Fri, 27 Sep 2019 11:31:13 +0800
#10071 MIPS Port of 8176100: [REDO][REDO] G1 Needs pre barrier on dereference of weak JNI handles
Summary: runtime/jni/CallWithJNIWeak/test.sh runtime/jni/ReturnJNIWeak/test.sh crash
Reviewed-by: aoqi
aoqi@1 | 1 | /* |
aoqi@1 | 2 | * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. |
aoqi@9459 | 3 | * Copyright (c) 2015, 2019, Loongson Technology. All rights reserved. |
aoqi@1 | 4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
aoqi@1 | 5 | * |
aoqi@1 | 6 | * This code is free software; you can redistribute it and/or modify it |
aoqi@1 | 7 | * under the terms of the GNU General Public License version 2 only, as |
aoqi@1 | 8 | * published by the Free Software Foundation. |
aoqi@1 | 9 | * |
aoqi@1 | 10 | * This code is distributed in the hope that it will be useful, but WITHOUT |
aoqi@1 | 11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
aoqi@1 | 12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
aoqi@1 | 13 | * version 2 for more details (a copy is included in the LICENSE file that |
aoqi@1 | 14 | * accompanied this code). |
aoqi@1 | 15 | * |
aoqi@1 | 16 | * You should have received a copy of the GNU General Public License version |
aoqi@1 | 17 | * 2 along with this work; if not, write to the Free Software Foundation, |
aoqi@1 | 18 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
aoqi@1 | 19 | * |
aoqi@1 | 20 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
aoqi@1 | 21 | * or visit www.oracle.com if you need additional information or have any |
aoqi@1 | 22 | * questions. |
aoqi@1 | 23 | * |
aoqi@1 | 24 | */ |
aoqi@1 | 25 | |
aoqi@1 | 26 | #include "precompiled.hpp" |
aoqi@1 | 27 | #include "asm/macroAssembler.hpp" |
aoqi@1 | 28 | #include "asm/macroAssembler.inline.hpp" |
aoqi@1 | 29 | #include "code/debugInfoRec.hpp" |
aoqi@1 | 30 | #include "code/icBuffer.hpp" |
aoqi@1 | 31 | #include "code/vtableStubs.hpp" |
aoqi@1 | 32 | #include "interpreter/interpreter.hpp" |
aoqi@1 | 33 | #include "oops/compiledICHolder.hpp" |
aoqi@1 | 34 | #include "prims/jvmtiRedefineClassesTrace.hpp" |
aoqi@1 | 35 | #include "runtime/sharedRuntime.hpp" |
aoqi@1 | 36 | #include "runtime/vframeArray.hpp" |
aoqi@1 | 37 | #include "vmreg_mips.inline.hpp" |
aoqi@1 | 38 | #ifdef COMPILER1 |
aoqi@1 | 39 | #include "c1/c1_Runtime1.hpp" |
aoqi@1 | 40 | #endif |
aoqi@1 | 41 | #ifdef COMPILER2 |
aoqi@1 | 42 | #include "opto/runtime.hpp" |
aoqi@1 | 43 | #endif |
aoqi@1 | 44 | |
wanghaomin@9639 | 45 | #include <alloca.h> |
wanghaomin@9639 | 46 | |
aoqi@1 | 47 | #define __ masm-> |
aoqi@6880 | 48 | |
aoqi@1 | 49 | const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; |
aoqi@1 | 50 | |
aoqi@1 | 51 | class RegisterSaver { |
aoqi@6880 | 52 | enum { FPU_regs_live = 32 }; |
aoqi@6880 | 53 | // Capture info about frame layout |
aoqi@6880 | 54 | enum layout { |
aoqi@1 | 55 | #define DEF_LAYOUT_OFFS(regname) regname ## _off, regname ## H_off, |
aoqi@6880 | 56 | DEF_LAYOUT_OFFS(for_16_bytes_aligned) |
aoqi@6880 | 57 | DEF_LAYOUT_OFFS(fpr0) |
aoqi@6880 | 58 | DEF_LAYOUT_OFFS(fpr1) |
aoqi@6880 | 59 | DEF_LAYOUT_OFFS(fpr2) |
aoqi@6880 | 60 | DEF_LAYOUT_OFFS(fpr3) |
aoqi@6880 | 61 | DEF_LAYOUT_OFFS(fpr4) |
aoqi@6880 | 62 | DEF_LAYOUT_OFFS(fpr5) |
aoqi@6880 | 63 | DEF_LAYOUT_OFFS(fpr6) |
aoqi@6880 | 64 | DEF_LAYOUT_OFFS(fpr7) |
aoqi@6880 | 65 | DEF_LAYOUT_OFFS(fpr8) |
aoqi@6880 | 66 | DEF_LAYOUT_OFFS(fpr9) |
aoqi@6880 | 67 | DEF_LAYOUT_OFFS(fpr10) |
aoqi@6880 | 68 | DEF_LAYOUT_OFFS(fpr11) |
aoqi@6880 | 69 | DEF_LAYOUT_OFFS(fpr12) |
aoqi@6880 | 70 | DEF_LAYOUT_OFFS(fpr13) |
aoqi@6880 | 71 | DEF_LAYOUT_OFFS(fpr14) |
aoqi@6880 | 72 | DEF_LAYOUT_OFFS(fpr15) |
aoqi@6880 | 73 | DEF_LAYOUT_OFFS(fpr16) |
aoqi@6880 | 74 | DEF_LAYOUT_OFFS(fpr17) |
aoqi@6880 | 75 | DEF_LAYOUT_OFFS(fpr18) |
aoqi@6880 | 76 | DEF_LAYOUT_OFFS(fpr19) |
aoqi@6880 | 77 | DEF_LAYOUT_OFFS(fpr20) |
aoqi@6880 | 78 | DEF_LAYOUT_OFFS(fpr21) |
aoqi@6880 | 79 | DEF_LAYOUT_OFFS(fpr22) |
aoqi@6880 | 80 | DEF_LAYOUT_OFFS(fpr23) |
aoqi@6880 | 81 | DEF_LAYOUT_OFFS(fpr24) |
aoqi@6880 | 82 | DEF_LAYOUT_OFFS(fpr25) |
aoqi@6880 | 83 | DEF_LAYOUT_OFFS(fpr26) |
aoqi@6880 | 84 | DEF_LAYOUT_OFFS(fpr27) |
aoqi@6880 | 85 | DEF_LAYOUT_OFFS(fpr28) |
aoqi@6880 | 86 | DEF_LAYOUT_OFFS(fpr29) |
aoqi@6880 | 87 | DEF_LAYOUT_OFFS(fpr30) |
aoqi@6880 | 88 | DEF_LAYOUT_OFFS(fpr31) |
aoqi@6880 | 89 | |
aoqi@6880 | 90 | DEF_LAYOUT_OFFS(v0) |
aoqi@6880 | 91 | DEF_LAYOUT_OFFS(v1) |
aoqi@6880 | 92 | DEF_LAYOUT_OFFS(a0) |
aoqi@6880 | 93 | DEF_LAYOUT_OFFS(a1) |
aoqi@6880 | 94 | DEF_LAYOUT_OFFS(a2) |
aoqi@6880 | 95 | DEF_LAYOUT_OFFS(a3) |
aoqi@6880 | 96 | DEF_LAYOUT_OFFS(a4) |
aoqi@6880 | 97 | DEF_LAYOUT_OFFS(a5) |
aoqi@6880 | 98 | DEF_LAYOUT_OFFS(a6) |
aoqi@6880 | 99 | DEF_LAYOUT_OFFS(a7) |
aoqi@6880 | 100 | DEF_LAYOUT_OFFS(t0) |
aoqi@6880 | 101 | DEF_LAYOUT_OFFS(t1) |
aoqi@6880 | 102 | DEF_LAYOUT_OFFS(t2) |
aoqi@6880 | 103 | DEF_LAYOUT_OFFS(t3) |
aoqi@6880 | 104 | DEF_LAYOUT_OFFS(s0) |
aoqi@6880 | 105 | DEF_LAYOUT_OFFS(s1) |
aoqi@6880 | 106 | DEF_LAYOUT_OFFS(s2) |
aoqi@6880 | 107 | DEF_LAYOUT_OFFS(s3) |
aoqi@6880 | 108 | DEF_LAYOUT_OFFS(s4) |
aoqi@6880 | 109 | DEF_LAYOUT_OFFS(s5) |
aoqi@6880 | 110 | DEF_LAYOUT_OFFS(s6) |
aoqi@6880 | 111 | DEF_LAYOUT_OFFS(s7) |
aoqi@6880 | 112 | DEF_LAYOUT_OFFS(t8) |
aoqi@6880 | 113 | DEF_LAYOUT_OFFS(t9) |
aoqi@6880 | 114 | |
aoqi@6880 | 115 | DEF_LAYOUT_OFFS(gp) |
aoqi@6880 | 116 | DEF_LAYOUT_OFFS(fp) |
aoqi@6880 | 117 | DEF_LAYOUT_OFFS(return) |
aoqi@6880 | 118 | reg_save_size |
aoqi@6880 | 119 | }; |
aoqi@1 | 120 | |
aoqi@1 | 121 | public: |
aoqi@1 | 122 | |
aoqi@6880 | 123 | static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors =false ); |
aoqi@6880 | 124 | static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false); |
aoqi@6880 | 125 | static int raOffset(void) { return return_off / 2; } |
aoqi@6880 | 126 | //Rmethod |
aoqi@6880 | 127 | static int methodOffset(void) { return s3_off / 2; } |
aoqi@6880 | 128 | |
aoqi@6880 | 129 | static int v0Offset(void) { return v0_off / 2; } |
aoqi@6880 | 130 | static int v1Offset(void) { return v1_off / 2; } |
aoqi@6880 | 131 | |
aoqi@6880 | 132 | static int fpResultOffset(void) { return fpr0_off / 2; } |
aoqi@6880 | 133 | |
aoqi@6880 | 134 | // During deoptimization only the result register need to be restored |
aoqi@6880 | 135 | // all the other values have already been extracted. |
aoqi@6880 | 136 | static void restore_result_registers(MacroAssembler* masm); |
aoqi@1 | 137 | }; |
aoqi@1 | 138 | |
aoqi@1 | 139 | OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors ) { |
aoqi@1 | 140 | |
aoqi@1 | 141 | // Always make the frame size 16-byte aligned |
aoqi@1 | 142 | int frame_size_in_bytes = round_to(additional_frame_words*wordSize + |
aoqi@1 | 143 | reg_save_size*BytesPerInt, 16); |
aoqi@1 | 144 | // OopMap frame size is in compiler stack slots (jint's) not bytes or words |
aoqi@1 | 145 | int frame_size_in_slots = frame_size_in_bytes / BytesPerInt; |
aoqi@1 | 146 | // The caller will allocate additional_frame_words |
aoqi@1 | 147 | int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt; |
aoqi@1 | 148 | // CodeBlob frame size is in words. |
aoqi@1 | 149 | int frame_size_in_words = frame_size_in_bytes / wordSize; |
aoqi@1 | 150 | *total_frame_words = frame_size_in_words; |
aoqi@1 | 151 | |
aoqi@9459 | 152 | // save registers |
aoqi@1 | 153 | |
aoqi@1 | 154 | __ daddiu(SP, SP, - reg_save_size * jintSize); |
aoqi@1 | 155 | |
aoqi@1 | 156 | __ sdc1(F0, SP, fpr0_off * jintSize); __ sdc1(F1, SP, fpr1_off * jintSize); |
aoqi@1 | 157 | __ sdc1(F2, SP, fpr2_off * jintSize); __ sdc1(F3, SP, fpr3_off * jintSize); |
aoqi@1 | 158 | __ sdc1(F4, SP, fpr4_off * jintSize); __ sdc1(F5, SP, fpr5_off * jintSize); |
aoqi@6880 | 159 | __ sdc1(F6, SP, fpr6_off * jintSize); __ sdc1(F7, SP, fpr7_off * jintSize); |
aoqi@6880 | 160 | __ sdc1(F8, SP, fpr8_off * jintSize); __ sdc1(F9, SP, fpr9_off * jintSize); |
aoqi@6880 | 161 | __ sdc1(F10, SP, fpr10_off * jintSize); __ sdc1(F11, SP, fpr11_off * jintSize); |
aoqi@6880 | 162 | __ sdc1(F12, SP, fpr12_off * jintSize); __ sdc1(F13, SP, fpr13_off * jintSize); |
aoqi@6880 | 163 | __ sdc1(F14, SP, fpr14_off * jintSize); __ sdc1(F15, SP, fpr15_off * jintSize); |
aoqi@6880 | 164 | __ sdc1(F16, SP, fpr16_off * jintSize); __ sdc1(F17, SP, fpr17_off * jintSize); |
aoqi@6880 | 165 | __ sdc1(F18, SP, fpr18_off * jintSize); __ sdc1(F19, SP, fpr19_off * jintSize); |
aoqi@6880 | 166 | __ sdc1(F20, SP, fpr20_off * jintSize); __ sdc1(F21, SP, fpr21_off * jintSize); |
aoqi@6880 | 167 | __ sdc1(F22, SP, fpr22_off * jintSize); __ sdc1(F23, SP, fpr23_off * jintSize); |
aoqi@6880 | 168 | __ sdc1(F24, SP, fpr24_off * jintSize); __ sdc1(F25, SP, fpr25_off * jintSize); |
aoqi@6880 | 169 | __ sdc1(F26, SP, fpr26_off * jintSize); __ sdc1(F27, SP, fpr27_off * jintSize); |
aoqi@6880 | 170 | __ sdc1(F28, SP, fpr28_off * jintSize); __ sdc1(F29, SP, fpr29_off * jintSize); |
aoqi@6880 | 171 | __ sdc1(F30, SP, fpr30_off * jintSize); __ sdc1(F31, SP, fpr31_off * jintSize); |
aoqi@6880 | 172 | __ sd(V0, SP, v0_off * jintSize); __ sd(V1, SP, v1_off * jintSize); |
aoqi@6880 | 173 | __ sd(A0, SP, a0_off * jintSize); __ sd(A1, SP, a1_off * jintSize); |
aoqi@6880 | 174 | __ sd(A2, SP, a2_off * jintSize); __ sd(A3, SP, a3_off * jintSize); |
aoqi@6880 | 175 | __ sd(A4, SP, a4_off * jintSize); __ sd(A5, SP, a5_off * jintSize); |
aoqi@6880 | 176 | __ sd(A6, SP, a6_off * jintSize); __ sd(A7, SP, a7_off * jintSize); |
aoqi@1 | 177 | __ sd(T0, SP, t0_off * jintSize); |
aoqi@1 | 178 | __ sd(T1, SP, t1_off * jintSize); |
aoqi@1 | 179 | __ sd(T2, SP, t2_off * jintSize); |
aoqi@1 | 180 | __ sd(T3, SP, t3_off * jintSize); |
aoqi@1 | 181 | __ sd(S0, SP, s0_off * jintSize); |
aoqi@1 | 182 | __ sd(S1, SP, s1_off * jintSize); |
aoqi@1 | 183 | __ sd(S2, SP, s2_off * jintSize); |
aoqi@1 | 184 | __ sd(S3, SP, s3_off * jintSize); |
aoqi@1 | 185 | __ sd(S4, SP, s4_off * jintSize); |
aoqi@1 | 186 | __ sd(S5, SP, s5_off * jintSize); |
aoqi@1 | 187 | __ sd(S6, SP, s6_off * jintSize); |
aoqi@1 | 188 | __ sd(S7, SP, s7_off * jintSize); |
aoqi@1 | 189 | |
aoqi@1 | 190 | __ sd(T8, SP, t8_off * jintSize); |
aoqi@1 | 191 | __ sd(T9, SP, t9_off * jintSize); |
aoqi@1 | 192 | |
aoqi@1 | 193 | __ sd(GP, SP, gp_off * jintSize); |
aoqi@1 | 194 | __ sd(FP, SP, fp_off * jintSize); |
aoqi@1 | 195 | __ sd(RA, SP, return_off * jintSize); |
aoqi@1 | 196 | __ daddi(FP, SP, fp_off * jintSize); |
aoqi@1 | 197 | |
aoqi@1 | 198 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@6880 | 199 | //OopMap* map = new OopMap( frame_words, 0 ); |
aoqi@6880 | 200 | OopMap* map = new OopMap( frame_size_in_slots, 0 ); |
aoqi@1 | 201 | |
aoqi@1 | 202 | |
aoqi@1 | 203 | //#define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words) |
aoqi@1 | 204 | #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_slots) |
aoqi@1 | 205 | map->set_callee_saved(STACK_OFFSET( v0_off), V0->as_VMReg()); |
aoqi@1 | 206 | map->set_callee_saved(STACK_OFFSET( v1_off), V1->as_VMReg()); |
aoqi@1 | 207 | map->set_callee_saved(STACK_OFFSET( a0_off), A0->as_VMReg()); |
aoqi@1 | 208 | map->set_callee_saved(STACK_OFFSET( a1_off), A1->as_VMReg()); |
aoqi@1 | 209 | map->set_callee_saved(STACK_OFFSET( a2_off), A2->as_VMReg()); |
aoqi@1 | 210 | map->set_callee_saved(STACK_OFFSET( a3_off), A3->as_VMReg()); |
aoqi@1 | 211 | map->set_callee_saved(STACK_OFFSET( a4_off), A4->as_VMReg()); |
aoqi@1 | 212 | map->set_callee_saved(STACK_OFFSET( a5_off), A5->as_VMReg()); |
aoqi@1 | 213 | map->set_callee_saved(STACK_OFFSET( a6_off), A6->as_VMReg()); |
aoqi@1 | 214 | map->set_callee_saved(STACK_OFFSET( a7_off), A7->as_VMReg()); |
aoqi@1 | 215 | map->set_callee_saved(STACK_OFFSET( t0_off), T0->as_VMReg()); |
aoqi@1 | 216 | map->set_callee_saved(STACK_OFFSET( t1_off), T1->as_VMReg()); |
aoqi@1 | 217 | map->set_callee_saved(STACK_OFFSET( t2_off), T2->as_VMReg()); |
aoqi@1 | 218 | map->set_callee_saved(STACK_OFFSET( t3_off), T3->as_VMReg()); |
aoqi@1 | 219 | map->set_callee_saved(STACK_OFFSET( s0_off), S0->as_VMReg()); |
aoqi@1 | 220 | map->set_callee_saved(STACK_OFFSET( s1_off), S1->as_VMReg()); |
aoqi@1 | 221 | map->set_callee_saved(STACK_OFFSET( s2_off), S2->as_VMReg()); |
aoqi@1 | 222 | map->set_callee_saved(STACK_OFFSET( s3_off), S3->as_VMReg()); |
aoqi@1 | 223 | map->set_callee_saved(STACK_OFFSET( s4_off), S4->as_VMReg()); |
aoqi@1 | 224 | map->set_callee_saved(STACK_OFFSET( s5_off), S5->as_VMReg()); |
aoqi@1 | 225 | map->set_callee_saved(STACK_OFFSET( s6_off), S6->as_VMReg()); |
aoqi@1 | 226 | map->set_callee_saved(STACK_OFFSET( s7_off), S7->as_VMReg()); |
aoqi@1 | 227 | map->set_callee_saved(STACK_OFFSET( t8_off), T8->as_VMReg()); |
aoqi@1 | 228 | map->set_callee_saved(STACK_OFFSET( t9_off), T9->as_VMReg()); |
aoqi@1 | 229 | map->set_callee_saved(STACK_OFFSET( gp_off), GP->as_VMReg()); |
aoqi@1 | 230 | map->set_callee_saved(STACK_OFFSET( fp_off), FP->as_VMReg()); |
aoqi@1 | 231 | map->set_callee_saved(STACK_OFFSET( return_off), RA->as_VMReg()); |
aoqi@1 | 232 | |
aoqi@1 | 233 | map->set_callee_saved(STACK_OFFSET( fpr0_off), F0->as_VMReg()); |
aoqi@1 | 234 | map->set_callee_saved(STACK_OFFSET( fpr1_off), F1->as_VMReg()); |
aoqi@1 | 235 | map->set_callee_saved(STACK_OFFSET( fpr2_off), F2->as_VMReg()); |
aoqi@1 | 236 | map->set_callee_saved(STACK_OFFSET( fpr3_off), F3->as_VMReg()); |
aoqi@1 | 237 | map->set_callee_saved(STACK_OFFSET( fpr4_off), F4->as_VMReg()); |
aoqi@1 | 238 | map->set_callee_saved(STACK_OFFSET( fpr5_off), F5->as_VMReg()); |
aoqi@1 | 239 | map->set_callee_saved(STACK_OFFSET( fpr6_off), F6->as_VMReg()); |
aoqi@1 | 240 | map->set_callee_saved(STACK_OFFSET( fpr7_off), F7->as_VMReg()); |
aoqi@1 | 241 | map->set_callee_saved(STACK_OFFSET( fpr8_off), F8->as_VMReg()); |
aoqi@1 | 242 | map->set_callee_saved(STACK_OFFSET( fpr9_off), F9->as_VMReg()); |
aoqi@1 | 243 | map->set_callee_saved(STACK_OFFSET( fpr10_off), F10->as_VMReg()); |
aoqi@1 | 244 | map->set_callee_saved(STACK_OFFSET( fpr11_off), F11->as_VMReg()); |
aoqi@1 | 245 | map->set_callee_saved(STACK_OFFSET( fpr12_off), F12->as_VMReg()); |
aoqi@1 | 246 | map->set_callee_saved(STACK_OFFSET( fpr13_off), F13->as_VMReg()); |
aoqi@1 | 247 | map->set_callee_saved(STACK_OFFSET( fpr14_off), F14->as_VMReg()); |
aoqi@1 | 248 | map->set_callee_saved(STACK_OFFSET( fpr15_off), F15->as_VMReg()); |
aoqi@1 | 249 | map->set_callee_saved(STACK_OFFSET( fpr16_off), F16->as_VMReg()); |
aoqi@1 | 250 | map->set_callee_saved(STACK_OFFSET( fpr17_off), F17->as_VMReg()); |
aoqi@1 | 251 | map->set_callee_saved(STACK_OFFSET( fpr18_off), F18->as_VMReg()); |
aoqi@1 | 252 | map->set_callee_saved(STACK_OFFSET( fpr19_off), F19->as_VMReg()); |
aoqi@1 | 253 | map->set_callee_saved(STACK_OFFSET( fpr20_off), F20->as_VMReg()); |
aoqi@1 | 254 | map->set_callee_saved(STACK_OFFSET( fpr21_off), F21->as_VMReg()); |
aoqi@1 | 255 | map->set_callee_saved(STACK_OFFSET( fpr22_off), F22->as_VMReg()); |
aoqi@1 | 256 | map->set_callee_saved(STACK_OFFSET( fpr23_off), F23->as_VMReg()); |
aoqi@1 | 257 | map->set_callee_saved(STACK_OFFSET( fpr24_off), F24->as_VMReg()); |
aoqi@1 | 258 | map->set_callee_saved(STACK_OFFSET( fpr25_off), F25->as_VMReg()); |
aoqi@1 | 259 | map->set_callee_saved(STACK_OFFSET( fpr26_off), F26->as_VMReg()); |
aoqi@1 | 260 | map->set_callee_saved(STACK_OFFSET( fpr27_off), F27->as_VMReg()); |
aoqi@1 | 261 | map->set_callee_saved(STACK_OFFSET( fpr28_off), F28->as_VMReg()); |
aoqi@1 | 262 | map->set_callee_saved(STACK_OFFSET( fpr29_off), F29->as_VMReg()); |
aoqi@1 | 263 | map->set_callee_saved(STACK_OFFSET( fpr30_off), F30->as_VMReg()); |
aoqi@1 | 264 | map->set_callee_saved(STACK_OFFSET( fpr31_off), F31->as_VMReg()); |
aoqi@1 | 265 | |
aoqi@1 | 266 | #undef STACK_OFFSET |
aoqi@1 | 267 | return map; |
aoqi@1 | 268 | } |
aoqi@1 | 269 | |
aoqi@1 | 270 | |
aoqi@1 | 271 | // Pop the current frame and restore all the registers that we |
aoqi@1 | 272 | // saved. |
aoqi@1 | 273 | void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) { |
aoqi@1 | 274 | __ ldc1(F0, SP, fpr0_off * jintSize); __ ldc1(F1, SP, fpr1_off * jintSize); |
aoqi@1 | 275 | __ ldc1(F2, SP, fpr2_off * jintSize); __ ldc1(F3, SP, fpr3_off * jintSize); |
aoqi@1 | 276 | __ ldc1(F4, SP, fpr4_off * jintSize); __ ldc1(F5, SP, fpr5_off * jintSize); |
aoqi@6880 | 277 | __ ldc1(F6, SP, fpr6_off * jintSize); __ ldc1(F7, SP, fpr7_off * jintSize); |
aoqi@6880 | 278 | __ ldc1(F8, SP, fpr8_off * jintSize); __ ldc1(F9, SP, fpr9_off * jintSize); |
aoqi@6880 | 279 | __ ldc1(F10, SP, fpr10_off * jintSize); __ ldc1(F11, SP, fpr11_off * jintSize); |
aoqi@6880 | 280 | __ ldc1(F12, SP, fpr12_off * jintSize); __ ldc1(F13, SP, fpr13_off * jintSize); |
aoqi@6880 | 281 | __ ldc1(F14, SP, fpr14_off * jintSize); __ ldc1(F15, SP, fpr15_off * jintSize); |
aoqi@6880 | 282 | __ ldc1(F16, SP, fpr16_off * jintSize); __ ldc1(F17, SP, fpr17_off * jintSize); |
aoqi@6880 | 283 | __ ldc1(F18, SP, fpr18_off * jintSize); __ ldc1(F19, SP, fpr19_off * jintSize); |
aoqi@6880 | 284 | __ ldc1(F20, SP, fpr20_off * jintSize); __ ldc1(F21, SP, fpr21_off * jintSize); |
aoqi@6880 | 285 | __ ldc1(F22, SP, fpr22_off * jintSize); __ ldc1(F23, SP, fpr23_off * jintSize); |
aoqi@6880 | 286 | __ ldc1(F24, SP, fpr24_off * jintSize); __ ldc1(F25, SP, fpr25_off * jintSize); |
aoqi@6880 | 287 | __ ldc1(F26, SP, fpr26_off * jintSize); __ ldc1(F27, SP, fpr27_off * jintSize); |
aoqi@6880 | 288 | __ ldc1(F28, SP, fpr28_off * jintSize); __ ldc1(F29, SP, fpr29_off * jintSize); |
aoqi@6880 | 289 | __ ldc1(F30, SP, fpr30_off * jintSize); __ ldc1(F31, SP, fpr31_off * jintSize); |
aoqi@6880 | 290 | |
aoqi@6880 | 291 | __ ld(V0, SP, v0_off * jintSize); __ ld(V1, SP, v1_off * jintSize); |
aoqi@6880 | 292 | __ ld(A0, SP, a0_off * jintSize); __ ld(A1, SP, a1_off * jintSize); |
aoqi@6880 | 293 | __ ld(A2, SP, a2_off * jintSize); __ ld(A3, SP, a3_off * jintSize); |
aoqi@6880 | 294 | __ ld(A4, SP, a4_off * jintSize); __ ld(A5, SP, a5_off * jintSize); |
aoqi@6880 | 295 | __ ld(A6, SP, a6_off * jintSize); __ ld(A7, SP, a7_off * jintSize); |
aoqi@1 | 296 | __ ld(T0, SP, t0_off * jintSize); |
aoqi@1 | 297 | __ ld(T1, SP, t1_off * jintSize); |
