6880:       StubFrame f(sasm, "jvmti_exception_throw", dont_gc_arguments);
aoqi@6880:       // Preserve all registers across this potentially blocking call
aoqi@6880:       const int num_rt_args = 2;  // thread, exception oop
aoqi@6880:       //OopMap* map = save_live_registers(sasm, num_rt_args);
aoqi@6880:       OopMap* map = save_live_registers(sasm, 0);
aoqi@6880:       int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address,
aoqi@6880:             Runtime1::post_jvmti_exception_throw), V0);
aoqi@6880:       oop_maps = new OopMapSet();
aoqi@6880:       oop_maps->add_gc_map(call_offset,  map);
aoqi@6880:       restore_live_registers(sasm);
aoqi@6880:     }*/
aoqi@8865:   case load_mirror_patching_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "load_mirror_patching" , dont_gc_arguments);
aoqi@8865:       oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
aoqi@8865:     }
aoqi@6880:     break;
aoqi@8865: 
aoqi@8865:   case load_appendix_patching_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments);
aoqi@8865:       // we should set up register map
aoqi@8865:       oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
aoqi@8865:     }
aoqi@8865:     break;
aoqi@8865: 
aoqi@6880:   case dtrace_object_alloc_id:
aoqi@6880:     {
aoqi@6880:       // V0:object
aoqi@6880:       StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
aoqi@6880:       // we can't gc here so skip the oopmap but make sure that all
aoqi@6880:       // the live registers get saved.
aoqi@6880:       save_live_registers(sasm, 0);
aoqi@1: 
aoqi@6880:       __ push_reg(V0);
aoqi@8865:       __ move(A0, V0);
aoqi@6880:       __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
aoqi@6880:           relocInfo::runtime_call_type);
fujie@9134:       __ delayed()->nop();
aoqi@6880:       __ super_pop(V0);
aoqi@1: 
aoqi@6880:       restore_live_registers(sasm);
aoqi@6880:     }
aoqi@6880:     break;
aoqi@8865: 
aoqi@6880:   case fpu2long_stub_id:
aoqi@6880:     {
aoqi@8865:       //FIXME, I hava no idea how to port this
aoqi@8865:       //tty->print_cr("fpu2long_stub_id unimplemented yet!");
aoqi@6880:     }
aoqi@8865:     break;
aoqi@8865: 
aoqi@8865:   case deoptimize_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "deoptimize", dont_gc_arguments);
aoqi@8865:       const int num_rt_args = 1;  // thread
aoqi@8865:       OopMap* oop_map = save_live_registers(sasm, num_rt_args);
aoqi@8865:       int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
aoqi@8865:       oop_maps = new OopMapSet();
aoqi@8865:       oop_maps->add_gc_map(call_offset, oop_map);
aoqi@8865:       restore_live_registers(sasm);
aoqi@8865:       DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
aoqi@8865:       assert(deopt_blob != NULL, "deoptimization blob must have been created");
aoqi@8865:       __ leave();
aoqi@8865:       __ jmp(deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
zhaixiang@9135:       __ delayed()->nop();
aoqi@8865:     }
aoqi@8865:    break;
aoqi@8865: 
aoqi@8865:   case predicate_failed_trap_id:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments);
aoqi@8865: 
aoqi@8865:       OopMap* map = save_live_registers(sasm, 1);
aoqi@8865: 
aoqi@8865:       int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
aoqi@8865:       oop_maps = new OopMapSet();
aoqi@8865:       oop_maps->add_gc_map(call_offset, map);
aoqi@8865:       restore_live_registers(sasm);
aoqi@8865:       __ leave();
aoqi@8865:       DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
aoqi@8865:       assert(deopt_blob != NULL, "deoptimization blob must have been created");
aoqi@8865: 
aoqi@8865:       __ jmp(deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
zhaixiang@9135:       __ delayed()->nop();
aoqi@8865:     }
aoqi@8865:    break;
aoqi@8865: 
aoqi@6880:   default:
aoqi@8865:     {
aoqi@8865:       StubFrame f(sasm, "unimplemented entry", dont_gc_arguments);
aoqi@6880:       __ move(A1, (int)id);
aoqi@6880:       __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), A1);
aoqi@6880:       __ should_not_reach_here();
aoqi@6880:     }
aoqi@6880:     break;
aoqi@6880:   }
aoqi@6880:   return oop_maps;
aoqi@1: }
aoqi@1: 
aoqi@1: #undef __
aoqi@1: 
aoqi@1: const char *Runtime1::pd_name_for_address(address entry) {
aoqi@1:   return "<unknown function>";
aoqi@1: }