aoqi@1 | 298 | __ ld(T2, SP, t2_off * jintSize); |
aoqi@1 | 299 | __ ld(T3, SP, t3_off * jintSize); |
aoqi@1 | 300 | __ ld(S0, SP, s0_off * jintSize); |
aoqi@1 | 301 | __ ld(S1, SP, s1_off * jintSize); |
aoqi@1 | 302 | __ ld(S2, SP, s2_off * jintSize); |
aoqi@1 | 303 | __ ld(S3, SP, s3_off * jintSize); |
aoqi@1 | 304 | __ ld(S4, SP, s4_off * jintSize); |
aoqi@1 | 305 | __ ld(S5, SP, s5_off * jintSize); |
aoqi@1 | 306 | __ ld(S6, SP, s6_off * jintSize); |
aoqi@1 | 307 | __ ld(S7, SP, s7_off * jintSize); |
aoqi@1 | 308 | |
aoqi@1 | 309 | __ ld(T8, SP, t8_off * jintSize); |
aoqi@1 | 310 | __ ld(T9, SP, t9_off * jintSize); |
aoqi@1 | 311 | |
aoqi@1 | 312 | __ ld(GP, SP, gp_off * jintSize); |
aoqi@1 | 313 | __ ld(FP, SP, fp_off * jintSize); |
aoqi@1 | 314 | __ ld(RA, SP, return_off * jintSize); |
aoqi@1 | 315 | |
aoqi@1 | 316 | __ addiu(SP, SP, reg_save_size * jintSize); |
aoqi@1 | 317 | } |
aoqi@1 | 318 | |
aoqi@1 | 319 | // Pop the current frame and restore the registers that might be holding |
aoqi@1 | 320 | // a result. |
aoqi@1 | 321 | // FIXME, if the result is float? |
aoqi@1 | 322 | void RegisterSaver::restore_result_registers(MacroAssembler* masm) { |
aoqi@6880 | 323 | |
aoqi@1 | 324 | // Just restore result register. Only used by deoptimization. By |
aoqi@1 | 325 | // now any callee save register that needs to be restore to a c2 |
aoqi@1 | 326 | // caller of the deoptee has been extracted into the vframeArray |
aoqi@1 | 327 | // and will be stuffed into the c2i adapter we create for later |
aoqi@1 | 328 | // restoration so only result registers need to be restored here. |
aoqi@6880 | 329 | |
aoqi@1 | 330 | __ ld(V0, SP, v0_off * jintSize); |
aoqi@1 | 331 | __ ld(V1, SP, v1_off * jintSize); |
aoqi@6880 | 332 | __ addiu(SP, SP, return_off * jintSize); |
aoqi@1 | 333 | } |
aoqi@1 | 334 | |
aoqi@6880 | 335 | // Is vector's size (in bytes) bigger than a size saved by default? |
aoqi@6880 | 336 | // 16 bytes XMM registers are saved by default using fxsave/fxrstor instructions. |
aoqi@6880 | 337 | bool SharedRuntime::is_wide_vector(int size) { |
aoqi@6880 | 338 | return size > 16; |
aoqi@6880 | 339 | } |
aoqi@1 | 340 | |
aoqi@1 | 341 | // The java_calling_convention describes stack locations as ideal slots on |
aoqi@1 | 342 | // a frame with no abi restrictions. Since we must observe abi restrictions |
aoqi@1 | 343 | // (like the placement of the register window) the slots must be biased by |
aoqi@1 | 344 | // the following value. |
aoqi@1 | 345 | |
aoqi@6880 | 346 | static int reg2offset_in(VMReg r) { |
aoqi@9459 | 347 | // Account for saved fp and return address |
aoqi@6880 | 348 | // This should really be in_preserve_stack_slots |
aoqi@6880 | 349 | return (r->reg2stack() + 2 * VMRegImpl::slots_per_word) * VMRegImpl::stack_slot_size; // + 2 * VMRegImpl::stack_slot_size); |
aoqi@1 | 350 | } |
aoqi@1 | 351 | |
aoqi@6880 | 352 | static int reg2offset_out(VMReg r) { |
aoqi@6880 | 353 | return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; |
aoqi@1 | 354 | } |
aoqi@1 | 355 | |
aoqi@1 | 356 | // --------------------------------------------------------------------------- |
aoqi@1 | 357 | // Read the array of BasicTypes from a signature, and compute where the |
aoqi@1 | 358 | // arguments should go. Values in the VMRegPair regs array refer to 4-byte |
aoqi@1 | 359 | // quantities. Values less than SharedInfo::stack0 are registers, those above |
aoqi@1 | 360 | // refer to 4-byte stack slots. All stack slots are based off of the stack pointer |
aoqi@1 | 361 | // as framesizes are fixed. |
aoqi@1 | 362 | // VMRegImpl::stack0 refers to the first slot 0(sp). |
aoqi@1 | 363 | // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register |
aoqi@1 | 364 | // up to RegisterImpl::number_of_registers) are the 32-bit |
aoqi@1 | 365 | // integer registers. |
aoqi@1 | 366 | |
aoqi@1 | 367 | // Pass first five oop/int args in registers T0, A0 - A3. |
aoqi@1 | 368 | // Pass float/double/long args in stack. |
aoqi@1 | 369 | // Doubles have precedence, so if you pass a mix of floats and doubles |
aoqi@1 | 370 | // the doubles will grab the registers before the floats will. |
aoqi@1 | 371 | |
aoqi@1 | 372 | // Note: the INPUTS in sig_bt are in units of Java argument words, which are |
aoqi@1 | 373 | // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit |
aoqi@9459 | 374 | // units regardless of build. |
aoqi@1 | 375 | |
aoqi@1 | 376 | |
aoqi@1 | 377 | // --------------------------------------------------------------------------- |
aoqi@1 | 378 | // The compiled Java calling convention. |
aoqi@1 | 379 | // Pass first five oop/int args in registers T0, A0 - A3. |
aoqi@1 | 380 | // Pass float/double/long args in stack. |
aoqi@1 | 381 | // Doubles have precedence, so if you pass a mix of floats and doubles |
aoqi@1 | 382 | // the doubles will grab the registers before the floats will. |
aoqi@1 | 383 | |
aoqi@1 | 384 | int SharedRuntime::java_calling_convention(const BasicType *sig_bt, |
aoqi@1 | 385 | VMRegPair *regs, |
aoqi@1 | 386 | int total_args_passed, |
aoqi@1 | 387 | int is_outgoing) { |
aoqi@1 | 388 | |
aoqi@1 | 389 | // Create the mapping between argument positions and |
aoqi@1 | 390 | // registers. |
aoqi@1 | 391 | //static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = { |
aoqi@1 | 392 | static const Register INT_ArgReg[Argument::n_register_parameters + 1] = { |
aoqi@1 | 393 | T0, A0, A1, A2, A3, A4, A5, A6, A7 |
aoqi@1 | 394 | }; |
aoqi@1 | 395 | //static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = { |
aoqi@1 | 396 | static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters] = { |
aoqi@1 | 397 | F12, F13, F14, F15, F16, F17, F18, F19 |
aoqi@1 | 398 | }; |
aoqi@1 | 399 | |
aoqi@1 | 400 | |
aoqi@1 | 401 | uint args = 0; |
aoqi@1 | 402 | uint stk_args = 0; // inc by 2 each time |
aoqi@1 | 403 | |
aoqi@1 | 404 | for (int i = 0; i < total_args_passed; i++) { |
aoqi@1 | 405 | switch (sig_bt[i]) { |
aoqi@1 | 406 | case T_VOID: |
aoqi@1 | 407 | // halves of T_LONG or T_DOUBLE |
aoqi@1 | 408 | assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
aoqi@1 | 409 | regs[i].set_bad(); |
aoqi@1 | 410 | break; |
aoqi@1 | 411 | case T_BOOLEAN: |
aoqi@1 | 412 | case T_CHAR: |
aoqi@1 | 413 | case T_BYTE: |
aoqi@1 | 414 | case T_SHORT: |
aoqi@1 | 415 | case T_INT: |
aoqi@1 | 416 | if (args < Argument::n_register_parameters) { |
aoqi@1 | 417 | regs[i].set1(INT_ArgReg[args++]->as_VMReg()); |
aoqi@1 | 418 | } else { |
aoqi@1 | 419 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
aoqi@1 | 420 | stk_args += 2; |
aoqi@1 | 421 | } |
aoqi@1 | 422 | break; |
aoqi@1 | 423 | case T_LONG: |
aoqi@1 | 424 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
aoqi@1 | 425 | // fall through |
aoqi@1 | 426 | case T_OBJECT: |
aoqi@1 | 427 | case T_ARRAY: |
aoqi@1 | 428 | case T_ADDRESS: |
aoqi@1 | 429 | if (args < Argument::n_register_parameters) { |
aoqi@1 | 430 | regs[i].set2(INT_ArgReg[args++]->as_VMReg()); |
aoqi@1 | 431 | } else { |
aoqi@1 | 432 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
aoqi@1 | 433 | stk_args += 2; |
aoqi@1 | 434 | } |
aoqi@1 | 435 | break; |
aoqi@1 | 436 | case T_FLOAT: |
aoqi@1 | 437 | if (args < Argument::n_float_register_parameters) { |
aoqi@1 | 438 | regs[i].set1(FP_ArgReg[args++]->as_VMReg()); |
aoqi@1 | 439 | } else { |
aoqi@1 | 440 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
aoqi@1 | 441 | stk_args += 2; |
aoqi@1 | 442 | } |
aoqi@1 | 443 | break; |
aoqi@1 | 444 | case T_DOUBLE: |
aoqi@1 | 445 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
aoqi@1 | 446 | if (args < Argument::n_float_register_parameters) { |
aoqi@1 | 447 | regs[i].set2(FP_ArgReg[args++]->as_VMReg()); |
aoqi@1 | 448 | } else { |
aoqi@1 | 449 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
aoqi@1 | 450 | stk_args += 2; |
aoqi@1 | 451 | } |
aoqi@1 | 452 | break; |
aoqi@1 | 453 | default: |
aoqi@1 | 454 | ShouldNotReachHere(); |
aoqi@1 | 455 | break; |
aoqi@1 | 456 | } |
aoqi@1 | 457 | } |
aoqi@1 | 458 | |
aoqi@1 | 459 | return round_to(stk_args, 2); |
aoqi@1 | 460 | } |
aoqi@1 | 461 | |
aoqi@1 | 462 | // Helper class mostly to avoid passing masm everywhere, and handle store |
aoqi@1 | 463 | // displacement overflow logic for LP64 |
aoqi@1 | 464 | class AdapterGenerator { |
aoqi@1 | 465 | MacroAssembler *masm; |
aoqi@1 | 466 | #ifdef _LP64 |
aoqi@1 | 467 | Register Rdisp; |
aoqi@1 | 468 | void set_Rdisp(Register r) { Rdisp = r; } |
aoqi@1 | 469 | #endif // _LP64 |
aoqi@1 | 470 | |
aoqi@1 | 471 | void patch_callers_callsite(); |
aoqi@1 | 472 | |
aoqi@1 | 473 | // base+st_off points to top of argument |
aoqi@1 | 474 | int arg_offset(const int st_off) { return st_off; } |
aoqi@1 | 475 | int next_arg_offset(const int st_off) { |
aoqi@1 | 476 | return st_off - Interpreter::stackElementSize; |
aoqi@1 | 477 | } |
aoqi@1 | 478 | |
aoqi@1 | 479 | #ifdef _LP64 |
aoqi@1 | 480 | // On _LP64 argument slot values are loaded first into a register |
aoqi@1 | 481 | // because they might not fit into displacement. |
aoqi@1 | 482 | Register arg_slot(const int st_off); |
aoqi@1 | 483 | Register next_arg_slot(const int st_off); |
aoqi@1 | 484 | #else |
aoqi@1 | 485 | int arg_slot(const int st_off) { return arg_offset(st_off); } |
aoqi@1 | 486 | int next_arg_slot(const int st_off) { return next_arg_offset(st_off); } |
aoqi@1 | 487 | #endif // _LP64 |
aoqi@1 | 488 | |
aoqi@1 | 489 | // Stores long into offset pointed to by base |
aoqi@1 | 490 | void store_c2i_long(Register r, Register base, |
aoqi@1 | 491 | const int st_off, bool is_stack); |
aoqi@1 | 492 | void store_c2i_object(Register r, Register base, |
aoqi@1 | 493 | const int st_off); |
aoqi@1 | 494 | void store_c2i_int(Register r, Register base, |
aoqi@1 | 495 | const int st_off); |
aoqi@1 | 496 | void store_c2i_double(VMReg r_2, |
aoqi@1 | 497 | VMReg r_1, Register base, const int st_off); |
aoqi@1 | 498 | void store_c2i_float(FloatRegister f, Register base, |
aoqi@1 | 499 | const int st_off); |
aoqi@1 | 500 | |
aoqi@1 | 501 | public: |
aoqi@1 | 502 | //void tag_stack(const BasicType sig, int st_off); |
aoqi@1 | 503 | void gen_c2i_adapter(int total_args_passed, |
aoqi@1 | 504 | // VMReg max_arg, |
aoqi@1 | 505 | int comp_args_on_stack, // VMRegStackSlots |
aoqi@1 | 506 | const BasicType *sig_bt, |
aoqi@1 | 507 | const VMRegPair *regs, |
aoqi@1 | 508 | Label& skip_fixup); |
aoqi@1 | 509 | void gen_i2c_adapter(int total_args_passed, |
aoqi@1 | 510 | // VMReg max_arg, |
aoqi@1 | 511 | int comp_args_on_stack, // VMRegStackSlots |
aoqi@1 | 512 | const BasicType *sig_bt, |
aoqi@1 | 513 | const VMRegPair *regs); |
aoqi@1 | 514 | |
aoqi@1 | 515 | AdapterGenerator(MacroAssembler *_masm) : masm(_masm) {} |
aoqi@1 | 516 | }; |
aoqi@1 | 517 | |
aoqi@1 | 518 | |
aoqi@1 | 519 | // Patch the callers callsite with entry to compiled code if it exists. |
aoqi@1 | 520 | void AdapterGenerator::patch_callers_callsite() { |
aoqi@6880 | 521 | Label L; |
aoqi@6880 | 522 | __ verify_oop(Rmethod); |
aoqi@6880 | 523 | __ ld_ptr(AT, Rmethod, in_bytes(Method::code_offset())); |
huangjia@9645 | 524 | __ beq(AT, R0, L); |
aoqi@6880 | 525 | __ delayed()->nop(); |
aoqi@6880 | 526 | // Schedule the branch target address early. |
aoqi@6880 | 527 | // Call into the VM to patch the caller, then jump to compiled callee |
aoqi@9459 | 528 | // V0 isn't live so capture return address while we easily can |
aoqi@6880 | 529 | __ move(V0, RA); |
aoqi@6880 | 530 | |
aoqi@6880 | 531 | __ pushad(); |
aoqi@1 | 532 | #ifdef COMPILER2 |
aoqi@6880 | 533 | // C2 may leave the stack dirty if not in SSE2+ mode |
aoqi@6880 | 534 | __ empty_FPU_stack(); |
aoqi@9459 | 535 | #endif |
aoqi@1 | 536 | |
aoqi@6880 | 537 | // VM needs caller's callsite |
aoqi@6880 | 538 | // VM needs target method |
aoqi@6880 | 539 | |
aoqi@6880 | 540 | __ move(A0, Rmethod); |
aoqi@6880 | 541 | __ move(A1, V0); |
aoqi@9459 | 542 | // we should preserve the return address |
aoqi@6880 | 543 | __ verify_oop(Rmethod); |
aoqi@6880 | 544 | __ move(S0, SP); |
aoqi@6880 | 545 | __ move(AT, -(StackAlignmentInBytes)); // align the stack |
aoqi@6880 | 546 | __ andr(SP, SP, AT); |
aoqi@6880 | 547 | __ call(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite), |
aoqi@6880 | 548 | relocInfo::runtime_call_type); |
aoqi@6880 | 549 | |
aoqi@6880 | 550 | __ delayed()->nop(); |
aoqi@6880 | 551 | __ move(SP, S0); |
aoqi@6880 | 552 | __ popad(); |
aoqi@6880 | 553 | __ bind(L); |
aoqi@1 | 554 | } |
aoqi@1 | 555 | |
aoqi@1 | 556 | #ifdef _LP64 |
aoqi@1 | 557 | Register AdapterGenerator::arg_slot(const int st_off) { |
aoqi@6880 | 558 | Unimplemented(); |
aoqi@1 | 559 | } |
aoqi@1 | 560 | |
aoqi@1 | 561 | Register AdapterGenerator::next_arg_slot(const int st_off){ |
aoqi@6880 | 562 | Unimplemented(); |
aoqi@1 | 563 | } |
aoqi@1 | 564 | #endif // _LP64 |
aoqi@1 | 565 | |
aoqi@1 | 566 | // Stores long into offset pointed to by base |
aoqi@1 | 567 | void AdapterGenerator::store_c2i_long(Register r, Register base, |
aoqi@1 | 568 | const int st_off, bool is_stack) { |
aoqi@6880 | 569 | Unimplemented(); |
aoqi@1 | 570 | } |
aoqi@1 | 571 | |
aoqi@1 | 572 | void AdapterGenerator::store_c2i_object(Register r, Register base, |
aoqi@6880 | 573 | const int st_off) { |
aoqi@6880 | 574 | Unimplemented(); |
aoqi@1 | 575 | } |
aoqi@1 | 576 | |
aoqi@1 | 577 | void AdapterGenerator::store_c2i_int(Register r, Register base, |
aoqi@6880 | 578 | const int st_off) { |
aoqi@6880 | 579 | Unimplemented(); |
aoqi@1 | 580 | } |
aoqi@1 | 581 | |
aoqi@1 | 582 | // Stores into offset pointed to by base |
aoqi@1 | 583 | void AdapterGenerator::store_c2i_double(VMReg r_2, |
aoqi@1 | 584 | VMReg r_1, Register base, const int st_off) { |
aoqi@6880 | 585 | Unimplemented(); |
aoqi@1 | 586 | } |
aoqi@1 | 587 | |
aoqi@1 | 588 | void AdapterGenerator::store_c2i_float(FloatRegister f, Register base, |
aoqi@1 | 589 | const int st_off) { |
aoqi@6880 | 590 | Unimplemented(); |
aoqi@1 | 591 | } |
aoqi@1 | 592 | |
aoqi@1 | 593 | void AdapterGenerator::gen_c2i_adapter( |
aoqi@1 | 594 | int total_args_passed, |
aoqi@1 | 595 | // VMReg max_arg, |
aoqi@1 | 596 | int comp_args_on_stack, // VMRegStackSlots |
aoqi@1 | 597 | const BasicType *sig_bt, |
aoqi@1 | 598 | const VMRegPair *regs, |
aoqi@1 | 599 | Label& skip_fixup) { |
aoqi@1 | 600 | |
aoqi@1 | 601 | // Before we get into the guts of the C2I adapter, see if we should be here |
aoqi@1 | 602 | // at all. We've come from compiled code and are attempting to jump to the |
aoqi@1 | 603 | // interpreter, which means the caller made a static call to get here |
aoqi@1 | 604 | // (vcalls always get a compiled target if there is one). Check for a |
aoqi@1 | 605 | // compiled target. If there is one, we need to patch the caller's call. |
aoqi@1 | 606 | // However we will run interpreted if we come thru here. The next pass |
aoqi@1 | 607 | // thru the call site will run compiled. If we ran compiled here then |
aoqi@1 | 608 | // we can (theorectically) do endless i2c->c2i->i2c transitions during |
aoqi@1 | 609 | // deopt/uncommon trap cycles. If we always go interpreted here then |
aoqi@1 | 610 | // we can have at most one and don't need to play any tricks to keep |
aoqi@1 | 611 | // from endlessly growing the stack. |
aoqi@1 | 612 | // |
aoqi@1 | 613 | // Actually if we detected that we had an i2c->c2i transition here we |
aoqi@1 | 614 | // ought to be able to reset the world back to the state of the interpreted |
aoqi@1 | 615 | // call and not bother building another interpreter arg area. We don't |
aoqi@1 | 616 | // do that at this point. |
aoqi@1 | 617 | |
aoqi@6880 | 618 | patch_callers_callsite(); |
aoqi@6880 | 619 | |
aoqi@6880 | 620 | __ bind(skip_fixup); |
aoqi@1 | 621 | |
aoqi@1 | 622 | #ifdef COMPILER2 |
aoqi@6880 | 623 | __ empty_FPU_stack(); |
aoqi@9459 | 624 | #endif |
aoqi@6880 | 625 | //this is for native ? |
aoqi@6880 | 626 | // Since all args are passed on the stack, total_args_passed * interpreter_ |
aoqi@6880 | 627 | // stack_element_size is the |
aoqi@6880 | 628 | // space we need. |
aoqi@6880 | 629 | int extraspace = total_args_passed * Interpreter::stackElementSize; |
aoqi@6880 | 630 | |
aoqi@6880 | 631 | // stack is aligned, keep it that way |
aoqi@6880 | 632 | extraspace = round_to(extraspace, 2*wordSize); |
aoqi@6880 | 633 | |
aoqi@6880 | 634 | // Get return address |
aoqi@6880 | 635 | __ move(V0, RA); |
aoqi@6880 | 636 | // set senderSP value |
aoqi@6880 | 637 | //refer to interpreter_mips.cpp:generate_asm_entry |
aoqi@6880 | 638 | __ move(Rsender, SP); |
aoqi@6880 | 639 | __ addi(SP, SP, -extraspace); |
aoqi@6880 | 640 | |
aoqi@6880 | 641 | // Now write the args into the outgoing interpreter space |
aoqi@6880 | 642 | for (int i = 0; i < total_args_passed; i++) { |
aoqi@6880 | 643 | if (sig_bt[i] == T_VOID) { |
aoqi@6880 | 644 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
aoqi@6880 | 645 | continue; |
aoqi@6880 | 646 | } |
aoqi@6880 | 647 | |
aoqi@6880 | 648 | // st_off points to lowest address on stack. |
aoqi@6880 | 649 | int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize; |
aoqi@6880 | 650 | // Say 4 args: |
aoqi@6880 | 651 | // i st_off |
aoqi@6880 | 652 | // 0 12 T_LONG |
aoqi@6880 | 653 | // 1 8 T_VOID |
aoqi@6880 | 654 | // 2 4 T_OBJECT |
aoqi@6880 | 655 | // 3 0 T_BOOL |
aoqi@6880 | 656 | VMReg r_1 = regs[i].first(); |
aoqi@6880 | 657 | VMReg r_2 = regs[i].second(); |
aoqi@6880 | 658 | if (!r_1->is_valid()) { |
aoqi@6880 | 659 | assert(!r_2->is_valid(), ""); |
aoqi@6880 | 660 | continue; |
aoqi@6880 | 661 | } |
aoqi@6880 | 662 | if (r_1->is_stack()) { |
aoqi@6880 | 663 | // memory to memory use fpu stack top |
aoqi@6880 | 664 | int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; |
aoqi@6880 | 665 | if (!r_2->is_valid()) { |
aoqi@6880 | 666 | __ ld_ptr(AT, SP, ld_off); |
aoqi@6880 | 667 | __ st_ptr(AT, SP, st_off); |
aoqi@6880 | 668 | |
aoqi@6880 | 669 | } else { |
aoqi@6880 | 670 | |
aoqi@6880 | 671 | |
aoqi@6880 | 672 | int next_off = st_off - Interpreter::stackElementSize; |
aoqi@6880 | 673 | __ ld_ptr(AT, SP, ld_off); |
aoqi@6880 | 674 | __ st_ptr(AT, SP, st_off); |
aoqi@6880 | 675 | |
aoqi@9459 | 676 | // Ref to is_Register condition |
aoqi@6880 | 677 | if(sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) |
huangjia@9645 | 678 | __ st_ptr(AT, SP, st_off - 8); |
aoqi@6880 | 679 | } |
aoqi@6880 | 680 | } else if (r_1->is_Register()) { |
aoqi@6880 | 681 | Register r = r_1->as_Register(); |
aoqi@6880 | 682 | if (!r_2->is_valid()) { |
huangjia@9645 | 683 | __ sd(r, SP, st_off); |
aoqi@6880 | 684 | } else { |
aoqi@6880 | 685 | //FIXME, mips will not enter here |
aoqi@6880 | 686 | // long/double in gpr |
huangjia@9645 | 687 | __ sd(r, SP, st_off); |
huangjia@9645 | 688 | // In [java/util/zip/ZipFile.java] |
aoqi@9459 | 689 | // |
aoqi@9459 | 690 | // private static native long open(String name, int mode, long lastModified); |
aoqi@9459 | 691 | // private static native int getTotal(long jzfile); |
aoqi@9459 | 692 | // |
aoqi@9459 | 693 | // We need to transfer T_LONG paramenters from a compiled method to a native method. |
aoqi@9459 | 694 | // It's a complex process: |
aoqi@9459 | 695 | // |
aoqi@9459 | 696 | // Caller -> lir_static_call -> gen_resolve_stub |
aoqi@9459 | 697 | // -> -- resolve_static_call_C |
aoqi@9459 | 698 | // `- gen_c2i_adapter() [*] |
aoqi@9459 | 699 | // | |
aoqi@9459 | 700 | // `- AdapterHandlerLibrary::get_create_apapter_index |
aoqi@9459 | 701 | // -> generate_native_entry |
aoqi@9459 | 702 | // -> InterpreterRuntime::SignatureHandlerGenerator::pass_long [**] |
aoqi@9459 | 703 | // |
aoqi@9459 | 704 | // In [**], T_Long parameter is stored in stack as: |
aoqi@9459 | 705 | // |
aoqi@9459 | 706 | // (high) |
aoqi@9459 | 707 | // | | |
aoqi@9459 | 708 | // ----------- |
aoqi@9459 | 709 | // | 8 bytes | |
aoqi@9459 | 710 | // | (void) | |
aoqi@9459 | 711 | // ----------- |
aoqi@9459 | 712 | // | 8 bytes | |
aoqi@9459 | 713 | // | (long) | |
aoqi@9459 | 714 | // ----------- |
aoqi@9459 | 715 | // | | |
aoqi@9459 | 716 | // (low) |
aoqi@9459 | 717 | // |
aoqi@9459 | 718 | // However, the sequence is reversed here: |
aoqi@9459 | 719 | // |
aoqi@9459 | 720 | // (high) |
aoqi@9459 | 721 | // | | |
aoqi@9459 | 722 | // ----------- |
aoqi@9459 | 723 | // | 8 bytes | |
aoqi@9459 | 724 | // | (long) | |
aoqi@9459 | 725 | // ----------- |
aoqi@9459 | 726 | // | 8 bytes | |
aoqi@9459 | 727 | // | (void) | |
aoqi@9459 | 728 | // ----------- |
aoqi@9459 | 729 | // | | |
aoqi@9459 | 730 | // (low) |
aoqi@9459 | 731 | // |
aoqi@9459 | 732 | // So I stored another 8 bytes in the T_VOID slot. It then can be accessed from generate_native_entry(). |
aoqi@9459 | 733 | // |
aoqi@6880 | 734 | if (sig_bt[i] == T_LONG) |
huangjia@9645 | 735 | __ sd(r, SP, st_off - 8); |
aoqi@6880 | 736 | } |
aoqi@6880 | 737 | } else if (r_1->is_FloatRegister()) { |
aoqi@6880 | 738 | assert(sig_bt[i] == T_FLOAT || sig_bt[i] == T_DOUBLE, "Must be a float register"); |
aoqi@6880 | 739 | |
aoqi@6880 | 740 | FloatRegister fr = r_1->as_FloatRegister(); |
aoqi@6880 | 741 | if (sig_bt[i] == T_FLOAT) |
huangjia@9645 | 742 | __ swc1(fr, SP, st_off); |
aoqi@6880 | 743 | else { |
huangjia@9645 | 744 | __ sdc1(fr, SP, st_off); |
huangjia@9645 | 745 | __ sdc1(fr, SP, st_off - 8); // T_DOUBLE needs two slots |
aoqi@6880 | 746 | } |
aoqi@6880 | 747 | } |
aoqi@6880 | 748 | } |
aoqi@6880 | 749 | |
aoqi@6880 | 750 | // Schedule the branch target address early. |
huangjia@9645 | 751 | __ ld_ptr(AT, Rmethod, in_bytes(Method::interpreter_entry_offset()) ); |
aoqi@6880 | 752 | // And repush original return address |
aoqi@6880 | 753 | __ move(RA, V0); |
aoqi@6880 | 754 | __ jr (AT); |
aoqi@6880 | 755 | __ delayed()->nop(); |
aoqi@1 | 756 | } |
aoqi@1 | 757 | |
aoqi@1 | 758 | void AdapterGenerator::gen_i2c_adapter( |
aoqi@6880 | 759 | int total_args_passed, |
aoqi@6880 | 760 | // VMReg max_arg, |
aoqi@6880 | 761 | int comp_args_on_stack, // VMRegStackSlots |
aoqi@6880 | 762 | const BasicType *sig_bt, |
aoqi@6880 | 763 | const VMRegPair *regs) { |
aoqi@1 | 764 | |
aoqi@1 | 765 | // Generate an I2C adapter: adjust the I-frame to make space for the C-frame |
aoqi@1 | 766 | // layout. Lesp was saved by the calling I-frame and will be restored on |
aoqi@1 | 767 | // return. Meanwhile, outgoing arg space is all owned by the callee |
aoqi@1 | 768 | // C-frame, so we can mangle it at will. After adjusting the frame size, |
aoqi@1 | 769 | // hoist register arguments and repack other args according to the compiled |
aoqi@1 | 770 | // code convention. Finally, end in a jump to the compiled code. The entry |
aoqi@1 | 771 | // point address is the start of the buffer. |
aoqi@1 | 772 | |
aoqi@1 | 773 | // We will only enter here from an interpreted frame and never from after |
aoqi@1 | 774 | // passing thru a c2i. Azul allowed this but we do not. If we lose the |
aoqi@1 | 775 | // race and use a c2i we will remain interpreted for the race loser(s). |
aoqi@1 | 776 | // This removes all sorts of headaches on the mips side and also eliminates |
aoqi@1 | 777 | // the possibility of having c2i -> i2c -> c2i -> ... endless transitions. |
aoqi@1 | 778 | |
aoqi@1 | 779 | |
aoqi@1 | 780 | __ move(T9, SP); |
aoqi@1 | 781 | |
aoqi@1 | 782 | // Cut-out for having no stack args. Since up to 2 int/oop args are passed |
aoqi@1 | 783 | // in registers, we will occasionally have no stack args. |
aoqi@1 | 784 | int comp_words_on_stack = 0; |
aoqi@1 | 785 | if (comp_args_on_stack) { |
aoqi@1 | 786 | // Sig words on the stack are greater-than VMRegImpl::stack0. Those in |
aoqi@1 | 787 | // registers are below. By subtracting stack0, we either get a negative |
aoqi@1 | 788 | // number (all values in registers) or the maximum stack slot accessed. |
aoqi@1 | 789 | // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg); |
aoqi@1 | 790 | // Convert 4-byte stack slots to words. |
aoqi@1 | 791 | // did mips need round? FIXME aoqi |
aoqi@1 | 792 | comp_words_on_stack = round_to(comp_args_on_stack*4, wordSize)>>LogBytesPerWord; |
aoqi@1 | 793 | // Round up to miminum stack alignment, in wordSize |
aoqi@1 | 794 | comp_words_on_stack = round_to(comp_words_on_stack, 2); |
aoqi@1 | 795 | __ daddi(SP, SP, -comp_words_on_stack * wordSize); |
aoqi@1 | 796 | } |
aoqi@1 | 797 | |
aoqi@1 | 798 | // Align the outgoing SP |
aoqi@1 | 799 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@6880 | 800 | __ andr(SP, SP, AT); |
aoqi@1 | 801 | // push the return address on the stack (note that pushing, rather |
aoqi@1 | 802 | // than storing it, yields the correct frame alignment for the callee) |
aoqi@1 | 803 | // Put saved SP in another register |
aoqi@1 | 804 | const Register saved_sp = V0; |
aoqi@1 | 805 | __ move(saved_sp, T9); |
aoqi@1 | 806 | |
aoqi@1 | 807 | |
aoqi@1 | 808 | // Will jump to the compiled code just as if compiled code was doing it. |
aoqi@1 | 809 | // Pre-load the register-jump target early, to schedule it better. |
aoqi@1 | 810 | __ ld(T9, Rmethod, in_bytes(Method::from_compiled_offset())); |
aoqi@1 | 811 | |
aoqi@1 | 812 | // Now generate the shuffle code. Pick up all register args and move the |
aoqi@1 | 813 | // rest through the floating point stack top. |
aoqi@1 | 814 | for (int i = 0; i < total_args_passed; i++) { |
aoqi@1 | 815 | if (sig_bt[i] == T_VOID) { |
aoqi@1 | 816 | // Longs and doubles are passed in native word order, but misaligned |
aoqi@1 | 817 | // in the 32-bit build. |
aoqi@1 | 818 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
aoqi@1 | 819 | continue; |
aoqi@1 | 820 | } |
aoqi@1 | 821 | |
aoqi@6880 | 822 | // Pick up 0, 1 or 2 words from SP+offset. |
aoqi@6880 | 823 | |
aoqi@6880 | 824 | //FIXME. aoqi. just delete the assert |
aoqi@1 | 825 | //assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), "scrambled load targets?"); |
aoqi@1 | 826 | // Load in argument order going down. |
aoqi@1 | 827 | int ld_off = (total_args_passed -1 - i)*Interpreter::stackElementSize; |
aoqi@1 | 828 | // Point to interpreter value (vs. tag) |
aoqi@1 | 829 | int next_off = ld_off - Interpreter::stackElementSize; |
aoqi@1 | 830 | VMReg r_1 = regs[i].first(); |
aoqi@1 | 831 | VMReg r_2 = regs[i].second(); |
aoqi@1 | 832 | if (!r_1->is_valid()) { |
aoqi@1 | 833 | assert(!r_2->is_valid(), ""); |
aoqi@1 | 834 | continue; |
aoqi@1 | 835 | } |
aoqi@6880 | 836 | if (r_1->is_stack()) { |
aoqi@6880 | 837 | // Convert stack slot to an SP offset (+ wordSize to |
aoqi@1 | 838 | // account for return address ) |
aoqi@9459 | 839 | // NOTICE HERE!!!! I sub a wordSize here |
aoqi@6880 | 840 | int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size; |
aoqi@1 | 841 | //+ wordSize; |
aoqi@1 | 842 | |
aoqi@1 | 843 | if (!r_2->is_valid()) { |
aoqi@6880 | 844 | __ ld(AT, saved_sp, ld_off); |
aoqi@6880 | 845 | __ sd(AT, SP, st_off); |
aoqi@1 | 846 | } else { |
aoqi@6880 | 847 | // Interpreter local[n] == MSW, local[n+1] == LSW however locals |
aoqi@6880 | 848 | // are accessed as negative so LSW is at LOW address |
aoqi@6880 | 849 | |
aoqi@6880 | 850 | // ld_off is MSW so get LSW |
aoqi@6880 | 851 | // st_off is LSW (i.e. reg.first()) |
aoqi@9459 | 852 | |
aoqi@9459 | 853 | // [./org/eclipse/swt/graphics/GC.java] |
aoqi@9459 | 854 | // void drawImageXRender(Image srcImage, int srcX, int srcY, int srcWidth, int srcHeight, |
aoqi@9459 | 855 | // int destX, int destY, int destWidth, int destHeight, |
aoqi@9459 | 856 | // boolean simple, |
aoqi@9459 | 857 | // int imgWidth, int imgHeight, |
aoqi@9459 | 858 | // long maskPixmap, <-- Pass T_LONG in stack |
aoqi@9459 | 859 | // int maskType); |
aoqi@9459 | 860 | // Before this modification, Eclipse displays icons with solid black background. |
aoqi@9459 | 861 | // |
aoqi@6880 | 862 | __ ld(AT, saved_sp, ld_off); |
aoqi@1 | 863 | if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) |
aoqi@6880 | 864 | __ ld(AT, saved_sp, ld_off - 8); |
aoqi@6880 | 865 | __ sd(AT, SP, st_off); |
aoqi@1 | 866 | } |
aoqi@1 | 867 | } else if (r_1->is_Register()) { // Register argument |
aoqi@1 | 868 | Register r = r_1->as_Register(); |
aoqi@1 | 869 | if (r_2->is_valid()) { |
aoqi@6880 | 870 | // Remember r_1 is low address (and LSB on mips) |
aoqi@6880 | 871 | // So r_2 gets loaded from high address regardless of the platform |
aoqi@6880 | 872 | assert(r_2->as_Register() == r_1->as_Register(), ""); |
aoqi@6880 | 873 | __ ld(r, saved_sp, ld_off); |
aoqi@6880 | 874 | |
aoqi@9459 | 875 | // |
aoqi@9459 | 876 | // For T_LONG type, the real layout is as below: |
aoqi@9459 | 877 | // |
aoqi@9459 | 878 | // (high) |
aoqi@9459 | 879 | // | | |
aoqi@9459 | 880 | // ----------- |
aoqi@9459 | 881 | // | 8 bytes | |
aoqi@9459 | 882 | // | (void) | |
aoqi@9459 | 883 | // ----------- |
aoqi@9459 | 884 | // | 8 bytes | |
aoqi@9459 | 885 | // | (long) | |
aoqi@9459 | 886 | // ----------- |
aoqi@9459 | 887 | // | | |
aoqi@9459 | 888 | // (low) |
aoqi@9459 | 889 | // |
aoqi@9459 | 890 | // We should load the low-8 bytes. |
aoqi@9459 | 891 | // |
aoqi@6880 | 892 | if (sig_bt[i] == T_LONG) |
aoqi@6880 | 893 | __ ld(r, saved_sp, ld_off - 8); |
aoqi@1 | 894 | } else { |
aoqi@6880 | 895 | __ lw(r, saved_sp, ld_off); |
aoqi@1 | 896 | } |
aoqi@1 | 897 | } else if (r_1->is_FloatRegister()) { // Float Register |
aoqi@6880 | 898 | assert(sig_bt[i] == T_FLOAT || sig_bt[i] == T_DOUBLE, "Must be a float register"); |
aoqi@6880 | 899 | |
aoqi@6880 | 900 | FloatRegister fr = r_1->as_FloatRegister(); |
aoqi@6880 | 901 | if (sig_bt[i] == T_FLOAT) |
aoqi@6880 | 902 | __ lwc1(fr, saved_sp, ld_off); |
aoqi@6880 | 903 | else { |
aoqi@6880 | 904 | __ ldc1(fr, saved_sp, ld_off); |
aoqi@6880 | 905 | __ ldc1(fr, saved_sp, ld_off - 8); |
aoqi@6880 | 906 | } |
aoqi@6880 | 907 | } |
aoqi@1 | 908 | } |
aoqi@1 | 909 | |
aoqi@1 | 910 | // 6243940 We might end up in handle_wrong_method if |
aoqi@1 | 911 | // the callee is deoptimized as we race thru here. If that |
aoqi@1 | 912 | // happens we don't want to take a safepoint because the |
aoqi@1 | 913 | // caller frame will look interpreted and arguments are now |
aoqi@1 | 914 | // "compiled" so it is much better to make this transition |
aoqi@1 | 915 | // invisible to the stack walking code. Unfortunately if |
aoqi@1 | 916 | // we try and find the callee by normal means a safepoint |
aoqi@1 | 917 | // is possible. So we stash the desired callee in the thread |
aoqi@1 | 918 | // and the vm will find there should this case occur. |
aoqi@1 | 919 | __ get_thread(T8); |
aoqi@1 | 920 | __ sd(Rmethod, T8, in_bytes(JavaThread::callee_target_offset())); |
aoqi@1 | 921 | |
aoqi@9459 | 922 | // move methodOop to V0 in case we end up in an c2i adapter. |
aoqi@9459 | 923 | // the c2i adapters expect methodOop in V0 (c2) because c2's |
aoqi@9459 | 924 | // resolve stubs return the result (the method) in V0. |
aoqi@6880 | 925 | // I'd love to fix this. |
aoqi@6880 | 926 | __ move(V0, Rmethod); |
aoqi@1 | 927 | __ jr(T9); |
aoqi@1 | 928 | __ delayed()->nop(); |
aoqi@1 | 929 | } |
aoqi@1 | 930 | |
aoqi@1 | 931 | // --------------------------------------------------------------- |
aoqi@1 | 932 | AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, |
aoqi@1 | 933 | int total_args_passed, |
aoqi@1 | 934 | // VMReg max_arg, |
aoqi@1 | 935 | int comp_args_on_stack, // VMRegStackSlots |
aoqi@1 | 936 | const BasicType *sig_bt, |
aoqi@1 | 937 | const VMRegPair *regs, |
aoqi@1 | 938 | AdapterFingerPrint* fingerprint) { |
aoqi@1 | 939 | address i2c_entry = __ pc(); |
aoqi@1 | 940 | |
aoqi@1 | 941 | AdapterGenerator agen(masm); |
aoqi@1 | 942 | |
aoqi@1 | 943 | agen.gen_i2c_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs); |
aoqi@1 | 944 | |
aoqi@1 | 945 | |
aoqi@1 | 946 | // ------------------------------------------------------------------------- |
aoqi@1 | 947 | // Generate a C2I adapter. On entry we know G5 holds the methodOop. The |
aoqi@1 | 948 | // args start out packed in the compiled layout. They need to be unpacked |
aoqi@1 | 949 | // into the interpreter layout. This will almost always require some stack |
aoqi@1 | 950 | // space. We grow the current (compiled) stack, then repack the args. We |
aoqi@1 | 951 | // finally end in a jump to the generic interpreter entry point. On exit |
aoqi@1 | 952 | // from the interpreter, the interpreter will restore our SP (lest the |
aoqi@1 | 953 | // compiled code, which relys solely on SP and not FP, get sick). |
aoqi@1 | 954 | |
aoqi@1 | 955 | address c2i_unverified_entry = __ pc(); |
aoqi@1 | 956 | Label skip_fixup; |
aoqi@1 | 957 | { |
aoqi@1 | 958 | Register holder = T1; |
aoqi@1 | 959 | Register receiver = T0; |
aoqi@1 | 960 | Register temp = T8; |
aoqi@1 | 961 | address ic_miss = SharedRuntime::get_ic_miss_stub(); |
aoqi@1 | 962 | |
aoqi@1 | 963 | Label missed; |
aoqi@1 | 964 | |
aoqi@1 | 965 | __ verify_oop(holder); |
aoqi@1 | 966 | //add for compressedoops |
aoqi@1 | 967 | __ load_klass(temp, receiver); |
aoqi@1 | 968 | __ verify_oop(temp); |
aoqi@1 | 969 | |
aoqi@6880 | 970 | __ ld_ptr(AT, holder, CompiledICHolder::holder_klass_offset()); |
aoqi@9043 | 971 | __ ld_ptr(Rmethod, holder, CompiledICHolder::holder_metadata_offset()); |
aoqi@6880 | 972 | __ bne(AT, temp, missed); |
aoqi@6880 | 973 | __ delayed()->nop(); |
aoqi@1 | 974 | // Method might have been compiled since the call site was patched to |
aoqi@1 | 975 | // interpreted if that is the case treat it as a miss so we can get |
aoqi@1 | 976 | // the call site corrected. |
aoqi@1 | 977 | __ ld_ptr(AT, Rmethod, in_bytes(Method::code_offset())); |
aoqi@6880 | 978 | __ beq(AT, R0, skip_fixup); |
aoqi@6880 | 979 | __ delayed()->nop(); |
aoqi@1 | 980 | __ bind(missed); |
aoqi@1 | 981 | |
aoqi@1 | 982 | __ jmp(ic_miss, relocInfo::runtime_call_type); |
aoqi@6880 | 983 | __ delayed()->nop(); |
aoqi@1 | 984 | } |
aoqi@1 | 985 | |
aoqi@1 | 986 | address c2i_entry = __ pc(); |
aoqi@1 | 987 | |
aoqi@1 | 988 | agen.gen_c2i_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); |
aoqi@1 | 989 | |
aoqi@1 | 990 | __ flush(); |
huangjia@9645 | 991 | return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); |
aoqi@1 | 992 | } |
aoqi@1 | 993 | |
aoqi@1 | 994 | int SharedRuntime::c_calling_convention(const BasicType *sig_bt, |
aoqi@1 | 995 | VMRegPair *regs, |
aoqi@1 | 996 | VMRegPair *regs2, |
aoqi@1 | 997 | int total_args_passed) { |
aoqi@6880 | 998 | assert(regs2 == NULL, "not needed on MIPS"); |
aoqi@6880 | 999 | // Return the number of VMReg stack_slots needed for the args. |
aoqi@6880 | 1000 | // This value does not include an abi space (like register window |
aoqi@6880 | 1001 | // save area). |
aoqi@6880 | 1002 | |
aoqi@6880 | 1003 | // The native convention is V8 if !LP64 |
aoqi@6880 | 1004 | // The LP64 convention is the V9 convention which is slightly more sane. |
aoqi@6880 | 1005 | |
aoqi@6880 | 1006 | // We return the amount of VMReg stack slots we need to reserve for all |
aoqi@6880 | 1007 | // the arguments NOT counting out_preserve_stack_slots. Since we always |
aoqi@6880 | 1008 | // have space for storing at least 6 registers to memory we start with that. |
aoqi@6880 | 1009 | // See int_stk_helper for a further discussion. |
aoqi@6880 | 1010 | // We return the amount of VMRegImpl stack slots we need to reserve for all |
aoqi@6880 | 1011 | // the arguments NOT counting out_preserve_stack_slots. |
aoqi@1 | 1012 | static const Register INT_ArgReg[Argument::n_register_parameters] = { |
aoqi@1 | 1013 | A0, A1, A2, A3, A4, A5, A6, A7 |
aoqi@1 | 1014 | }; |
aoqi@1 | 1015 | static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters] = { |
aoqi@1 | 1016 | F12, F13, F14, F15, F16, F17, F18, F19 |
aoqi@1 | 1017 | }; |
aoqi@6880 | 1018 | uint args = 0; |
aoqi@6880 | 1019 | uint stk_args = 0; // inc by 2 each time |
aoqi@1 | 1020 | |
aoqi@9459 | 1021 | // Example: |
aoqi@9459 | 1022 | // n java.lang.UNIXProcess::forkAndExec |
aoqi@9459 | 1023 | // private native int forkAndExec(byte[] prog, |
aoqi@9459 | 1024 | // byte[] argBlock, int argc, |
aoqi@9459 | 1025 | // byte[] envBlock, int envc, |
aoqi@9459 | 1026 | // byte[] dir, |
aoqi@9459 | 1027 | // boolean redirectErrorStream, |
aoqi@9459 | 1028 | // FileDescriptor stdin_fd, |
aoqi@9459 | 1029 | // FileDescriptor stdout_fd, |
aoqi@9459 | 1030 | // FileDescriptor stderr_fd) |
aoqi@9459 | 1031 | // JNIEXPORT jint JNICALL |
aoqi@9459 | 1032 | // Java_java_lang_UNIXProcess_forkAndExec(JNIEnv *env, |
aoqi@9459 | 1033 | // jobject process, |
aoqi@9459 | 1034 | // jbyteArray prog, |
aoqi@9459 | 1035 | // jbyteArray argBlock, jint argc, |
aoqi@9459 | 1036 | // jbyteArray envBlock, jint envc, |
aoqi@9459 | 1037 | // jbyteArray dir, |
aoqi@9459 | 1038 | // jboolean redirectErrorStream, |
aoqi@9459 | 1039 | // jobject stdin_fd, |
aoqi@9459 | 1040 | // jobject stdout_fd, |
aoqi@9459 | 1041 | // jobject stderr_fd) |
aoqi@9459 | 1042 | // |
aoqi@9459 | 1043 | // ::c_calling_convention |
aoqi@9459 | 1044 | // 0: // env <-- a0 |
aoqi@9459 | 1045 | // 1: L // klass/obj <-- t0 => a1 |
aoqi@9459 | 1046 | // 2: [ // prog[] <-- a0 => a2 |
aoqi@9459 | 1047 | // 3: [ // argBlock[] <-- a1 => a3 |
aoqi@9459 | 1048 | // 4: I // argc |
aoqi@9459 | 1049 | // 5: [ // envBlock[] <-- a3 => a5 |
aoqi@9459 | 1050 | // 6: I // envc |
aoqi@9459 | 1051 | // 7: [ // dir[] <-- a5 => a7 |
aoqi@9459 | 1052 | // 8: Z // redirectErrorStream a6 => sp[0] |
aoqi@9459 | 1053 | // 9: L // stdin a7 => sp[8] |
aoqi@9459 | 1054 | // 10: L // stdout fp[16] => sp[16] |
aoqi@9459 | 1055 | // 11: L // stderr fp[24] => sp[24] |
aoqi@9459 | 1056 | // |
aoqi@6880 | 1057 | for (int i = 0; i < total_args_passed; i++) { |
aoqi@6880 | 1058 | switch (sig_bt[i]) { |
aoqi@6880 | 1059 | case T_VOID: // Halves of longs and doubles |
aoqi@6880 | 1060 | assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
aoqi@6880 | 1061 | regs[i].set_bad(); |
aoqi@6880 | 1062 | break; |
aoqi@1 | 1063 | case T_BOOLEAN: |
aoqi@1 | 1064 | case T_CHAR: |
aoqi@1 | 1065 | case T_BYTE: |
aoqi@1 | 1066 | case T_SHORT: |
aoqi@1 | 1067 | case T_INT: |
aoqi@6880 | 1068 | if (args < Argument::n_register_parameters) { |
aoqi@6880 | 1069 | regs[i].set1(INT_ArgReg[args++]->as_VMReg()); |
aoqi@6880 | 1070 | } else { |
aoqi@6880 | 1071 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
aoqi@6880 | 1072 | stk_args += 2; |
aoqi@6880 | 1073 | } |
aoqi@6880 | 1074 | break; |
aoqi@6880 | 1075 | case T_LONG: |
aoqi@6880 | 1076 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
aoqi@6880 | 1077 | // fall through |
aoqi@1 | 1078 | case T_OBJECT: |
aoqi@1 | 1079 | case T_ARRAY: |
aoqi@1 | 1080 | case T_ADDRESS: |
aoqi@6880 | 1081 | case T_METADATA: |
aoqi@6880 | 1082 | if (args < Argument::n_register_parameters) { |
aoqi@6880 | 1083 | regs[i].set2(INT_ArgReg[args++]->as_VMReg()); |
aoqi@6880 | 1084 | } else { |
aoqi@6880 | 1085 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
aoqi@6880 | 1086 | stk_args += 2; |
aoqi@6880 | 1087 | } |
aoqi@1 | 1088 | break; |
aoqi@6880 | 1089 | case T_FLOAT: |
aoqi@6880 | 1090 | if (args < Argument::n_float_register_parameters) { |
aoqi@6880 | 1091 | regs[i].set1(FP_ArgReg[args++]->as_VMReg()); |
aoqi@6880 | 1092 | } else { |
aoqi@6880 | 1093 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
aoqi@6880 | 1094 | stk_args += 2; |
aoqi@6880 | 1095 | } |
aoqi@1 | 1096 | break; |
aoqi@6880 | 1097 | case T_DOUBLE: |
aoqi@6880 | 1098 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
aoqi@6880 | 1099 | if (args < Argument::n_float_register_parameters) { |
aoqi@6880 | 1100 | regs[i].set2(FP_ArgReg[args++]->as_VMReg()); |
aoqi@6880 | 1101 | } else { |
aoqi@6880 | 1102 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
aoqi@6880 | 1103 | stk_args += 2; |
aoqi@6880 | 1104 | } |
aoqi@6880 | 1105 | break; |
aoqi@1 | 1106 | default: |
aoqi@1 | 1107 | ShouldNotReachHere(); |
aoqi@1 | 1108 | break; |
aoqi@1 | 1109 | } |
aoqi@1 | 1110 | } |
aoqi@6880 | 1111 | |
aoqi@6880 | 1112 | return round_to(stk_args, 2); |
aoqi@1 | 1113 | } |
aoqi@1 | 1114 | |
aoqi@1 | 1115 | // --------------------------------------------------------------------------- |
aoqi@1 | 1116 | void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
aoqi@6880 | 1117 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
aoqi@6880 | 1118 | // which by this time is free to use |
aoqi@6880 | 1119 | switch (ret_type) { |
aoqi@6880 | 1120 | case T_FLOAT: |
aoqi@6880 | 1121 | __ swc1(FSF, FP, -wordSize); |
aoqi@6880 | 1122 | break; |
aoqi@6880 | 1123 | case T_DOUBLE: |
aoqi@6880 | 1124 | __ sdc1(FSF, FP, -wordSize ); |
aoqi@6880 | 1125 | break; |
aoqi@6880 | 1126 | case T_VOID: break; |
aoqi@6880 | 1127 | case T_LONG: |
aoqi@8009 | 1128 | __ sd(V0, FP, -wordSize); |
aoqi@8009 | 1129 | break; |
aoqi@6880 | 1130 | case T_OBJECT: |
aoqi@6880 | 1131 | case T_ARRAY: |
aoqi@6880 | 1132 | __ sd(V0, FP, -wordSize); |
aoqi@6880 | 1133 | break; |
aoqi@6880 | 1134 | default: { |
aoqi@8009 | 1135 | __ sw(V0, FP, -wordSize); |
aoqi@8009 | 1136 | } |
aoqi@6880 | 1137 | } |
aoqi@1 | 1138 | } |
aoqi@1 | 1139 | |
aoqi@1 | 1140 | void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
aoqi@6880 | 1141 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
aoqi@6880 | 1142 | // which by this time is free to use |
aoqi@6880 | 1143 | switch (ret_type) { |
aoqi@6880 | 1144 | case T_FLOAT: |
aoqi@6880 | 1145 | __ lwc1(FSF, FP, -wordSize); |
aoqi@6880 | 1146 | break; |
aoqi@6880 | 1147 | case T_DOUBLE: |
aoqi@6880 | 1148 | __ ldc1(FSF, FP, -wordSize ); |
aoqi@6880 | 1149 | break; |
aoqi@6880 | 1150 | case T_LONG: |
aoqi@6880 | 1151 | __ ld(V0, FP, -wordSize); |
aoqi@6880 | 1152 | break; |
aoqi@6880 | 1153 | case T_VOID: break; |
aoqi@6880 | 1154 | case T_OBJECT: |
aoqi@6880 | 1155 | case T_ARRAY: |
aoqi@6880 | 1156 | __ ld(V0, FP, -wordSize); |
aoqi@6880 | 1157 | break; |
aoqi@6880 | 1158 | default: { |
aoqi@8009 | 1159 | __ lw(V0, FP, -wordSize); |
aoqi@8009 | 1160 | } |
aoqi@6880 | 1161 | } |
aoqi@1 | 1162 | } |
aoqi@1 | 1163 | |
aoqi@1 | 1164 | static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
aoqi@6880 | 1165 | for ( int i = first_arg ; i < arg_count ; i++ ) { |
aoqi@6880 | 1166 | if (args[i].first()->is_Register()) { |
aoqi@6880 | 1167 | __ push(args[i].first()->as_Register()); |
aoqi@6880 | 1168 | } else if (args[i].first()->is_FloatRegister()) { |
aoqi@6880 | 1169 | __ push(args[i].first()->as_FloatRegister()); |
aoqi@1 | 1170 | } |
aoqi@6880 | 1171 | } |
aoqi@1 | 1172 | } |
aoqi@1 | 1173 | |
aoqi@1 | 1174 | static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
aoqi@6880 | 1175 | for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) { |
aoqi@6880 | 1176 | if (args[i].first()->is_Register()) { |
aoqi@6880 | 1177 | __ pop(args[i].first()->as_Register()); |
aoqi@6880 | 1178 | } else if (args[i].first()->is_FloatRegister()) { |
aoqi@6880 | 1179 | __ pop(args[i].first()->as_FloatRegister()); |
aoqi@1 | 1180 | } |
aoqi@6880 | 1181 | } |
aoqi@1 | 1182 | } |
aoqi@1 | 1183 | |
aoqi@1 | 1184 | // A simple move of integer like type |
aoqi@1 | 1185 | static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
aoqi@1 | 1186 | if (src.first()->is_stack()) { |
aoqi@1 | 1187 | if (dst.first()->is_stack()) { |
aoqi@1 | 1188 | // stack to stack |
aoqi@6880 | 1189 | __ lw(AT, FP, reg2offset_in(src.first())); |
huangjia@9645 | 1190 | __ sd(AT, SP, reg2offset_out(dst.first())); |
aoqi@1 | 1191 | } else { |
aoqi@1 | 1192 | // stack to reg |
aoqi@6880 | 1193 | __ lw(dst.first()->as_Register(), FP, reg2offset_in(src.first())); |
aoqi@1 | 1194 | } |
aoqi@1 | 1195 | } else if (dst.first()->is_stack()) { |
aoqi@1 | 1196 | // reg to stack |
aoqi@6880 | 1197 | __ sd(src.first()->as_Register(), SP, reg2offset_out(dst.first())); |
aoqi@1 | 1198 | } else { |
aoqi@6880 | 1199 | if (dst.first() != src.first()){ |
aoqi@6880 | 1200 | __ move(dst.first()->as_Register(), src.first()->as_Register()); // fujie error:dst.first() |
aoqi@6880 | 1201 | } |
aoqi@1 | 1202 | } |
aoqi@1 | 1203 | } |
aoqi@1 | 1204 | |
aoqi@1 | 1205 | // An oop arg. Must pass a handle not the oop itself |
aoqi@1 | 1206 | static void object_move(MacroAssembler* masm, |
aoqi@1 | 1207 | OopMap* map, |
aoqi@1 | 1208 | int oop_handle_offset, |
aoqi@1 | 1209 | int framesize_in_slots, |
aoqi@1 | 1210 | VMRegPair src, |
aoqi@1 | 1211 | VMRegPair dst, |
aoqi@1 | 1212 | bool is_receiver, |
aoqi@1 | 1213 | int* receiver_offset) { |
aoqi@1 | 1214 | |
aoqi@1 | 1215 | // must pass a handle. First figure out the location we use as a handle |
aoqi@1 | 1216 | |
aoqi@6880 | 1217 | //FIXME, for mips, dst can be register |
aoqi@6880 | 1218 | if (src.first()->is_stack()) { |
aoqi@6880 | 1219 | // Oop is already on the stack as an argument |
aoqi@6880 | 1220 | Register rHandle = V0; |
aoqi@6880 | 1221 | Label nil; |
aoqi@6880 | 1222 | __ xorr(rHandle, rHandle, rHandle); |
aoqi@6880 | 1223 | __ ld(AT, FP, reg2offset_in(src.first())); |
huangjia@9645 | 1224 | __ beq(AT, R0, nil); |
aoqi@6880 | 1225 | __ delayed()->nop(); |
aoqi@6880 | 1226 | __ lea(rHandle, Address(FP, reg2offset_in(src.first()))); |
aoqi@6880 | 1227 | __ bind(nil); |
aoqi@6880 | 1228 | if(dst.first()->is_stack())__ sd( rHandle, SP, reg2offset_out(dst.first())); |
huangjia@9645 | 1229 | else __ move( (dst.first())->as_Register(), rHandle); |
aoqi@6880 | 1230 | //if dst is register |
aoqi@6880 | 1231 | //FIXME, do mips need out preserve stack slots? |
aoqi@6880 | 1232 | int offset_in_older_frame = src.first()->reg2stack() |
aoqi@6880 | 1233 | + SharedRuntime::out_preserve_stack_slots(); |
aoqi@6880 | 1234 | map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); |
aoqi@6880 | 1235 | if (is_receiver) { |
aoqi@6880 | 1236 | *receiver_offset = (offset_in_older_frame |
aoqi@6880 | 1237 | + framesize_in_slots) * VMRegImpl::stack_slot_size; |
aoqi@6880 | 1238 | } |
aoqi@6880 | 1239 | } else { |
aoqi@6880 | 1240 | // Oop is in an a register we must store it to the space we reserve |
aoqi@6880 | 1241 | // on the stack for oop_handles |
aoqi@6880 | 1242 | const Register rOop = src.first()->as_Register(); |
aoqi@6880 | 1243 | assert( (rOop->encoding() >= A0->encoding()) && (rOop->encoding() <= T0->encoding()),"wrong register"); |
aoqi@6880 | 1244 | const Register rHandle = V0; |
aoqi@6880 | 1245 | //Important: refer to java_calling_convertion |
aoqi@6880 | 1246 | int oop_slot = (rOop->encoding() - A0->encoding()) * VMRegImpl::slots_per_word + oop_handle_offset; |
aoqi@6880 | 1247 | int offset = oop_slot*VMRegImpl::stack_slot_size; |
aoqi@6880 | 1248 | Label skip; |
aoqi@6880 | 1249 | __ sd( rOop , SP, offset ); |
aoqi@6880 | 1250 | map->set_oop(VMRegImpl::stack2reg(oop_slot)); |
aoqi@6880 | 1251 | __ xorr( rHandle, rHandle, rHandle); |
aoqi@6880 | 1252 | __ beq(rOop, R0, skip); |
aoqi@6880 | 1253 | __ delayed()->nop(); |
aoqi@6880 | 1254 | __ lea(rHandle, Address(SP, offset)); |
aoqi@6880 | 1255 | __ bind(skip); |
aoqi@6880 | 1256 | // Store the handle parameter |
aoqi@6880 | 1257 | if(dst.first()->is_stack())__ sd( rHandle, SP, reg2offset_out(dst.first())); |
aoqi@6880 | 1258 | else __ move((dst.first())->as_Register(), rHandle); |
aoqi@6880 | 1259 | //if dst is register |
aoqi@6880 | 1260 | |
aoqi@6880 | 1261 | if (is_receiver) { |
aoqi@6880 | 1262 | *receiver_offset = offset; |
aoqi@6880 | 1263 | } |
aoqi@6880 | 1264 | } |
aoqi@1 | 1265 | } |
aoqi@1 | 1266 | |
aoqi@1 | 1267 | // A float arg may have to do float reg int reg conversion |
aoqi@1 | 1268 | static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
aoqi@1 | 1269 | assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); |
aoqi@1 | 1270 | |
aoqi@6880 | 1271 | if (src.first()->is_stack()) { |
aoqi@6880 | 1272 | if (dst.first()->is_stack()) { |
huangjia@9645 | 1273 | __ lwc1(F12, FP, reg2offset_in(src.first())); |
huangjia@9645 | 1274 | __ swc1(F12, SP, reg2offset_out(dst.first())); |
aoqi@6880 | 1275 | } |
aoqi@6880 | 1276 | else |
huangjia@9645 | 1277 | __ lwc1(dst.first()->as_FloatRegister(), FP, reg2offset_in(src.first())); |
aoqi@6880 | 1278 | } else { |
aoqi@6880 | 1279 | // reg to stack |
aoqi@6880 | 1280 | if(dst.first()->is_stack()) |
huangjia@9645 | 1281 | __ swc1(src.first()->as_FloatRegister(), SP, reg2offset_out(dst.first())); |
aoqi@6880 | 1282 | else |
huangjia@9645 | 1283 | __ mov_s(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister()); |
aoqi@6880 | 1284 | } |
aoqi@1 | 1285 | } |
aoqi@6880 | 1286 | |
aoqi@1 | 1287 | // A long move |
aoqi@1 | 1288 | static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
aoqi@1 | 1289 | |
aoqi@6880 | 1290 | // The only legal possibility for a long_move VMRegPair is: |
aoqi@6880 | 1291 | // 1: two stack slots (possibly unaligned) |
aoqi@6880 | 1292 | // as neither the java or C calling convention will use registers |
aoqi@6880 | 1293 | // for longs. |
aoqi@6880 | 1294 | |
aoqi@6880 | 1295 | if (src.first()->is_stack()) { |
aoqi@6880 | 1296 | assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack"); |
aoqi@6880 | 1297 | if( dst.first()->is_stack()){ |
aoqi@6880 | 1298 | __ ld(AT, FP, reg2offset_in(src.first())); |
aoqi@6880 | 1299 | __ sd(AT, SP, reg2offset_out(dst.first())); |
aoqi@6880 | 1300 | } else { |
aoqi@6880 | 1301 | __ ld( (dst.first())->as_Register() , FP, reg2offset_in(src.first())); |
aoqi@6880 | 1302 | } |
aoqi@6880 | 1303 | } else { |
aoqi@6880 | 1304 | if( dst.first()->is_stack()){ |
aoqi@6880 | 1305 | __ sd( (src.first())->as_Register(), SP, reg2offset_out(dst.first())); |
huangjia@9645 | 1306 | } else { |
aoqi@6880 | 1307 | __ move( (dst.first())->as_Register() , (src.first())->as_Register()); |
aoqi@6880 | 1308 | } |
aoqi@6880 | 1309 | } |
aoqi@1 | 1310 | } |
aoqi@1 | 1311 | |
aoqi@1 | 1312 | // A double move |
aoqi@1 | 1313 | static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
aoqi@1 | 1314 | |
aoqi@6880 | 1315 | // The only legal possibilities for a double_move VMRegPair are: |
aoqi@6880 | 1316 | // The painful thing here is that like long_move a VMRegPair might be |
aoqi@6880 | 1317 | |
aoqi@6880 | 1318 | // Because of the calling convention we know that src is either |
aoqi@6880 | 1319 | // 1: a single physical register (xmm registers only) |
aoqi@6880 | 1320 | // 2: two stack slots (possibly unaligned) |
aoqi@6880 | 1321 | // dst can only be a pair of stack slots. |
aoqi@6880 | 1322 | |
aoqi@6880 | 1323 | |
aoqi@6880 | 1324 | if (src.first()->is_stack()) { |
aoqi@6880 | 1325 | // source is all stack |
aoqi@6880 | 1326 | if( dst.first()->is_stack()){ |
aoqi@6880 | 1327 | __ ldc1(F12, FP, reg2offset_in(src.first())); |
aoqi@6880 | 1328 | |
aoqi@6880 | 1329 | __ sdc1(F12, SP, reg2offset_out(dst.first())); |
huangjia@9645 | 1330 | } else { |
aoqi@6880 | 1331 | __ ldc1( (dst.first())->as_FloatRegister(), FP, reg2offset_in(src.first())); |
aoqi@6880 | 1332 | } |
aoqi@6880 | 1333 | |
aoqi@6880 | 1334 | } else { |
aoqi@6880 | 1335 | // reg to stack |
aoqi@6880 | 1336 | // No worries about stack alignment |
aoqi@6880 | 1337 | if( dst.first()->is_stack()){ |
huangjia@9645 | 1338 | __ sdc1(src.first()->as_FloatRegister(), SP, reg2offset_out(dst.first())); |
aoqi@6880 | 1339 | } |
aoqi@6880 | 1340 | else |
aoqi@6880 | 1341 | __ mov_d( dst.first()->as_FloatRegister(), src.first()->as_FloatRegister()); |
aoqi@6880 | 1342 | |
aoqi@6880 | 1343 | } |
aoqi@1 | 1344 | } |
aoqi@1 | 1345 | |
aoqi@1 | 1346 | static void verify_oop_args(MacroAssembler* masm, |
aoqi@1 | 1347 | methodHandle method, |
aoqi@1 | 1348 | const BasicType* sig_bt, |
aoqi@1 | 1349 | const VMRegPair* regs) { |
aoqi@1 | 1350 | Register temp_reg = T9; // not part of any compiled calling seq |
aoqi@1 | 1351 | if (VerifyOops) { |
aoqi@1 | 1352 | for (int i = 0; i < method->size_of_parameters(); i++) { |
aoqi@1 | 1353 | if (sig_bt[i] == T_OBJECT || |
aoqi@1 | 1354 | sig_bt[i] == T_ARRAY) { |
aoqi@1 | 1355 | VMReg r = regs[i].first(); |
aoqi@1 | 1356 | assert(r->is_valid(), "bad oop arg"); |
aoqi@1 | 1357 | if (r->is_stack()) { |
aoqi@1 | 1358 | __ ld(temp_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
aoqi@1 | 1359 | __ verify_oop(temp_reg); |
aoqi@1 | 1360 | } else { |
aoqi@1 | 1361 | __ verify_oop(r->as_Register()); |
aoqi@1 | 1362 | } |
aoqi@1 | 1363 | } |
aoqi@1 | 1364 | } |
aoqi@1 | 1365 | } |
aoqi@1 | 1366 | } |
aoqi@1 | 1367 | |
aoqi@1 | 1368 | static void gen_special_dispatch(MacroAssembler* masm, |
aoqi@1 | 1369 | methodHandle method, |
aoqi@1 | 1370 | const BasicType* sig_bt, |
aoqi@1 | 1371 | const VMRegPair* regs) { |
aoqi@1 | 1372 | verify_oop_args(masm, method, sig_bt, regs); |
aoqi@1 | 1373 | vmIntrinsics::ID iid = method->intrinsic_id(); |
aoqi@1 | 1374 | |
aoqi@1 | 1375 | // Now write the args into the outgoing interpreter space |
aoqi@1 | 1376 | bool has_receiver = false; |
aoqi@1 | 1377 | Register receiver_reg = noreg; |
aoqi@1 | 1378 | int member_arg_pos = -1; |
aoqi@1 | 1379 | Register member_reg = noreg; |
aoqi@1 | 1380 | int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid); |
aoqi@1 | 1381 | if (ref_kind != 0) { |
aoqi@1 | 1382 | member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument |
aoqi@1 | 1383 | member_reg = S3; // known to be free at this point |
aoqi@1 | 1384 | has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind); |
aoqi@1 | 1385 | } else if (iid == vmIntrinsics::_invokeBasic) { |
aoqi@1 | 1386 | has_receiver = true; |
aoqi@1 | 1387 | } else { |
aoqi@1 | 1388 | fatal(err_msg_res("unexpected intrinsic id %d", iid)); |
aoqi@1 | 1389 | } |
aoqi@1 | 1390 | |
aoqi@1 | 1391 | if (member_reg != noreg) { |
aoqi@1 | 1392 | // Load the member_arg into register, if necessary. |
aoqi@1 | 1393 | SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs); |
aoqi@1 | 1394 | VMReg r = regs[member_arg_pos].first(); |
aoqi@1 | 1395 | if (r->is_stack()) { |
fujie@410 | 1396 | __ ld(member_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size)); |
aoqi@1 | 1397 | } else { |
aoqi@1 | 1398 | // no data motion is needed |
aoqi@1 | 1399 | member_reg = r->as_Register(); |
aoqi@1 | 1400 | } |
aoqi@1 | 1401 | } |
aoqi@1 | 1402 | |
aoqi@1 | 1403 | if (has_receiver) { |
aoqi@1 | 1404 | // Make sure the receiver is loaded into a register. |
aoqi@1 | 1405 | assert(method->size_of_parameters() > 0, "oob"); |
aoqi@1 | 1406 | assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object"); |
aoqi@1 | 1407 | VMReg r = regs[0].first(); |
aoqi@1 | 1408 | assert(r->is_valid(), "bad receiver arg"); |
aoqi@1 | 1409 | if (r->is_stack()) { |
aoqi@1 | 1410 | // Porting note: This assumes that compiled calling conventions always |
aoqi@1 | 1411 | // pass the receiver oop in a register. If this is not true on some |
aoqi@1 | 1412 | // platform, pick a temp and load the receiver from stack. |
aoqi@1 | 1413 | fatal("receiver always in a register"); |
aoqi@1 | 1414 | receiver_reg = SSR; // known to be free at this point |
fujie@410 | 1415 | __ ld(receiver_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size)); |
aoqi@1 | 1416 | } else { |
aoqi@1 | 1417 | // no data motion is needed |
aoqi@1 | 1418 | receiver_reg = r->as_Register(); |
aoqi@1 | 1419 | } |
aoqi@1 | 1420 | } |
aoqi@1 | 1421 | |
aoqi@1 | 1422 | // Figure out which address we are really jumping to: |
aoqi@1 | 1423 | MethodHandles::generate_method_handle_dispatch(masm, iid, |
aoqi@1 | 1424 | receiver_reg, member_reg, /*for_compiler_entry:*/ true); |
aoqi@1 | 1425 | } |
aoqi@1 | 1426 | |
aoqi@1 | 1427 | // --------------------------------------------------------------------------- |
aoqi@1 | 1428 | // Generate a native wrapper for a given method. The method takes arguments |
aoqi@1 | 1429 | // in the Java compiled code convention, marshals them to the native |
aoqi@1 | 1430 | // convention (handlizes oops, etc), transitions to native, makes the call, |
aoqi@1 | 1431 | // returns to java state (possibly blocking), unhandlizes any result and |
aoqi@1 | 1432 | // returns. |
aoqi@1 | 1433 | nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler* masm, |
aoqi@1 | 1434 | methodHandle method, |
aoqi@1 | 1435 | int compile_id, |
aoqi@6880 | 1436 | BasicType* in_sig_bt, |
aoqi@6880 | 1437 | VMRegPair* in_regs, |
aoqi@1 | 1438 | BasicType ret_type) { |
aoqi@1 | 1439 | if (method->is_method_handle_intrinsic()) { |
aoqi@1 | 1440 | vmIntrinsics::ID iid = method->intrinsic_id(); |
aoqi@1 | 1441 | intptr_t start = (intptr_t)__ pc(); |
aoqi@1 | 1442 | int vep_offset = ((intptr_t)__ pc()) - start; |
aoqi@1 | 1443 | gen_special_dispatch(masm, |
aoqi@1 | 1444 | method, |
aoqi@1 | 1445 | in_sig_bt, |
aoqi@1 | 1446 | in_regs); |
aoqi@1 | 1447 | int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period |
aoqi@1 | 1448 | __ flush(); |
aoqi@1 | 1449 | int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually |
aoqi@1 | 1450 | return nmethod::new_native_nmethod(method, |
aoqi@1 | 1451 | compile_id, |
aoqi@1 | 1452 | masm->code(), |
aoqi@1 | 1453 | vep_offset, |
aoqi@1 | 1454 | frame_complete, |
aoqi@1 | 1455 | stack_slots / VMRegImpl::slots_per_word, |
aoqi@1 | 1456 | in_ByteSize(-1), |
aoqi@1 | 1457 | in_ByteSize(-1), |
aoqi@1 | 1458 | (OopMapSet*)NULL); |
aoqi@1 | 1459 | } |
aoqi@1 | 1460 | bool is_critical_native = true; |
aoqi@1 | 1461 | address native_func = method->critical_native_function(); |
aoqi@1 | 1462 | if (native_func == NULL) { |
aoqi@1 | 1463 | native_func = method->native_function(); |
aoqi@1 | 1464 | is_critical_native = false; |
aoqi@1 | 1465 | } |
aoqi@1 | 1466 | assert(native_func != NULL, "must have function"); |
aoqi@1 | 1467 | |
aoqi@1 | 1468 | // Native nmethod wrappers never take possesion of the oop arguments. |
aoqi@1 | 1469 | // So the caller will gc the arguments. The only thing we need an |
aoqi@1 | 1470 | // oopMap for is if the call is static |
aoqi@1 | 1471 | // |
aoqi@1 | 1472 | // An OopMap for lock (and class if static), and one for the VM call itself |
aoqi@1 | 1473 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@1 | 1474 | |
aoqi@6880 | 1475 | // We have received a description of where all the java arg are located |
aoqi@6880 | 1476 | // on entry to the wrapper. We need to convert these args to where |
aoqi@6880 | 1477 | // the jni function will expect them. To figure out where they go |
aoqi@6880 | 1478 | // we convert the java signature to a C signature by inserting |
aoqi@6880 | 1479 | // the hidden arguments as arg[0] and possibly arg[1] (static method) |
aoqi@1 | 1480 | |
aoqi@1 | 1481 | const int total_in_args = method->size_of_parameters(); |
aoqi@1 | 1482 | int total_c_args = total_in_args; |
aoqi@1 | 1483 | if (!is_critical_native) { |
aoqi@1 | 1484 | total_c_args += 1; |
aoqi@1 | 1485 | if (method->is_static()) { |
aoqi@1 | 1486 | total_c_args++; |
aoqi@1 | 1487 | } |
aoqi@1 | 1488 | } else { |
aoqi@1 | 1489 | for (int i = 0; i < total_in_args; i++) { |
aoqi@1 | 1490 | if (in_sig_bt[i] == T_ARRAY) { |
aoqi@1 | 1491 | total_c_args++; |
aoqi@1 | 1492 | } |
aoqi@1 | 1493 | } |
aoqi@1 | 1494 | } |
aoqi@1 | 1495 | |
aoqi@6880 | 1496 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); |
aoqi@6880 | 1497 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); |
aoqi@1 | 1498 | BasicType* in_elem_bt = NULL; |
aoqi@1 | 1499 | |
aoqi@1 | 1500 | int argc = 0; |
aoqi@1 | 1501 | if (!is_critical_native) { |
aoqi@1 | 1502 | out_sig_bt[argc++] = T_ADDRESS; |
aoqi@1 | 1503 | if (method->is_static()) { |
aoqi@1 | 1504 | out_sig_bt[argc++] = T_OBJECT; |
aoqi@1 | 1505 | } |
aoqi@1 | 1506 | |
aoqi@1 | 1507 | for (int i = 0; i < total_in_args ; i++ ) { |
aoqi@1 | 1508 | out_sig_bt[argc++] = in_sig_bt[i]; |
aoqi@1 | 1509 | } |
aoqi@1 | 1510 | } else { |
aoqi@1 | 1511 | Thread* THREAD = Thread::current(); |
aoqi@1 | 1512 | in_elem_bt = NEW_RESOURCE_ARRAY(BasicType, total_in_args); |
aoqi@1 | 1513 | SignatureStream ss(method->signature()); |
aoqi@1 | 1514 | for (int i = 0; i < total_in_args ; i++ ) { |
aoqi@1 | 1515 | if (in_sig_bt[i] == T_ARRAY) { |
aoqi@1 | 1516 | // Arrays are passed as int, elem* pair |
aoqi@1 | 1517 | out_sig_bt[argc++] = T_INT; |
aoqi@1 | 1518 | out_sig_bt[argc++] = T_ADDRESS; |
aoqi@1 | 1519 | Symbol* atype = ss.as_symbol(CHECK_NULL); |
aoqi@1 | 1520 | const char* at = atype->as_C_string(); |
aoqi@1 | 1521 | if (strlen(at) == 2) { |
aoqi@1 | 1522 | assert(at[0] == '[', "must be"); |
aoqi@1 | 1523 | switch (at[1]) { |
aoqi@1 | 1524 | case 'B': in_elem_bt[i] = T_BYTE; break; |
aoqi@1 | 1525 | case 'C': in_elem_bt[i] = T_CHAR; break; |
aoqi@1 | 1526 | case 'D': in_elem_bt[i] = T_DOUBLE; break; |
aoqi@1 | 1527 | case 'F': in_elem_bt[i] = T_FLOAT; break; |
aoqi@1 | 1528 | case 'I': in_elem_bt[i] = T_INT; break; |
aoqi@1 | 1529 | case 'J': in_elem_bt[i] = T_LONG; break; |
aoqi@1 | 1530 | case 'S': in_elem_bt[i] = T_SHORT; break; |
aoqi@1 | 1531 | case 'Z': in_elem_bt[i] = T_BOOLEAN; break; |
aoqi@1 | 1532 | default: ShouldNotReachHere(); |
aoqi@1 | 1533 | } |
aoqi@1 | 1534 | } |
aoqi@1 | 1535 | } else { |
aoqi@1 | 1536 | out_sig_bt[argc++] = in_sig_bt[i]; |
aoqi@1 | 1537 | in_elem_bt[i] = T_VOID; |
aoqi@1 | 1538 | } |
aoqi@1 | 1539 | if (in_sig_bt[i] != T_VOID) { |
aoqi@1 | 1540 | assert(in_sig_bt[i] == ss.type(), "must match"); |
aoqi@1 | 1541 | ss.next(); |
aoqi@1 | 1542 | } |
aoqi@1 | 1543 | } |
aoqi@1 | 1544 | } |
aoqi@1 | 1545 | |
aoqi@1 | 1546 | // Now figure out where the args must be stored and how much stack space |
aoqi@1 | 1547 | // they require (neglecting out_preserve_stack_slots but space for storing |
aoqi@1 | 1548 | // the 1st six register arguments). It's weird see int_stk_helper. |
aoqi@1 | 1549 | // |
aoqi@1 | 1550 | int out_arg_slots; |
aoqi@6880 | 1551 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args); |
aoqi@1 | 1552 | |
aoqi@1 | 1553 | // Compute framesize for the wrapper. We need to handlize all oops in |
aoqi@1 | 1554 | // registers. We must create space for them here that is disjoint from |
aoqi@1 | 1555 | // the windowed save area because we have no control over when we might |
aoqi@1 | 1556 | // flush the window again and overwrite values that gc has since modified. |
aoqi@1 | 1557 | // (The live window race) |
aoqi@1 | 1558 | // |
aoqi@1 | 1559 | // We always just allocate 6 word for storing down these object. This allow |
aoqi@1 | 1560 | // us to simply record the base and use the Ireg number to decide which |
aoqi@1 | 1561 | // slot to use. (Note that the reg number is the inbound number not the |
aoqi@1 | 1562 | // outbound number). |
aoqi@1 | 1563 | // We must shuffle args to match the native convention, and include var-args space. |
aoqi@1 | 1564 | |
aoqi@1 | 1565 | // Calculate the total number of stack slots we will need. |
aoqi@1 | 1566 | |
aoqi@1 | 1567 | // First count the abi requirement plus all of the outgoing args |
aoqi@1 | 1568 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
aoqi@1 | 1569 | |
aoqi@1 | 1570 | // Now the space for the inbound oop handle area |
aoqi@1 | 1571 | int total_save_slots = 9 * VMRegImpl::slots_per_word; // 9 arguments passed in registers |
aoqi@1 | 1572 | if (is_critical_native) { |
aoqi@1 | 1573 | // Critical natives may have to call out so they need a save area |
aoqi@1 | 1574 | // for register arguments. |
aoqi@1 | 1575 | int double_slots = 0; |
aoqi@1 | 1576 | int single_slots = 0; |
aoqi@1 | 1577 | for ( int i = 0; i < total_in_args; i++) { |
aoqi@1 | 1578 | if (in_regs[i].first()->is_Register()) { |
aoqi@1 | 1579 | const Register reg = in_regs[i].first()->as_Register(); |
aoqi@1 | 1580 | switch (in_sig_bt[i]) { |
aoqi@1 | 1581 | case T_BOOLEAN: |
aoqi@1 | 1582 | case T_BYTE: |
aoqi@1 | 1583 | case T_SHORT: |
aoqi@1 | 1584 | case T_CHAR: |
aoqi@1 | 1585 | case T_INT: single_slots++; break; |
aoqi@1 | 1586 | case T_ARRAY: // specific to LP64 (7145024) |
aoqi@1 | 1587 | case T_LONG: double_slots++; break; |
aoqi@1 | 1588 | default: ShouldNotReachHere(); |
aoqi@1 | 1589 | } |
aoqi@1 | 1590 | } else if (in_regs[i].first()->is_FloatRegister()) { |
aoqi@1 | 1591 | switch (in_sig_bt[i]) { |
aoqi@1 | 1592 | case T_FLOAT: single_slots++; break; |
aoqi@1 | 1593 | case T_DOUBLE: double_slots++; break; |
aoqi@1 | 1594 | default: ShouldNotReachHere(); |
aoqi@1 | 1595 | } |
aoqi@1 | 1596 | } |
aoqi@1 | 1597 | } |
aoqi@1 | 1598 | total_save_slots = double_slots * 2 + single_slots; |
aoqi@1 | 1599 | // align the save area |
aoqi@1 | 1600 | if (double_slots != 0) { |
aoqi@1 | 1601 | stack_slots = round_to(stack_slots, 2); |
aoqi@1 | 1602 | } |
aoqi@1 | 1603 | } |
aoqi@1 | 1604 | |
aoqi@1 | 1605 | int oop_handle_offset = stack_slots; |
aoqi@1 | 1606 | stack_slots += total_save_slots; |
aoqi@1 | 1607 | |
aoqi@1 | 1608 | // Now any space we need for handlizing a klass if static method |
aoqi@1 | 1609 | |
aoqi@6880 | 1610 | int klass_slot_offset = 0; |
aoqi@6880 | 1611 | int klass_offset = -1; |
aoqi@6880 | 1612 | int lock_slot_offset = 0; |
aoqi@6880 | 1613 | bool is_static = false; |
aoqi@1 | 1614 | |
aoqi@1 | 1615 | if (method->is_static()) { |
aoqi@1 | 1616 | klass_slot_offset = stack_slots; |
aoqi@1 | 1617 | stack_slots += VMRegImpl::slots_per_word; |
aoqi@1 | 1618 | klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; |
aoqi@1 | 1619 | is_static = true; |
aoqi@1 | 1620 | } |
aoqi@1 | 1621 | |
aoqi@1 | 1622 | // Plus a lock if needed |
aoqi@1 | 1623 | |
aoqi@1 | 1624 | if (method->is_synchronized()) { |
aoqi@1 | 1625 | lock_slot_offset = stack_slots; |
aoqi@1 | 1626 | stack_slots += VMRegImpl::slots_per_word; |
aoqi@1 | 1627 | } |
aoqi@1 | 1628 | |
aoqi@1 | 1629 | // Now a place to save return value or as a temporary for any gpr -> fpr moves |
aoqi@9459 | 1630 | // + 2 for return address (which we own) and saved fp |
aoqi@6880 | 1631 | stack_slots += 2 + 9 * VMRegImpl::slots_per_word; // (T0, A0, A1, A2, A3, A4, A5, A6, A7) |
aoqi@1 | 1632 | |
aoqi@1 | 1633 | // Ok The space we have allocated will look like: |
aoqi@1 | 1634 | // |
aoqi@1 | 1635 | // |
aoqi@1 | 1636 | // FP-> | | |
aoqi@1 | 1637 | // |---------------------| |
aoqi@1 | 1638 | // | 2 slots for moves | |
aoqi@1 | 1639 | // |---------------------| |
aoqi@1 | 1640 | // | lock box (if sync) | |
aoqi@1 | 1641 | // |---------------------| <- lock_slot_offset |
aoqi@1 | 1642 | // | klass (if static) | |
aoqi@1 | 1643 | // |---------------------| <- klass_slot_offset |
aoqi@1 | 1644 | // | oopHandle area | |
aoqi@1 | 1645 | // |---------------------| <- oop_handle_offset |
aoqi@1 | 1646 | // | outbound memory | |
aoqi@1 | 1647 | // | based arguments | |
aoqi@1 | 1648 | // | | |
aoqi@1 | 1649 | // |---------------------| |
aoqi@1 | 1650 | // | vararg area | |
aoqi@1 | 1651 | // |---------------------| |
aoqi@1 | 1652 | // | | |
aoqi@1 | 1653 | // SP-> | out_preserved_slots | |
aoqi@1 | 1654 | // |
aoqi@1 | 1655 | // |
aoqi@1 | 1656 | |
aoqi@1 | 1657 | |
aoqi@1 | 1658 | // Now compute actual number of stack words we need rounding to make |
aoqi@1 | 1659 | // stack properly aligned. |
aoqi@1 | 1660 | stack_slots = round_to(stack_slots, StackAlignmentInSlots); |
aoqi@1 | 1661 | |
aoqi@1 | 1662 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
aoqi@1 | 1663 | |
aoqi@6880 | 1664 | intptr_t start = (intptr_t)__ pc(); |
aoqi@6880 | 1665 | |
aoqi@6880 | 1666 | |
aoqi@6880 | 1667 | |
aoqi@6880 | 1668 | // First thing make an ic check to see if we should even be here |
aoqi@6880 | 1669 | address ic_miss = SharedRuntime::get_ic_miss_stub(); |
aoqi@6880 | 1670 | |
aoqi@6880 | 1671 | // We are free to use all registers as temps without saving them and |
aoqi@9459 | 1672 | // restoring them except fp. fp is the only callee save register |
aoqi@6880 | 1673 | // as far as the interpreter and the compiler(s) are concerned. |
aoqi@1 | 1674 | |
aoqi@1 | 1675 | //refer to register_mips.hpp:IC_Klass |
aoqi@6880 | 1676 | const Register ic_reg = T1; |
aoqi@6880 | 1677 | const Register receiver = T0; |
aoqi@6880 | 1678 | |
aoqi@6880 | 1679 | Label hit; |
aoqi@6880 | 1680 | Label exception_pending; |
aoqi@6880 | 1681 | |
aoqi@6880 | 1682 | __ verify_oop(receiver); |
aoqi@6880 | 1683 | //add for compressedoops |
aoqi@6880 | 1684 | __ load_klass(T9, receiver); |
aoqi@6880 | 1685 | __ beq(T9, ic_reg, hit); |
aoqi@6880 | 1686 | __ delayed()->nop(); |
aoqi@6880 | 1687 | __ jmp(ic_miss, relocInfo::runtime_call_type); |
aoqi@6880 | 1688 | __ delayed()->nop(); |
aoqi@6880 | 1689 | // verified entry must be aligned for code patching. |
aoqi@6880 | 1690 | // and the first 5 bytes must be in the same cache line |
aoqi@6880 | 1691 | // if we align at 8 then we will be sure 5 bytes are in the same line |
aoqi@6880 | 1692 | __ align(8); |
aoqi@6880 | 1693 | |
aoqi@6880 | 1694 | __ bind(hit); |
aoqi@6880 | 1695 | |
aoqi@6880 | 1696 | |
aoqi@6880 | 1697 | int vep_offset = ((intptr_t)__ pc()) - start; |
aoqi@1 | 1698 | #ifdef COMPILER1 |
aoqi@6880 | 1699 | if (InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) { |
aoqi@6880 | 1700 | // Object.hashCode can pull the hashCode from the header word |
aoqi@6880 | 1701 | // instead of doing a full VM transition once it's been computed. |
aoqi@6880 | 1702 | // Since hashCode is usually polymorphic at call sites we can't do |
aoqi@6880 | 1703 | // this optimization at the call site without a lot of work. |
aoqi@6880 | 1704 | Label slowCase; |
aoqi@6880 | 1705 | Register receiver = T0; |
aoqi@6880 | 1706 | Register result = V0; |
aoqi@6880 | 1707 | __ ld ( result, receiver, oopDesc::mark_offset_in_bytes()); |
aoqi@6880 | 1708 | // check if locked |
aoqi@6880 | 1709 | __ andi(AT, result, markOopDesc::unlocked_value); |
aoqi@6880 | 1710 | __ beq(AT, R0, slowCase); |
aoqi@6880 | 1711 | __ delayed()->nop(); |
aoqi@6880 | 1712 | if (UseBiasedLocking) { |
aoqi@6880 | 1713 | // Check if biased and fall through to runtime if so |
aoqi@6880 | 1714 | __ andi (AT, result, markOopDesc::biased_lock_bit_in_place); |
huangjia@9645 | 1715 | __ bne(AT, R0, slowCase); |
aoqi@6880 | 1716 | __ delayed()->nop(); |
aoqi@6880 | 1717 | } |
aoqi@6880 | 1718 | // get hash |
aoqi@6880 | 1719 | __ li(AT, markOopDesc::hash_mask_in_place); |
aoqi@6880 | 1720 | __ andr (AT, result, AT); |
aoqi@6880 | 1721 | // test if hashCode exists |
aoqi@6880 | 1722 | __ beq (AT, R0, slowCase); |
aoqi@6880 | 1723 | __ delayed()->nop(); |
aoqi@6880 | 1724 | __ shr(result, markOopDesc::hash_shift); |
aoqi@6880 | 1725 | __ jr(RA); |
aoqi@6880 | 1726 | __ delayed()->nop(); |
aoqi@6880 | 1727 | __ bind (slowCase); |
aoqi@6880 | 1728 | } |
aoqi@1 | 1729 | #endif // COMPILER1 |
aoqi@1 | 1730 | |
aoqi@6880 | 1731 | // The instruction at the verified entry point must be 5 bytes or longer |
aoqi@6880 | 1732 | // because it can be patched on the fly by make_non_entrant. The stack bang |
aoqi@6880 | 1733 | // instruction fits that requirement. |
aoqi@6880 | 1734 | |
aoqi@6880 | 1735 | // Generate stack overflow check |
aoqi@6880 | 1736 | |
aoqi@6880 | 1737 | if (UseStackBanging) { |
aoqi@6880 | 1738 | __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
aoqi@6880 | 1739 | } else { |
aoqi@6880 | 1740 | // need a 5 byte instruction to allow MT safe patching to non-entrant |
aoqi@6880 | 1741 | __ nop(); |
aoqi@6880 | 1742 | __ nop(); |
aoqi@6880 | 1743 | __ nop(); |
aoqi@6880 | 1744 | __ nop(); |
aoqi@6880 | 1745 | __ nop(); |
aoqi@6880 | 1746 | } |
aoqi@6880 | 1747 | // Generate a new frame for the wrapper. |
aoqi@6880 | 1748 | // do mips need this ? |
aoqi@1 | 1749 | #ifndef OPT_THREAD |
aoqi@6880 | 1750 | __ get_thread(TREG); |
aoqi@1 | 1751 | #endif |
aoqi@6880 | 1752 | __ st_ptr(SP, TREG, in_bytes(JavaThread::last_Java_sp_offset())); |
aoqi@6880 | 1753 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@6880 | 1754 | __ andr(SP, SP, AT); |
aoqi@6880 | 1755 | |
aoqi@6880 | 1756 | __ enter(); |
aoqi@9459 | 1757 | // -2 because return address is already present and so is saved fp |
aoqi@6880 | 1758 | __ addiu(SP, SP, -1 * (stack_size - 2*wordSize)); |
aoqi@6880 | 1759 | |
aoqi@6880 | 1760 | // Frame is now completed as far a size and linkage. |
aoqi@6880 | 1761 | |
aoqi@6880 | 1762 | int frame_complete = ((intptr_t)__ pc()) - start; |
aoqi@6880 | 1763 | |
aoqi@9459 | 1764 | // Calculate the difference between sp and fp. We need to know it |
aoqi@6880 | 1765 | // after the native call because on windows Java Natives will pop |
aoqi@9459 | 1766 | // the arguments and it is painful to do sp relative addressing |
aoqi@6880 | 1767 | // in a platform independent way. So after the call we switch to |
aoqi@9459 | 1768 | // fp relative addressing. |
aoqi@9459 | 1769 | //FIXME actually , the fp_adjustment may not be the right, because andr(sp, sp, at) may change |
aoqi@7997 | 1770 | //the SP |
aoqi@6880 | 1771 | int fp_adjustment = stack_size - 2*wordSize; |
aoqi@1 | 1772 | |
aoqi@1 | 1773 | #ifdef COMPILER2 |
aoqi@6880 | 1774 | // C2 may leave the stack dirty if not in SSE2+ mode |
aoqi@6880 | 1775 | __ empty_FPU_stack(); |
aoqi@9459 | 1776 | #endif |
aoqi@9459 | 1777 | |
aoqi@9459 | 1778 | // Compute the fp offset for any slots used after the jni call |
aoqi@9459 | 1779 | |
aoqi@9459 | 1780 | int lock_slot_fp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; |
aoqi@9459 | 1781 | // We use TREG as a thread pointer because it is callee save and |
aoqi@6880 | 1782 | // if we load it once it is usable thru the entire wrapper |
aoqi@6880 | 1783 | const Register thread = TREG; |
aoqi@6880 | 1784 | |
aoqi@9459 | 1785 | // We use S4 as the oop handle for the receiver/klass |
aoqi@6880 | 1786 | // It is callee save so it survives the call to native |
aoqi@6880 | 1787 | |
aoqi@6880 | 1788 | const Register oop_handle_reg = S4; |
aoqi@1 | 1789 | if (is_critical_native) { |
aoqi@1 | 1790 | __ stop("generate_native_wrapper in sharedRuntime <2>"); |
aoqi@9459 | 1791 | //TODO:Fu |
aoqi@9459 | 1792 | // check_needs_gc_for_critical_native(masm, stack_slots, total_c_args, total_in_args, |
aoqi@9459 | 1793 | // oop_handle_offset, oop_maps, in_regs, in_sig_bt); |
aoqi@1 | 1794 | } |
aoqi@1 | 1795 | |
aoqi@1 | 1796 | #ifndef OPT_THREAD |
aoqi@6880 | 1797 | __ get_thread(thread); |
aoqi@1 | 1798 | #endif |
aoqi@1 | 1799 | |
aoqi@1 | 1800 | // |
aoqi@1 | 1801 | // We immediately shuffle the arguments so that any vm call we have to |
aoqi@1 | 1802 | // make from here on out (sync slow path, jvmpi, etc.) we will have |
aoqi@1 | 1803 | // captured the oops from our caller and have a valid oopMap for |
aoqi@1 | 1804 | // them. |
aoqi@1 | 1805 | |
aoqi@1 | 1806 | // ----------------- |
aoqi@6880 | 1807 | // The Grand Shuffle |
aoqi@1 | 1808 | // |
aoqi@1 | 1809 | // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv* |
aoqi@1 | 1810 | // and, if static, the class mirror instead of a receiver. This pretty much |
aoqi@1 | 1811 | // guarantees that register layout will not match (and mips doesn't use reg |
aoqi@1 | 1812 | // parms though amd does). Since the native abi doesn't use register args |
aoqi@1 | 1813 | // and the java conventions does we don't have to worry about collisions. |
aoqi@1 | 1814 | // All of our moved are reg->stack or stack->stack. |
aoqi@1 | 1815 | // We ignore the extra arguments during the shuffle and handle them at the |
aoqi@1 | 1816 | // last moment. The shuffle is described by the two calling convention |
aoqi@1 | 1817 | // vectors we have in our possession. We simply walk the java vector to |
aoqi@1 | 1818 | // get the source locations and the c vector to get the destinations. |
aoqi@1 | 1819 | |
aoqi@6880 | 1820 | int c_arg = method->is_static() ? 2 : 1 ; |
aoqi@6880 | 1821 | |
aoqi@9459 | 1822 | // Record sp-based slot for receiver on stack for non-static methods |
aoqi@6880 | 1823 | int receiver_offset = -1; |
aoqi@6880 | 1824 | |
aoqi@6880 | 1825 | // This is a trick. We double the stack slots so we can claim |
aoqi@6880 | 1826 | // the oops in the caller's frame. Since we are sure to have |
aoqi@6880 | 1827 | // more args than the caller doubling is enough to make |
aoqi@6880 | 1828 | // sure we can capture all the incoming oop args from the |
aoqi@6880 | 1829 | // caller. |
aoqi@6880 | 1830 | // |
aoqi@6880 | 1831 | OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); |
aoqi@1 | 1832 | |
aoqi@9459 | 1833 | // Mark location of fp (someday) |
aoqi@9459 | 1834 | // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(fp)); |
aoqi@1 | 1835 | |
aoqi@1 | 1836 | #ifdef ASSERT |
aoqi@1 | 1837 | bool reg_destroyed[RegisterImpl::number_of_registers]; |
aoqi@1 | 1838 | bool freg_destroyed[FloatRegisterImpl::number_of_registers]; |
aoqi@1 | 1839 | for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { |
aoqi@1 | 1840 | reg_destroyed[r] = false; |
aoqi@1 | 1841 | } |
aoqi@1 | 1842 | for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) { |
aoqi@1 | 1843 | freg_destroyed[f] = false; |
aoqi@1 | 1844 | } |
aoqi@1 | 1845 | |
aoqi@1 | 1846 | #endif /* ASSERT */ |
aoqi@1 | 1847 | |
aoqi@1 | 1848 | // This may iterate in two different directions depending on the |
aoqi@1 | 1849 | // kind of native it is. The reason is that for regular JNI natives |
aoqi@1 | 1850 | // the incoming and outgoing registers are offset upwards and for |
aoqi@1 | 1851 | // critical natives they are offset down. |
aoqi@1 | 1852 | GrowableArray<int> arg_order(2 * total_in_args); |
aoqi@1 | 1853 | VMRegPair tmp_vmreg; |
aoqi@1 | 1854 | tmp_vmreg.set1(T8->as_VMReg()); |
aoqi@1 | 1855 | |
aoqi@1 | 1856 | if (!is_critical_native) { |
aoqi@1 | 1857 | for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) { |
aoqi@1 | 1858 | arg_order.push(i); |
aoqi@1 | 1859 | arg_order.push(c_arg); |
aoqi@1 | 1860 | } |
aoqi@1 | 1861 | } else { |
aoqi@1 | 1862 | // Compute a valid move order, using tmp_vmreg to break any cycles |
aoqi@1 | 1863 | __ stop("generate_native_wrapper in sharedRuntime <2>"); |
aoqi@9459 | 1864 | //TODO:Fu |
aoqi@9459 | 1865 | // ComputeMoveOrder cmo(total_in_args, in_regs, total_c_args, out_regs, in_sig_bt, arg_order, tmp_vmreg); |
aoqi@1 | 1866 | } |
aoqi@1 | 1867 | |
aoqi@1 | 1868 | int temploc = -1; |
aoqi@1 | 1869 | for (int ai = 0; ai < arg_order.length(); ai += 2) { |
aoqi@1 | 1870 | int i = arg_order.at(ai); |
aoqi@1 | 1871 | int c_arg = arg_order.at(ai + 1); |
aoqi@1 | 1872 | __ block_comment(err_msg("move %d -> %d", i, c_arg)); |
aoqi@1 | 1873 | if (c_arg == -1) { |
aoqi@1 | 1874 | assert(is_critical_native, "should only be required for critical natives"); |
aoqi@1 | 1875 | // This arg needs to be moved to a temporary |
aoqi@1 | 1876 | __ move(tmp_vmreg.first()->as_Register(), in_regs[i].first()->as_Register()); |
aoqi@1 | 1877 | in_regs[i] = tmp_vmreg; |
aoqi@1 | 1878 | temploc = i; |
aoqi@1 | 1879 | continue; |
aoqi@1 | 1880 | } else if (i == -1) { |
aoqi@1 | 1881 | assert(is_critical_native, "should only be required for critical natives"); |
aoqi@1 | 1882 | // Read from the temporary location |
aoqi@1 | 1883 | assert(temploc != -1, "must be valid"); |
aoqi@1 | 1884 | i = temploc; |
aoqi@1 | 1885 | temploc = -1; |
aoqi@1 | 1886 | } |
aoqi@1 | 1887 | #ifdef ASSERT |
aoqi@1 | 1888 | if (in_regs[i].first()->is_Register()) { |
aoqi@1 | 1889 | assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!"); |
aoqi@1 | 1890 | } else if (in_regs[i].first()->is_FloatRegister()) { |
aoqi@1 | 1891 | assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!"); |
aoqi@1 | 1892 | } |
aoqi@1 | 1893 | if (out_regs[c_arg].first()->is_Register()) { |
aoqi@1 | 1894 | reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; |
aoqi@1 | 1895 | } else if (out_regs[c_arg].first()->is_FloatRegister()) { |
aoqi@1 | 1896 | freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true; |
aoqi@1 | 1897 | } |
aoqi@1 | 1898 | #endif /* ASSERT */ |
aoqi@1 | 1899 | switch (in_sig_bt[i]) { |
aoqi@1 | 1900 | case T_ARRAY: |
aoqi@1 | 1901 | if (is_critical_native) { |
aoqi@9459 | 1902 | __ stop("generate_native_wrapper in sharedRuntime <2>"); |
aoqi@9459 | 1903 | //TODO:Fu |
aoqi@9459 | 1904 | // unpack_array_argument(masm, in_regs[i], in_elem_bt[i], out_regs[c_arg + 1], out_regs[c_arg]); |
aoqi@1 | 1905 | c_arg++; |
aoqi@1 | 1906 | #ifdef ASSERT |
aoqi@1 | 1907 | if (out_regs[c_arg].first()->is_Register()) { |
aoqi@1 | 1908 | reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; |
aoqi@1 | 1909 | } else if (out_regs[c_arg].first()->is_FloatRegister()) { |
aoqi@1 | 1910 | freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true; |
aoqi@1 | 1911 | } |
aoqi@1 | 1912 | #endif |
aoqi@1 | 1913 | break; |
aoqi@1 | 1914 | } |
aoqi@1 | 1915 | case T_OBJECT: |
aoqi@1 | 1916 | assert(!is_critical_native, "no oop arguments"); |
aoqi@1 | 1917 | object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], |
aoqi@1 | 1918 | ((i == 0) && (!is_static)), |
aoqi@1 | 1919 | &receiver_offset); |
aoqi@1 | 1920 | break; |
aoqi@1 | 1921 | case T_VOID: |
aoqi@1 | 1922 | break; |
aoqi@1 | 1923 | |
aoqi@1 | 1924 | case T_FLOAT: |
aoqi@1 | 1925 | float_move(masm, in_regs[i], out_regs[c_arg]); |
aoqi@1 | 1926 | break; |
aoqi@1 | 1927 | |
aoqi@1 | 1928 | case T_DOUBLE: |
aoqi@1 | 1929 | assert( i + 1 < total_in_args && |
aoqi@1 | 1930 | in_sig_bt[i + 1] == T_VOID && |
aoqi@1 | 1931 | out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
aoqi@1 | 1932 | double_move(masm, in_regs[i], out_regs[c_arg]); |
aoqi@1 | 1933 | break; |
aoqi@1 | 1934 | |
aoqi@1 | 1935 | case T_LONG : |
aoqi@1 | 1936 | long_move(masm, in_regs[i], out_regs[c_arg]); |
aoqi@1 | 1937 | break; |
aoqi@1 | 1938 | |
aoqi@1 | 1939 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
aoqi@1 | 1940 | |
aoqi@1 | 1941 | default: |
aoqi@1 | 1942 | simple_move32(masm, in_regs[i], out_regs[c_arg]); |
aoqi@1 | 1943 | } |
aoqi@1 | 1944 | } |
aoqi@1 | 1945 | |
aoqi@1 | 1946 | // point c_arg at the first arg that is already loaded in case we |
aoqi@1 | 1947 | // need to spill before we call out |
aoqi@9459 | 1948 | c_arg = total_c_args - total_in_args; |
aoqi@9459 | 1949 | // Pre-load a static method's oop. Used both by locking code and |
aoqi@6880 | 1950 | // the normal JNI call code. |
aoqi@6880 | 1951 | |
aoqi@6880 | 1952 | __ move(oop_handle_reg, A1); |
aoqi@6880 | 1953 | |
aoqi@6880 | 1954 | if (method->is_static() && !is_critical_native) { |
aoqi@6880 | 1955 | |
aoqi@6880 | 1956 | // load opp into a register |
aoqi@6880 | 1957 | int oop_index = __ oop_recorder()->find_index(JNIHandles::make_local( |
aoqi@6880 | 1958 | (method->method_holder())->java_mirror())); |
aoqi@6880 | 1959 | |
aoqi@6880 | 1960 | |
aoqi@6880 | 1961 | RelocationHolder rspec = oop_Relocation::spec(oop_index); |
aoqi@6880 | 1962 | __ relocate(rspec); |
aoqi@6880 | 1963 | __ patchable_set48(oop_handle_reg, (long)JNIHandles::make_local((method->method_holder())->java_mirror())); |
aoqi@6880 | 1964 | // Now handlize the static class mirror it's known not-null. |
aoqi@6880 | 1965 | __ sd( oop_handle_reg, SP, klass_offset); |
aoqi@6880 | 1966 | map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); |
aoqi@6880 | 1967 | |
aoqi@6880 | 1968 | // Now get the handle |
aoqi@6880 | 1969 | __ lea(oop_handle_reg, Address(SP, klass_offset)); |
aoqi@6880 | 1970 | // store the klass handle as second argument |
aoqi@6880 | 1971 | __ move(A1, oop_handle_reg); |
aoqi@6880 | 1972 | // and protect the arg if we must spill |
aoqi@6880 | 1973 | c_arg--; |
aoqi@6880 | 1974 | } |
aoqi@6880 | 1975 | |
aoqi@1 | 1976 | // Change state to native (we save the return address in the thread, since it might not |
aoqi@1 | 1977 | // be pushed on the stack when we do a a stack traversal). It is enough that the pc() |
aoqi@1 | 1978 | // points into the right code segment. It does not have to be the correct return pc. |
aoqi@1 | 1979 | // We use the same pc/oopMap repeatedly when we call out |
aoqi@1 | 1980 | |
aoqi@6880 | 1981 | intptr_t the_pc = (intptr_t) __ pc(); |
aoqi@6880 | 1982 | oop_maps->add_gc_map(the_pc - start, map); |
aoqi@6880 | 1983 | |
aoqi@6880 | 1984 | __ set_last_Java_frame(SP, noreg, NULL); |
aoqi@6880 | 1985 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 1986 | { |
aoqi@6880 | 1987 | intptr_t save_pc = (intptr_t)the_pc ; |
aoqi@6880 | 1988 | __ patchable_set48(AT, save_pc); |
aoqi@6880 | 1989 | } |
aoqi@6880 | 1990 | __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
aoqi@6880 | 1991 | |
aoqi@6880 | 1992 | |
aoqi@6880 | 1993 | // We have all of the arguments setup at this point. We must not touch any register |
aoqi@6880 | 1994 | // argument registers at this point (what if we save/restore them there are no oop? |
aoqi@6880 | 1995 | { |
aoqi@6880 | 1996 | SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
aoqi@6880 | 1997 | int metadata_index = __ oop_recorder()->find_index(method()); |
aoqi@6880 | 1998 | RelocationHolder rspec = metadata_Relocation::spec(metadata_index); |
aoqi@6880 | 1999 | __ relocate(rspec); |
aoqi@6880 | 2000 | __ patchable_set48(AT, (long)(method())); |
aoqi@6880 | 2001 | |
aoqi@6880 | 2002 | __ call_VM_leaf( |
aoqi@6880 | 2003 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), |
aoqi@6880 | 2004 | thread, AT); |
aoqi@6880 | 2005 | |
aoqi@6880 | 2006 | } |
aoqi@6880 | 2007 | |
aoqi@6880 | 2008 | // These are register definitions we need for locking/unlocking |
aoqi@9459 | 2009 | const Register swap_reg = T8; // Must use T8 for cmpxchg instruction |
aoqi@6880 | 2010 | const Register obj_reg = T9; // Will contain the oop |
aoqi@6880 | 2011 | //const Register lock_reg = T6; // Address of compiler lock object (BasicLock) |
aoqi@6880 | 2012 | const Register lock_reg = c_rarg0; // Address of compiler lock object (BasicLock) |
aoqi@6880 | 2013 | |
aoqi@6880 | 2014 | |
aoqi@6880 | 2015 | |
aoqi@6880 | 2016 | Label slow_path_lock; |
aoqi@6880 | 2017 | Label lock_done; |
aoqi@6880 | 2018 | |
aoqi@6880 | 2019 | // Lock a synchronized method |
aoqi@6880 | 2020 | if (method->is_synchronized()) { |
aoqi@6880 | 2021 | assert(!is_critical_native, "unhandled"); |
aoqi@6880 | 2022 | |
aoqi@6880 | 2023 | const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); |
aoqi@6880 | 2024 | |
aoqi@6880 | 2025 | // Get the handle (the 2nd argument) |
aoqi@6880 | 2026 | __ move(oop_handle_reg, A1); |
aoqi@6880 | 2027 | |
aoqi@6880 | 2028 | // Get address of the box |
aoqi@9459 | 2029 | __ lea(lock_reg, Address(FP, lock_slot_fp_offset)); |
aoqi@6880 | 2030 | |
aoqi@6880 | 2031 | // Load the oop from the handle |
aoqi@6880 | 2032 | __ ld(obj_reg, oop_handle_reg, 0); |
aoqi@6880 | 2033 | |
aoqi@6880 | 2034 | if (UseBiasedLocking) { |
aoqi@6880 | 2035 | // Note that oop_handle_reg is trashed during this call |
aoqi@6880 | 2036 | __ biased_locking_enter(lock_reg, obj_reg, swap_reg, A1, false, lock_done, &slow_path_lock); |
aoqi@6880 | 2037 | } |
aoqi@6880 | 2038 | |
aoqi@9459 | 2039 | // Load immediate 1 into swap_reg %T8 |
aoqi@6880 | 2040 | __ move(swap_reg, 1); |
aoqi@6880 | 2041 | |
aoqi@6880 | 2042 | __ ld(AT, obj_reg, 0); |
aoqi@6880 | 2043 | __ orr(swap_reg, swap_reg, AT); |
aoqi@6880 | 2044 | |
aoqi@6880 | 2045 | __ sd( swap_reg, lock_reg, mark_word_offset); |
aoqi@6880 | 2046 | __ cmpxchg(lock_reg, Address(obj_reg, 0), swap_reg); |
aoqi@6880 | 2047 | __ bne(AT, R0, lock_done); |
aoqi@6880 | 2048 | __ delayed()->nop(); |
aoqi@6880 | 2049 | // Test if the oopMark is an obvious stack pointer, i.e., |
aoqi@6880 | 2050 | // 1) (mark & 3) == 0, and |
aoqi@9459 | 2051 | // 2) sp <= mark < mark + os::pagesize() |
aoqi@6880 | 2052 | // These 3 tests can be done by evaluating the following |
aoqi@9459 | 2053 | // expression: ((mark - sp) & (3 - os::vm_page_size())), |
aoqi@6880 | 2054 | // assuming both stack pointer and pagesize have their |
aoqi@6880 | 2055 | // least significant 2 bits clear. |
aoqi@9459 | 2056 | // NOTE: the oopMark is in swap_reg %T8 as the result of cmpxchg |
aoqi@6880 | 2057 | |
aoqi@8009 | 2058 | __ dsub(swap_reg, swap_reg, SP); |
aoqi@8009 | 2059 | __ move(AT, 3 - os::vm_page_size()); |
aoqi@6880 | 2060 | __ andr(swap_reg , swap_reg, AT); |
aoqi@6880 | 2061 | // Save the test result, for recursive case, the result is zero |
aoqi@6880 | 2062 | __ sd(swap_reg, lock_reg, mark_word_offset); |
aoqi@8009 | 2063 | __ bne(swap_reg, R0, slow_path_lock); |
aoqi@6880 | 2064 | __ delayed()->nop(); |
aoqi@6880 | 2065 | // Slow path will re-enter here |
aoqi@6880 | 2066 | __ bind(lock_done); |
aoqi@6880 | 2067 | |
aoqi@6880 | 2068 | if (UseBiasedLocking) { |
aoqi@6880 | 2069 | // Re-fetch oop_handle_reg as we trashed it above |
aoqi@6880 | 2070 | __ move(A1, oop_handle_reg); |
aoqi@6880 | 2071 | } |
aoqi@6880 | 2072 | } |
aoqi@6880 | 2073 | |
aoqi@6880 | 2074 | |
aoqi@6880 | 2075 | // Finally just about ready to make the JNI call |
aoqi@6880 | 2076 | |
aoqi@6880 | 2077 | |
aoqi@6880 | 2078 | // get JNIEnv* which is first argument to native |
aoqi@1 | 2079 | if (!is_critical_native) { |
aoqi@8009 | 2080 | __ addi(A0, thread, in_bytes(JavaThread::jni_environment_offset())); |
aoqi@1 | 2081 | } |
aoqi@1 | 2082 | |
aoqi@6880 | 2083 | // Example: Java_java_lang_ref_Finalizer_invokeFinalizeMethod(JNIEnv *env, jclass clazz, jobject ob) |
aoqi@9459 | 2084 | // Load the second arguments into A1 |
aoqi@6880 | 2085 | //__ ld(A1, SP , wordSize ); // klass |
aoqi@6880 | 2086 | |
aoqi@6880 | 2087 | // Now set thread in native |
aoqi@6880 | 2088 | __ addi(AT, R0, _thread_in_native); |
aoqi@6880 | 2089 | __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
aoqi@9459 | 2090 | // do the call |
aoqi@6880 | 2091 | __ call(method->native_function(), relocInfo::runtime_call_type); |
aoqi@6880 | 2092 | __ delayed()->nop(); |
aoqi@6880 | 2093 | // WARNING - on Windows Java Natives use pascal calling convention and pop the |
aoqi@6880 | 2094 | // arguments off of the stack. We could just re-adjust the stack pointer here |
aoqi@6880 | 2095 | // and continue to do SP relative addressing but we instead switch to FP |
aoqi@6880 | 2096 | // relative addressing. |
aoqi@6880 | 2097 | |
aoqi@6880 | 2098 | // Unpack native results. |
aoqi@6880 | 2099 | switch (ret_type) { |
aoqi@6880 | 2100 | case T_BOOLEAN: __ c2bool(V0); break; |
huangjia@9645 | 2101 | case T_CHAR : __ andi(V0, V0, 0xFFFF); break; |
aoqi@6880 | 2102 | case T_BYTE : __ sign_extend_byte (V0); break; |
aoqi@6880 | 2103 | case T_SHORT : __ sign_extend_short(V0); break; |
aoqi@6880 | 2104 | case T_INT : // nothing to do break; |
aoqi@6880 | 2105 | case T_DOUBLE : |
aoqi@6880 | 2106 | case T_FLOAT : |
aoqi@6880 | 2107 | // Result is in st0 we'll save as needed |
aoqi@6880 | 2108 | break; |
aoqi@6880 | 2109 | case T_ARRAY: // Really a handle |
aoqi@6880 | 2110 | case T_OBJECT: // Really a handle |
aoqi@6880 | 2111 | break; // can't de-handlize until after safepoint check |
aoqi@6880 | 2112 | case T_VOID: break; |
aoqi@6880 | 2113 | case T_LONG: break; |
aoqi@6880 | 2114 | default : ShouldNotReachHere(); |
aoqi@6880 | 2115 | } |
aoqi@6880 | 2116 | // Switch thread to "native transition" state before reading the synchronization state. |
aoqi@6880 | 2117 | // This additional state is necessary because reading and testing the synchronization |
aoqi@6880 | 2118 | // state is not atomic w.r.t. GC, as this scenario demonstrates: |
aoqi@6880 | 2119 | // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. |
aoqi@6880 | 2120 | // VM thread changes sync state to synchronizing and suspends threads for GC. |
aoqi@6880 | 2121 | // Thread A is resumed to finish this native method, but doesn't block here since it |
aoqi@6880 | 2122 | // didn't see any synchronization is progress, and escapes. |
aoqi@6880 | 2123 | __ addi(AT, R0, _thread_in_native_trans); |
aoqi@6880 | 2124 | __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
aoqi@6880 | 2125 | |
aoqi@8009 | 2126 | //if(os::is_MP()) {} |
aoqi@8009 | 2127 | |
aoqi@1 | 2128 | Label after_transition; |
aoqi@1 | 2129 | |
aoqi@6880 | 2130 | // check for safepoint operation in progress and/or pending suspend requests |
aoqi@8009 | 2131 | { |
aoqi@8009 | 2132 | Label Continue; |
aoqi@6880 | 2133 | __ li(AT, SafepointSynchronize::address_of_state()); |
aoqi@6880 | 2134 | __ lw(A0, AT, 0); |
aoqi@6880 | 2135 | __ addi(AT, A0, -SafepointSynchronize::_not_synchronized); |
aoqi@6880 | 2136 | Label L; |
huangjia@9645 | 2137 | __ bne(AT, R0, L); |
aoqi@6880 | 2138 | __ delayed()->nop(); |
aoqi@6880 | 2139 | __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset())); |
aoqi@6880 | 2140 | __ beq(AT, R0, Continue); |
aoqi@6880 | 2141 | __ delayed()->nop(); |
aoqi@6880 | 2142 | __ bind(L); |
aoqi@6880 | 2143 | |
aoqi@6880 | 2144 | // Don't use call_VM as it will see a possible pending exception and forward it |
aoqi@6880 | 2145 | // and never return here preventing us from clearing _last_native_pc down below. |
aoqi@6880 | 2146 | // |
aoqi@6880 | 2147 | save_native_result(masm, ret_type, stack_slots); |
aoqi@8009 | 2148 | __ move(A0, thread); |
aoqi@8009 | 2149 | __ addi(SP, SP, -wordSize); |
aoqi@21 | 2150 | __ push(S2); |
aoqi@21 | 2151 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@21 | 2152 | __ move(S2, SP); // use S2 as a sender SP holder |
aoqi@21 | 2153 | __ andr(SP, SP, AT); // align stack as required by ABI |
aoqi@1 | 2154 | if (!is_critical_native) { |
aoqi@1 | 2155 | __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::runtime_call_type); |
aoqi@6880 | 2156 | __ delayed()->nop(); |
aoqi@1 | 2157 | } else { |
aoqi@1 | 2158 | __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans_and_transition), relocInfo::runtime_call_type); |
aoqi@6880 | 2159 | __ delayed()->nop(); |
aoqi@1 | 2160 | } |
aoqi@21 | 2161 | __ move(SP, S2); // use S2 as a sender SP holder |
aoqi@21 | 2162 | __ pop(S2); |
huangjia@9645 | 2163 | __ addi(SP, SP, wordSize); |
aoqi@6880 | 2164 | //add for compressedoops |
aoqi@6880 | 2165 | __ reinit_heapbase(); |
aoqi@6880 | 2166 | // Restore any method result value |
aoqi@6880 | 2167 | restore_native_result(masm, ret_type, stack_slots); |
aoqi@1 | 2168 | |
aoqi@1 | 2169 | if (is_critical_native) { |
aoqi@1 | 2170 | // The call above performed the transition to thread_in_Java so |
aoqi@1 | 2171 | // skip the transition logic below. |
aoqi@1 | 2172 | __ beq(R0, R0, after_transition); |
aoqi@6880 | 2173 | __ delayed()->nop(); |
aoqi@1 | 2174 | } |
aoqi@1 | 2175 | |
aoqi@6880 | 2176 | __ bind(Continue); |
aoqi@6880 | 2177 | } |
aoqi@6880 | 2178 | |
aoqi@6880 | 2179 | // change thread state |
aoqi@6880 | 2180 | __ addi(AT, R0, _thread_in_Java); |
aoqi@6880 | 2181 | __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
aoqi@1 | 2182 | __ bind(after_transition); |
aoqi@6880 | 2183 | Label reguard; |
aoqi@6880 | 2184 | Label reguard_done; |
aoqi@6880 | 2185 | __ lw(AT, thread, in_bytes(JavaThread::stack_guard_state_offset())); |
aoqi@6880 | 2186 | __ addi(AT, AT, -JavaThread::stack_guard_yellow_disabled); |
aoqi@6880 | 2187 | __ beq(AT, R0, reguard); |
aoqi@6880 | 2188 | __ delayed()->nop(); |
aoqi@6880 | 2189 | // slow path reguard re-enters here |
aoqi@6880 | 2190 | __ bind(reguard_done); |
aoqi@6880 | 2191 | |
aoqi@6880 | 2192 | // Handle possible exception (will unlock if necessary) |
aoqi@6880 | 2193 | |
aoqi@6880 | 2194 | // native result if any is live |
aoqi@6880 | 2195 | |
aoqi@6880 | 2196 | // Unlock |
aoqi@6880 | 2197 | Label slow_path_unlock; |
aoqi@6880 | 2198 | Label unlock_done; |
aoqi@6880 | 2199 | if (method->is_synchronized()) { |
aoqi@6880 | 2200 | |
aoqi@6880 | 2201 | Label done; |
aoqi@6880 | 2202 | |
aoqi@6880 | 2203 | // Get locked oop from the handle we passed to jni |
aoqi@6880 | 2204 | __ ld( obj_reg, oop_handle_reg, 0); |
aoqi@6880 | 2205 | if (UseBiasedLocking) { |
aoqi@6880 | 2206 | __ biased_locking_exit(obj_reg, T8, done); |
aoqi@6880 | 2207 | |
aoqi@6880 | 2208 | } |
aoqi@6880 | 2209 | |
aoqi@6880 | 2210 | // Simple recursive lock? |
aoqi@6880 | 2211 | |
aoqi@9459 | 2212 | __ ld(AT, FP, lock_slot_fp_offset); |
aoqi@6880 | 2213 | __ beq(AT, R0, done); |
aoqi@6880 | 2214 | __ delayed()->nop(); |
aoqi@9459 | 2215 | // Must save FSF if if it is live now because cmpxchg must use it |
aoqi@6880 | 2216 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
aoqi@6880 | 2217 | save_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2218 | } |
aoqi@6880 | 2219 | |
aoqi@6880 | 2220 | // get old displaced header |
aoqi@9459 | 2221 | __ ld (T8, FP, lock_slot_fp_offset); |
aoqi@6880 | 2222 | // get address of the stack lock |
aoqi@9459 | 2223 | __ addi (c_rarg0, FP, lock_slot_fp_offset); |
aoqi@6880 | 2224 | // Atomic swap old header if oop still contains the stack lock |
aoqi@6880 | 2225 | __ cmpxchg(T8, Address(obj_reg, 0), c_rarg0); |
aoqi@6880 | 2226 | |
aoqi@6880 | 2227 | __ beq(AT, R0, slow_path_unlock); |
aoqi@6880 | 2228 | __ delayed()->nop(); |
aoqi@6880 | 2229 | // slow path re-enters here |
aoqi@6880 | 2230 | __ bind(unlock_done); |
aoqi@6880 | 2231 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
aoqi@6880 | 2232 | restore_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2233 | } |
aoqi@6880 | 2234 | |
aoqi@6880 | 2235 | __ bind(done); |
aoqi@6880 | 2236 | |
aoqi@6880 | 2237 | } |
aoqi@6880 | 2238 | { |
aoqi@6880 | 2239 | SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
aoqi@6880 | 2240 | // Tell dtrace about this method exit |
aoqi@6880 | 2241 | save_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2242 | int metadata_index = __ oop_recorder()->find_index( (method())); |
aoqi@6880 | 2243 | RelocationHolder rspec = metadata_Relocation::spec(metadata_index); |
aoqi@6880 | 2244 | __ relocate(rspec); |
aoqi@6880 | 2245 | __ patchable_set48(AT, (long)(method())); |
aoqi@6880 | 2246 | |
aoqi@6880 | 2247 | __ call_VM_leaf( |
aoqi@6880 | 2248 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
aoqi@6880 | 2249 | thread, AT); |
aoqi@6880 | 2250 | restore_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2251 | } |
aoqi@6880 | 2252 | |
aoqi@6880 | 2253 | // We can finally stop using that last_Java_frame we setup ages ago |
aoqi@6880 | 2254 | |
fujie@9171 | 2255 | __ reset_last_Java_frame(false); |
aoqi@6880 | 2256 | |
huangjia@9705 | 2257 | // Unpack oop result, e.g. JNIHandles::resolve value. |
aoqi@6880 | 2258 | if (ret_type == T_OBJECT || ret_type == T_ARRAY) { |
huangjia@9705 | 2259 | __ resolve_jobject(V0, thread, T9); |
aoqi@6880 | 2260 | } |
aoqi@1 | 2261 | |
aoqi@1 | 2262 | if (!is_critical_native) { |
aoqi@8009 | 2263 | // reset handle block |
aoqi@8009 | 2264 | __ ld(AT, thread, in_bytes(JavaThread::active_handles_offset())); |
aoqi@8009 | 2265 | __ sw(R0, AT, JNIHandleBlock::top_offset_in_bytes()); |
aoqi@1 | 2266 | } |
aoqi@1 | 2267 | |
aoqi@1 | 2268 | if (!is_critical_native) { |
aoqi@6880 | 2269 | // Any exception pending? |
aoqi@6880 | 2270 | __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 2271 | __ bne(AT, R0, exception_pending); |
aoqi@6880 | 2272 | __ delayed()->nop(); |
aoqi@1 | 2273 | } |
aoqi@6880 | 2274 | // no exception, we're almost done |
aoqi@6880 | 2275 | |
aoqi@6880 | 2276 | // check that only result value is on FPU stack |
aoqi@6880 | 2277 | __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit"); |
aoqi@6880 | 2278 | |
aoqi@1 | 2279 | // Return |
aoqi@1 | 2280 | #ifndef OPT_THREAD |
aoqi@6880 | 2281 | __ get_thread(TREG); |
aoqi@1 | 2282 | #endif |
aoqi@8009 | 2283 | //__ ld_ptr(SP, TREG, in_bytes(JavaThread::last_Java_sp_offset())); |
aoqi@6880 | 2284 | __ leave(); |
aoqi@6880 | 2285 | |
aoqi@6880 | 2286 | __ jr(RA); |
aoqi@6880 | 2287 | __ delayed()->nop(); |
aoqi@6880 | 2288 | // Unexpected paths are out of line and go here |
aoqi@6880 | 2289 | // Slow path locking & unlocking |
aoqi@6880 | 2290 | if (method->is_synchronized()) { |
aoqi@6880 | 2291 | |
aoqi@6880 | 2292 | // BEGIN Slow path lock |
aoqi@6880 | 2293 | __ bind(slow_path_lock); |
aoqi@6880 | 2294 | |
aoqi@6880 | 2295 | // protect the args we've loaded |
aoqi@6880 | 2296 | save_args(masm, total_c_args, c_arg, out_regs); |
aoqi@6880 | 2297 | |
aoqi@6880 | 2298 | // has last_Java_frame setup. No exceptions so do vanilla call not call_VM |
aoqi@6880 | 2299 | // args are (oop obj, BasicLock* lock, JavaThread* thread) |
aoqi@6880 | 2300 | |
aoqi@6880 | 2301 | __ move(A0, obj_reg); |
aoqi@6880 | 2302 | __ move(A1, lock_reg); |
aoqi@6880 | 2303 | __ move(A2, thread); |
aoqi@6880 | 2304 | __ addi(SP, SP, - 3*wordSize); |
aoqi@6880 | 2305 | |
aoqi@6880 | 2306 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@6880 | 2307 | __ move(S2, SP); // use S2 as a sender SP holder |
aoqi@6880 | 2308 | __ andr(SP, SP, AT); // align stack as required by ABI |
aoqi@6880 | 2309 | |
aoqi@6880 | 2310 | __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), relocInfo::runtime_call_type); |
aoqi@6880 | 2311 | __ delayed()->nop(); |
aoqi@6880 | 2312 | __ move(SP, S2); |
aoqi@6880 | 2313 | __ addi(SP, SP, 3*wordSize); |
aoqi@6880 | 2314 | |
aoqi@6880 | 2315 | restore_args(masm, total_c_args, c_arg, out_regs); |
aoqi@6880 | 2316 | |
aoqi@6880 | 2317 | #ifdef ASSERT |
aoqi@6880 | 2318 | { Label L; |
aoqi@6880 | 2319 | __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 2320 | __ beq(AT, R0, L); |
aoqi@6880 | 2321 | __ delayed()->nop(); |
aoqi@6880 | 2322 | __ stop("no pending exception allowed on exit from monitorenter"); |
aoqi@6880 | 2323 | __ bind(L); |
aoqi@6880 | 2324 | } |
aoqi@6880 | 2325 | #endif |
aoqi@6880 | 2326 | __ b(lock_done); |
aoqi@6880 | 2327 | __ delayed()->nop(); |
aoqi@6880 | 2328 | // END Slow path lock |
aoqi@6880 | 2329 | |
aoqi@6880 | 2330 | // BEGIN Slow path unlock |
aoqi@6880 | 2331 | __ bind(slow_path_unlock); |
aoqi@6880 | 2332 | |
aoqi@6880 | 2333 | // Slow path unlock |
aoqi@6880 | 2334 | |
aoqi@6880 | 2335 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
aoqi@6880 | 2336 | save_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2337 | } |
aoqi@6880 | 2338 | // Save pending exception around call to VM (which contains an EXCEPTION_MARK) |
aoqi@6880 | 2339 | |
aoqi@6880 | 2340 | __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 2341 | __ push(AT); |
aoqi@6880 | 2342 | __ sd(R0, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@1 | 2343 | |
aoqi@1 | 2344 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@1 | 2345 | __ move(S2, SP); // use S2 as a sender SP holder |
aoqi@1 | 2346 | __ andr(SP, SP, AT); // align stack as required by ABI |
aoqi@1 | 2347 | |
aoqi@6880 | 2348 | // should be a peal |
aoqi@6880 | 2349 | // +wordSize because of the push above |
aoqi@9459 | 2350 | __ addi(A1, FP, lock_slot_fp_offset); |
aoqi@6880 | 2351 | |
aoqi@6880 | 2352 | __ move(A0, obj_reg); |
aoqi@6880 | 2353 | __ addi(SP,SP, -2*wordSize); |
aoqi@6880 | 2354 | __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C), |
aoqi@6880 | 2355 | relocInfo::runtime_call_type); |
aoqi@6880 | 2356 | __ delayed()->nop(); |
huangjia@9645 | 2357 | __ addi(SP, SP, 2*wordSize); |
aoqi@1 | 2358 | __ move(SP, S2); |
aoqi@6880 | 2359 | //add for compressedoops |
aoqi@6880 | 2360 | __ reinit_heapbase(); |
aoqi@1 | 2361 | #ifdef ASSERT |
aoqi@6880 | 2362 | { |
aoqi@6880 | 2363 | Label L; |
aoqi@6880 | 2364 | __ lw( AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 2365 | __ beq(AT, R0, L); |
aoqi@6880 | 2366 | __ delayed()->nop(); |
aoqi@6880 | 2367 | __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); |
aoqi@6880 | 2368 | __ bind(L); |
aoqi@6880 | 2369 | } |
aoqi@1 | 2370 | #endif /* ASSERT */ |
aoqi@1 | 2371 | |
aoqi@6880 | 2372 | __ pop(AT); |
aoqi@6880 | 2373 | __ sd(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 2374 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
aoqi@6880 | 2375 | restore_native_result(masm, ret_type, stack_slots); |
aoqi@1 | 2376 | } |
aoqi@6880 | 2377 | __ b(unlock_done); |
aoqi@6880 | 2378 | __ delayed()->nop(); |
aoqi@6880 | 2379 | // END Slow path unlock |
aoqi@6880 | 2380 | |
aoqi@6880 | 2381 | } |
aoqi@6880 | 2382 | |
aoqi@6880 | 2383 | // SLOW PATH Reguard the stack if needed |
aoqi@6880 | 2384 | |
aoqi@6880 | 2385 | __ bind(reguard); |
aoqi@6880 | 2386 | save_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2387 | __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), |
aoqi@6880 | 2388 | relocInfo::runtime_call_type); |
aoqi@6880 | 2389 | __ delayed()->nop(); |
aoqi@6880 | 2390 | //add for compressedoops |
aoqi@6880 | 2391 | __ reinit_heapbase(); |
aoqi@6880 | 2392 | restore_native_result(masm, ret_type, stack_slots); |
aoqi@6880 | 2393 | __ b(reguard_done); |
aoqi@6880 | 2394 | __ delayed()->nop(); |
aoqi@6880 | 2395 | |
aoqi@6880 | 2396 | // BEGIN EXCEPTION PROCESSING |
aoqi@6880 | 2397 | if (!is_critical_native) { |
aoqi@6880 | 2398 | // Forward the exception |
aoqi@6880 | 2399 | __ bind(exception_pending); |
aoqi@6880 | 2400 | |
aoqi@6880 | 2401 | // remove possible return value from FPU register stack |
aoqi@6880 | 2402 | __ empty_FPU_stack(); |
aoqi@6880 | 2403 | |
aoqi@6880 | 2404 | // pop our frame |
aoqi@8009 | 2405 | //forward_exception_entry need return address on stack |
aoqi@8009 | 2406 | __ addiu(SP, FP, wordSize); |
aoqi@6880 | 2407 | __ ld(FP, SP, (-1) * wordSize); |
aoqi@6880 | 2408 | |
aoqi@6880 | 2409 | // and forward the exception |
aoqi@6880 | 2410 | __ jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); |
aoqi@6880 | 2411 | __ delayed()->nop(); |
aoqi@6880 | 2412 | } |
aoqi@6880 | 2413 | __ flush(); |
aoqi@6880 | 2414 | |
aoqi@6880 | 2415 | nmethod *nm = nmethod::new_native_nmethod(method, |
aoqi@6880 | 2416 | compile_id, |
aoqi@6880 | 2417 | masm->code(), |
aoqi@6880 | 2418 | vep_offset, |
aoqi@6880 | 2419 | frame_complete, |
aoqi@6880 | 2420 | stack_slots / VMRegImpl::slots_per_word, |
aoqi@6880 | 2421 | (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), |
aoqi@6880 | 2422 | in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), |
aoqi@6880 | 2423 | oop_maps); |
aoqi@1 | 2424 | |
aoqi@1 | 2425 | if (is_critical_native) { |
aoqi@1 | 2426 | nm->set_lazy_critical_native(true); |
aoqi@1 | 2427 | } |
aoqi@6880 | 2428 | |
aoqi@6880 | 2429 | return nm; |
aoqi@1 | 2430 | |
aoqi@1 | 2431 | } |
aoqi@1 | 2432 | |
aoqi@1 | 2433 | #ifdef HAVE_DTRACE_H |
aoqi@1 | 2434 | // --------------------------------------------------------------------------- |
aoqi@1 | 2435 | // Generate a dtrace nmethod for a given signature. The method takes arguments |
aoqi@1 | 2436 | // in the Java compiled code convention, marshals them to the native |
aoqi@1 | 2437 | // abi and then leaves nops at the position you would expect to call a native |
aoqi@1 | 2438 | // function. When the probe is enabled the nops are replaced with a trap |
aoqi@1 | 2439 | // instruction that dtrace inserts and the trace will cause a notification |
aoqi@1 | 2440 | // to dtrace. |
aoqi@1 | 2441 | // |
aoqi@1 | 2442 | // The probes are only able to take primitive types and java/lang/String as |
aoqi@1 | 2443 | // arguments. No other java types are allowed. Strings are converted to utf8 |
aoqi@1 | 2444 | // strings so that from dtrace point of view java strings are converted to C |
aoqi@1 | 2445 | // strings. There is an arbitrary fixed limit on the total space that a method |
aoqi@1 | 2446 | // can use for converting the strings. (256 chars per string in the signature). |
aoqi@1 | 2447 | // So any java string larger then this is truncated. |
aoqi@1 | 2448 | |
aoqi@1 | 2449 | static int fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 }; |
aoqi@1 | 2450 | static bool offsets_initialized = false; |
aoqi@1 | 2451 | |
aoqi@1 | 2452 | static VMRegPair reg64_to_VMRegPair(Register r) { |
aoqi@1 | 2453 | VMRegPair ret; |
aoqi@1 | 2454 | if (wordSize == 8) { |
aoqi@1 | 2455 | ret.set2(r->as_VMReg()); |
aoqi@1 | 2456 | } else { |
aoqi@1 | 2457 | ret.set_pair(r->successor()->as_VMReg(), r->as_VMReg()); |
aoqi@1 | 2458 | } |
aoqi@1 | 2459 | return ret; |
aoqi@1 | 2460 | } |
aoqi@1 | 2461 | |
aoqi@1 | 2462 | |
aoqi@6880 | 2463 | nmethod *SharedRuntime::generate_dtrace_nmethod(MacroAssembler *masm, |
aoqi@6880 | 2464 | methodHandle method) { |
aoqi@1 | 2465 | |
aoqi@1 | 2466 | |
aoqi@1 | 2467 | // generate_dtrace_nmethod is guarded by a mutex so we are sure to |
aoqi@1 | 2468 | // be single threaded in this method. |
aoqi@1 | 2469 | assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); |
aoqi@1 | 2470 | |
aoqi@1 | 2471 | // Fill in the signature array, for the calling-convention call. |
aoqi@1 | 2472 | int total_args_passed = method->size_of_parameters(); |
aoqi@1 | 2473 | |
aoqi@1 | 2474 | BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); |
aoqi@1 | 2475 | VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); |
aoqi@1 | 2476 | |
aoqi@1 | 2477 | // The signature we are going to use for the trap that dtrace will see |
aoqi@1 | 2478 | // java/lang/String is converted. We drop "this" and any other object |
aoqi@1 | 2479 | // is converted to NULL. (A one-slot java/lang/Long object reference |
aoqi@1 | 2480 | // is converted to a two-slot long, which is why we double the allocation). |
aoqi@1 | 2481 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); |
aoqi@1 | 2482 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); |
aoqi@1 | 2483 | |
aoqi@1 | 2484 | int i=0; |
aoqi@1 | 2485 | int total_strings = 0; |
aoqi@1 | 2486 | int first_arg_to_pass = 0; |
aoqi@1 | 2487 | int total_c_args = 0; |
aoqi@1 | 2488 | |
aoqi@1 | 2489 | // Skip the receiver as dtrace doesn't want to see it |
aoqi@1 | 2490 | if( !method->is_static() ) { |
aoqi@1 | 2491 | in_sig_bt[i++] = T_OBJECT; |
aoqi@1 | 2492 | first_arg_to_pass = 1; |
aoqi@1 | 2493 | } |
aoqi@1 | 2494 | |
aoqi@1 | 2495 | SignatureStream ss(method->signature()); |
aoqi@1 | 2496 | for ( ; !ss.at_return_type(); ss.next()) { |
aoqi@1 | 2497 | BasicType bt = ss.type(); |
aoqi@1 | 2498 | in_sig_bt[i++] = bt; // Collect remaining bits of signature |
aoqi@1 | 2499 | out_sig_bt[total_c_args++] = bt; |
aoqi@1 | 2500 | if( bt == T_OBJECT) { |
aoqi@1 | 2501 | symbolOop s = ss.as_symbol_or_null(); |
aoqi@1 | 2502 | if (s == vmSymbols::java_lang_String()) { |
aoqi@1 | 2503 | total_strings++; |
aoqi@1 | 2504 | out_sig_bt[total_c_args-1] = T_ADDRESS; |
aoqi@1 | 2505 | } else if (s == vmSymbols::java_lang_Boolean() || |
aoqi@1 | 2506 | s == vmSymbols::java_lang_Byte()) { |
aoqi@1 | 2507 | out_sig_bt[total_c_args-1] = T_BYTE; |
aoqi@1 | 2508 | } else if (s == vmSymbols::java_lang_Character() || |
aoqi@1 | 2509 | s == vmSymbols::java_lang_Short()) { |
aoqi@1 | 2510 | out_sig_bt[total_c_args-1] = T_SHORT; |
aoqi@1 | 2511 | } else if (s == vmSymbols::java_lang_Integer() || |
aoqi@1 | 2512 | s == vmSymbols::java_lang_Float()) { |
aoqi@1 | 2513 | out_sig_bt[total_c_args-1] = T_INT; |
aoqi@1 | 2514 | } else if (s == vmSymbols::java_lang_Long() || |
aoqi@1 | 2515 | s == vmSymbols::java_lang_Double()) { |
aoqi@1 | 2516 | out_sig_bt[total_c_args-1] = T_LONG; |
aoqi@1 | 2517 | out_sig_bt[total_c_args++] = T_VOID; |
aoqi@1 | 2518 | } |
aoqi@1 | 2519 | } else if ( bt == T_LONG || bt == T_DOUBLE ) { |
aoqi@1 | 2520 | in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots |
aoqi@1 | 2521 | // We convert double to long |
aoqi@1 | 2522 | out_sig_bt[total_c_args-1] = T_LONG; |
aoqi@1 | 2523 | out_sig_bt[total_c_args++] = T_VOID; |
aoqi@1 | 2524 | } else if ( bt == T_FLOAT) { |
aoqi@1 | 2525 | // We convert float to int |
aoqi@1 | 2526 | out_sig_bt[total_c_args-1] = T_INT; |
aoqi@1 | 2527 | } |
aoqi@1 | 2528 | } |
aoqi@1 | 2529 | |
aoqi@1 | 2530 | assert(i==total_args_passed, "validly parsed signature"); |
aoqi@1 | 2531 | |
aoqi@1 | 2532 | // Now get the compiled-Java layout as input arguments |
aoqi@1 | 2533 | int comp_args_on_stack; |
aoqi@1 | 2534 | comp_args_on_stack = SharedRuntime::java_calling_convention( |
aoqi@1 | 2535 | in_sig_bt, in_regs, total_args_passed, false); |
aoqi@1 | 2536 | |
aoqi@1 | 2537 | // We have received a description of where all the java arg are located |
aoqi@1 | 2538 | // on entry to the wrapper. We need to convert these args to where |
aoqi@1 | 2539 | // the a native (non-jni) function would expect them. To figure out |
aoqi@1 | 2540 | // where they go we convert the java signature to a C signature and remove |
aoqi@1 | 2541 | // T_VOID for any long/double we might have received. |
aoqi@1 | 2542 | |
aoqi@1 | 2543 | |
aoqi@1 | 2544 | // Now figure out where the args must be stored and how much stack space |
aoqi@1 | 2545 | // they require (neglecting out_preserve_stack_slots but space for storing |
aoqi@1 | 2546 | // the 1st six register arguments). It's weird see int_stk_helper. |
aoqi@6880 | 2547 | |
aoqi@1 | 2548 | int out_arg_slots; |
aoqi@1 | 2549 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args); |
aoqi@1 | 2550 | |
aoqi@1 | 2551 | // Calculate the total number of stack slots we will need. |
aoqi@1 | 2552 | |
aoqi@1 | 2553 | // First count the abi requirement plus all of the outgoing args |
aoqi@1 | 2554 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
aoqi@1 | 2555 | |
aoqi@1 | 2556 | // Plus a temp for possible converion of float/double/long register args |
aoqi@1 | 2557 | |
aoqi@1 | 2558 | int conversion_temp = stack_slots; |
aoqi@1 | 2559 | stack_slots += 2; |
aoqi@1 | 2560 | |
aoqi@1 | 2561 | |
aoqi@1 | 2562 | // Now space for the string(s) we must convert |
aoqi@1 | 2563 | |
aoqi@1 | 2564 | int string_locs = stack_slots; |
aoqi@1 | 2565 | stack_slots += total_strings * |
aoqi@1 | 2566 | (max_dtrace_string_size / VMRegImpl::stack_slot_size); |
aoqi@1 | 2567 | |
aoqi@1 | 2568 | // Ok The space we have allocated will look like: |
aoqi@1 | 2569 | // |
aoqi@1 | 2570 | // |
aoqi@1 | 2571 | // FP-> | | |
aoqi@1 | 2572 | // |---------------------| |
aoqi@1 | 2573 | // | string[n] | |
aoqi@1 | 2574 | // |---------------------| <- string_locs[n] |
aoqi@1 | 2575 | // | string[n-1] | |
aoqi@1 | 2576 | // |---------------------| <- string_locs[n-1] |
aoqi@1 | 2577 | // | ... | |
aoqi@1 | 2578 | // | ... | |
aoqi@1 | 2579 | // |---------------------| <- string_locs[1] |
aoqi@1 | 2580 | // | string[0] | |
aoqi@1 | 2581 | // |---------------------| <- string_locs[0] |
aoqi@1 | 2582 | // | temp | |
aoqi@1 | 2583 | // |---------------------| <- conversion_temp |
aoqi@1 | 2584 | // | outbound memory | |
aoqi@1 | 2585 | // | based arguments | |
aoqi@1 | 2586 | // | | |
aoqi@1 | 2587 | // |---------------------| |
aoqi@1 | 2588 | // | | |
aoqi@1 | 2589 | // SP-> | out_preserved_slots | |
aoqi@1 | 2590 | // |
aoqi@1 | 2591 | // |
aoqi@1 | 2592 | |
aoqi@1 | 2593 | // Now compute actual number of stack words we need rounding to make |
aoqi@1 | 2594 | // stack properly aligned. |
aoqi@1 | 2595 | stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word); |
aoqi@1 | 2596 | |
aoqi@1 | 2597 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
aoqi@1 | 2598 | |
aoqi@1 | 2599 | intptr_t start = (intptr_t)__ pc(); |
aoqi@1 | 2600 | |
aoqi@1 | 2601 | // First thing make an ic check to see if we should even be here |
aoqi@1 | 2602 | |
aoqi@1 | 2603 | { |
aoqi@1 | 2604 | Label L; |
aoqi@1 | 2605 | const Register temp_reg = G3_scratch; |
aoqi@1 | 2606 | Address ic_miss(temp_reg, SharedRuntime::get_ic_miss_stub()); |
aoqi@1 | 2607 | __ verify_oop(O0); |
aoqi@1 | 2608 | __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), temp_reg); |
aoqi@1 | 2609 | __ cmp(temp_reg, G5_inline_cache_reg); |
aoqi@1 | 2610 | __ brx(Assembler::equal, true, Assembler::pt, L); |
aoqi@1 | 2611 | __ delayed()->nop(); |
aoqi@1 | 2612 | |
aoqi@1 | 2613 | __ jump_to(ic_miss, 0); |
aoqi@1 | 2614 | __ delayed()->nop(); |
aoqi@1 | 2615 | __ align(CodeEntryAlignment); |
aoqi@1 | 2616 | __ bind(L); |
aoqi@1 | 2617 | } |
aoqi@1 | 2618 | |
aoqi@1 | 2619 | int vep_offset = ((intptr_t)__ pc()) - start; |
aoqi@1 | 2620 | |
aoqi@1 | 2621 | |
aoqi@1 | 2622 | // The instruction at the verified entry point must be 5 bytes or longer |
aoqi@1 | 2623 | // because it can be patched on the fly by make_non_entrant. The stack bang |
aoqi@1 | 2624 | // instruction fits that requirement. |
aoqi@1 | 2625 | |
aoqi@1 | 2626 | // Generate stack overflow check before creating frame |
aoqi@1 | 2627 | __ generate_stack_overflow_check(stack_size); |
aoqi@1 | 2628 | |
aoqi@1 | 2629 | assert(((intptr_t)__ pc() - start - vep_offset) >= 5, |
aoqi@1 | 2630 | "valid size for make_non_entrant"); |
aoqi@1 | 2631 | |
aoqi@1 | 2632 | // Generate a new frame for the wrapper. |
aoqi@1 | 2633 | __ save(SP, -stack_size, SP); |
aoqi@1 | 2634 | |
aoqi@1 | 2635 | // Frame is now completed as far a size and linkage. |
aoqi@1 | 2636 | |
aoqi@1 | 2637 | int frame_complete = ((intptr_t)__ pc()) - start; |
aoqi@1 | 2638 | |
aoqi@1 | 2639 | #ifdef ASSERT |
aoqi@1 | 2640 | bool reg_destroyed[RegisterImpl::number_of_registers]; |
aoqi@1 | 2641 | bool freg_destroyed[FloatRegisterImpl::number_of_registers]; |
aoqi@1 | 2642 | for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { |
aoqi@1 | 2643 | reg_destroyed[r] = false; |
aoqi@1 | 2644 | } |
aoqi@1 | 2645 | for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) { |
aoqi@1 | 2646 | freg_destroyed[f] = false; |
aoqi@1 | 2647 | } |
aoqi@1 | 2648 | |
aoqi@1 | 2649 | #endif /* ASSERT */ |
aoqi@1 | 2650 | |
aoqi@1 | 2651 | VMRegPair zero; |
aoqi@1 | 2652 | const Register g0 = G0; // without this we get a compiler warning (why??) |
aoqi@1 | 2653 | zero.set2(g0->as_VMReg()); |
aoqi@1 | 2654 | |
aoqi@1 | 2655 | int c_arg, j_arg; |
aoqi@1 | 2656 | |
aoqi@1 | 2657 | Register conversion_off = noreg; |
aoqi@1 | 2658 | |
aoqi@1 | 2659 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
aoqi@1 | 2660 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
aoqi@1 | 2661 | |
aoqi@1 | 2662 | VMRegPair src = in_regs[j_arg]; |
aoqi@1 | 2663 | VMRegPair dst = out_regs[c_arg]; |
aoqi@1 | 2664 | |
aoqi@1 | 2665 | #ifdef ASSERT |
aoqi@1 | 2666 | if (src.first()->is_Register()) { |
aoqi@1 | 2667 | assert(!reg_destroyed[src.first()->as_Register()->encoding()], "ack!"); |
aoqi@1 | 2668 | } else if (src.first()->is_FloatRegister()) { |
aoqi@1 | 2669 | assert(!freg_destroyed[src.first()->as_FloatRegister()->encoding( |
aoqi@1 | 2670 | FloatRegisterImpl::S)], "ack!"); |
aoqi@1 | 2671 | } |
aoqi@1 | 2672 | if (dst.first()->is_Register()) { |
aoqi@1 | 2673 | reg_destroyed[dst.first()->as_Register()->encoding()] = true; |
aoqi@1 | 2674 | } else if (dst.first()->is_FloatRegister()) { |
aoqi@1 | 2675 | freg_destroyed[dst.first()->as_FloatRegister()->encoding( |
aoqi@1 | 2676 | FloatRegisterImpl::S)] = true; |
aoqi@1 | 2677 | } |
aoqi@1 | 2678 | #endif /* ASSERT */ |
aoqi@1 | 2679 | |
aoqi@1 | 2680 | switch (in_sig_bt[j_arg]) { |
aoqi@1 | 2681 | case T_ARRAY: |
aoqi@1 | 2682 | case T_OBJECT: |
aoqi@1 | 2683 | { |
aoqi@1 | 2684 | if (out_sig_bt[c_arg] == T_BYTE || out_sig_bt[c_arg] == T_SHORT || |
aoqi@1 | 2685 | out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { |
aoqi@1 | 2686 | // need to unbox a one-slot value |
aoqi@1 | 2687 | Register in_reg = L0; |
aoqi@1 | 2688 | Register tmp = L2; |
aoqi@1 | 2689 | if ( src.first()->is_reg() ) { |
aoqi@1 | 2690 | in_reg = src.first()->as_Register(); |
aoqi@1 | 2691 | } else { |
aoqi@1 | 2692 | assert(Assembler::is_simm13(reg2offset(src.first()) + STACK_BIAS), |
aoqi@1 | 2693 | "must be"); |
aoqi@1 | 2694 | __ ld_ptr(FP, reg2offset(src.first()) + STACK_BIAS, in_reg); |
aoqi@1 | 2695 | } |
aoqi@1 | 2696 | // If the final destination is an acceptable register |
aoqi@1 | 2697 | if ( dst.first()->is_reg() ) { |
aoqi@1 | 2698 | if ( dst.is_single_phys_reg() || out_sig_bt[c_arg] != T_LONG ) { |
aoqi@1 | 2699 | tmp = dst.first()->as_Register(); |
aoqi@1 | 2700 | } |
aoqi@1 | 2701 | } |
aoqi@1 | 2702 | |
aoqi@1 | 2703 | Label skipUnbox; |
aoqi@1 | 2704 | if ( wordSize == 4 && out_sig_bt[c_arg] == T_LONG ) { |
aoqi@1 | 2705 | __ mov(G0, tmp->successor()); |
aoqi@1 | 2706 | } |
aoqi@1 | 2707 | __ br_null(in_reg, true, Assembler::pn, skipUnbox); |
aoqi@1 | 2708 | __ delayed()->mov(G0, tmp); |
aoqi@1 | 2709 | |
aoqi@1 | 2710 | BasicType bt = out_sig_bt[c_arg]; |
aoqi@1 | 2711 | int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); |
aoqi@1 | 2712 | switch (bt) { |
aoqi@1 | 2713 | case T_BYTE: |
aoqi@1 | 2714 | __ ldub(in_reg, box_offset, tmp); break; |
aoqi@1 | 2715 | case T_SHORT: |
aoqi@1 | 2716 | __ lduh(in_reg, box_offset, tmp); break; |
aoqi@1 | 2717 | case T_INT: |
aoqi@1 | 2718 | __ ld(in_reg, box_offset, tmp); break; |
aoqi@1 | 2719 | case T_LONG: |
aoqi@1 | 2720 | __ ld_long(in_reg, box_offset, tmp); break; |
aoqi@1 | 2721 | default: ShouldNotReachHere(); |
aoqi@1 | 2722 | } |
aoqi@1 | 2723 | |
aoqi@1 | 2724 | __ bind(skipUnbox); |
aoqi@1 | 2725 | // If tmp wasn't final destination copy to final destination |
aoqi@1 | 2726 | if (tmp == L2) { |
aoqi@1 | 2727 | VMRegPair tmp_as_VM = reg64_to_VMRegPair(L2); |
aoqi@1 | 2728 | if (out_sig_bt[c_arg] == T_LONG) { |
aoqi@1 | 2729 | long_move(masm, tmp_as_VM, dst); |
aoqi@1 | 2730 | } else { |
aoqi@1 | 2731 | move32_64(masm, tmp_as_VM, out_regs[c_arg]); |
aoqi@1 | 2732 | } |
aoqi@1 | 2733 | } |
aoqi@1 | 2734 | if (out_sig_bt[c_arg] == T_LONG) { |
aoqi@1 | 2735 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
aoqi@1 | 2736 | ++c_arg; // move over the T_VOID to keep the loop indices in sync |
aoqi@1 | 2737 | } |
aoqi@1 | 2738 | } else if (out_sig_bt[c_arg] == T_ADDRESS) { |
aoqi@1 | 2739 | Register s = |
aoqi@1 | 2740 | src.first()->is_reg() ? src.first()->as_Register() : L2; |
aoqi@1 | 2741 | Register d = |
aoqi@1 | 2742 | dst.first()->is_reg() ? dst.first()->as_Register() : L2; |
aoqi@1 | 2743 | |
aoqi@1 | 2744 | // We store the oop now so that the conversion pass can reach |
aoqi@1 | 2745 | // while in the inner frame. This will be the only store if |
aoqi@1 | 2746 | // the oop is NULL. |
aoqi@1 | 2747 | if (s != L2) { |
aoqi@1 | 2748 | // src is register |
aoqi@1 | 2749 | if (d != L2) { |
aoqi@1 | 2750 | // dst is register |
aoqi@1 | 2751 | __ mov(s, d); |
aoqi@1 | 2752 | } else { |
aoqi@1 | 2753 | assert(Assembler::is_simm13(reg2offset(dst.first()) + |
aoqi@1 | 2754 | STACK_BIAS), "must be"); |
aoqi@1 | 2755 | __ st_ptr(s, SP, reg2offset(dst.first()) + STACK_BIAS); |
aoqi@1 | 2756 | } |
aoqi@1 | 2757 | } else { |
aoqi@1 | 2758 | // src not a register |
aoqi@1 | 2759 | assert(Assembler::is_simm13(reg2offset(src.first()) + |
aoqi@1 | 2760 | STACK_BIAS), "must be"); |
aoqi@1 | 2761 | __ ld_ptr(FP, reg2offset(src.first()) + STACK_BIAS, d); |
aoqi@1 | 2762 | if (d == L2) { |
aoqi@1 | 2763 | assert(Assembler::is_simm13(reg2offset(dst.first()) + |
aoqi@1 | 2764 | STACK_BIAS), "must be"); |
aoqi@1 | 2765 | __ st_ptr(d, SP, reg2offset(dst.first()) + STACK_BIAS); |
aoqi@1 | 2766 | } |
aoqi@1 | 2767 | } |
aoqi@1 | 2768 | } else if (out_sig_bt[c_arg] != T_VOID) { |
aoqi@1 | 2769 | // Convert the arg to NULL |
aoqi@1 | 2770 | if (dst.first()->is_reg()) { |
aoqi@1 | 2771 | __ mov(G0, dst.first()->as_Register()); |
aoqi@1 | 2772 | } else { |
aoqi@1 | 2773 | assert(Assembler::is_simm13(reg2offset(dst.first()) + |
aoqi@1 | 2774 | STACK_BIAS), "must be"); |
aoqi@1 | 2775 | __ st_ptr(G0, SP, reg2offset(dst.first()) + STACK_BIAS); |
aoqi@1 | 2776 | } |
aoqi@1 | 2777 | } |
aoqi@1 | 2778 | } |
aoqi@1 | 2779 | break; |
aoqi@1 | 2780 | case T_VOID: |
aoqi@1 | 2781 | break; |
aoqi@1 | 2782 | |
aoqi@1 | 2783 | case T_FLOAT: |
aoqi@1 | 2784 | if (src.first()->is_stack()) { |
aoqi@1 | 2785 | // Stack to stack/reg is simple |
aoqi@1 | 2786 | move32_64(masm, src, dst); |
aoqi@1 | 2787 | } else { |
aoqi@1 | 2788 | if (dst.first()->is_reg()) { |
aoqi@1 | 2789 | // freg -> reg |
aoqi@1 | 2790 | int off = |
aoqi@1 | 2791 | STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
aoqi@1 | 2792 | Register d = dst.first()->as_Register(); |
aoqi@1 | 2793 | if (Assembler::is_simm13(off)) { |
aoqi@1 | 2794 | __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
aoqi@1 | 2795 | SP, off); |
aoqi@1 | 2796 | __ ld(SP, off, d); |
aoqi@1 | 2797 | } else { |
aoqi@1 | 2798 | if (conversion_off == noreg) { |
aoqi@1 | 2799 | __ set(off, L6); |
aoqi@1 | 2800 | conversion_off = L6; |
aoqi@1 | 2801 | } |
aoqi@1 | 2802 | __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
aoqi@1 | 2803 | SP, conversion_off); |
aoqi@1 | 2804 | __ ld(SP, conversion_off , d); |
aoqi@1 | 2805 | } |
aoqi@1 | 2806 | } else { |
aoqi@1 | 2807 | // freg -> mem |
aoqi@1 | 2808 | int off = STACK_BIAS + reg2offset(dst.first()); |
aoqi@1 | 2809 | if (Assembler::is_simm13(off)) { |
aoqi@1 | 2810 | __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
aoqi@1 | 2811 | SP, off); |
aoqi@1 | 2812 | } else { |
aoqi@1 | 2813 | if (conversion_off == noreg) { |
aoqi@1 | 2814 | __ set(off, L6); |
aoqi@1 | 2815 | conversion_off = L6; |
aoqi@1 | 2816 | } |
aoqi@1 | 2817 | __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
aoqi@1 | 2818 | SP, conversion_off); |
aoqi@1 | 2819 | } |
aoqi@1 | 2820 | } |
aoqi@1 | 2821 | } |
aoqi@1 | 2822 | break; |
aoqi@1 | 2823 | |
aoqi@1 | 2824 | case T_DOUBLE: |
aoqi@1 | 2825 | assert( j_arg + 1 < total_args_passed && |
aoqi@1 | 2826 | in_sig_bt[j_arg + 1] == T_VOID && |
aoqi@1 | 2827 | out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
aoqi@1 | 2828 | if (src.first()->is_stack()) { |
aoqi@1 | 2829 | // Stack to stack/reg is simple |
aoqi@1 | 2830 | long_move(masm, src, dst); |
aoqi@1 | 2831 | } else { |
aoqi@1 | 2832 | Register d = dst.first()->is_reg() ? dst.first()->as_Register() : L2; |
aoqi@1 | 2833 | |
aoqi@1 | 2834 | // Destination could be an odd reg on 32bit in which case |
aoqi@1 | 2835 | // we can't load direct to the destination. |
aoqi@1 | 2836 | |
aoqi@1 | 2837 | if (!d->is_even() && wordSize == 4) { |
aoqi@1 | 2838 | d = L2; |
aoqi@1 | 2839 | } |
aoqi@1 | 2840 | int off = STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
aoqi@1 | 2841 | if (Assembler::is_simm13(off)) { |
aoqi@1 | 2842 | __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), |
aoqi@1 | 2843 | SP, off); |
aoqi@1 | 2844 | __ ld_long(SP, off, d); |
aoqi@1 | 2845 | } else { |
aoqi@1 | 2846 | if (conversion_off == noreg) { |
aoqi@1 | 2847 | __ set(off, L6); |
aoqi@1 | 2848 | conversion_off = L6; |
aoqi@1 | 2849 | } |
aoqi@1 | 2850 | __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), |
aoqi@1 | 2851 | SP, conversion_off); |
aoqi@1 | 2852 | __ ld_long(SP, conversion_off, d); |
aoqi@1 | 2853 | } |
aoqi@1 | 2854 | if (d == L2) { |
aoqi@1 | 2855 | long_move(masm, reg64_to_VMRegPair(L2), dst); |
aoqi@1 | 2856 | } |
aoqi@1 | 2857 | } |
aoqi@1 | 2858 | break; |
aoqi@1 | 2859 | |
aoqi@1 | 2860 | case T_LONG : |
aoqi@1 | 2861 | // 32bit can't do a split move of something like g1 -> O0, O1 |
aoqi@1 | 2862 | // so use a memory temp |
aoqi@1 | 2863 | if (src.is_single_phys_reg() && wordSize == 4) { |
aoqi@1 | 2864 | Register tmp = L2; |
aoqi@1 | 2865 | if (dst.first()->is_reg() && |
aoqi@1 | 2866 | (wordSize == 8 || dst.first()->as_Register()->is_even())) { |
aoqi@1 | 2867 | tmp = dst.first()->as_Register(); |
aoqi@1 | 2868 | } |
aoqi@1 | 2869 | |
aoqi@1 | 2870 | int off = STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
aoqi@1 | 2871 | if (Assembler::is_simm13(off)) { |
aoqi@1 | 2872 | __ stx(src.first()->as_Register(), SP, off); |
aoqi@1 | 2873 | __ ld_long(SP, off, tmp); |
aoqi@1 | 2874 | } else { |
aoqi@1 | 2875 | if (conversion_off == noreg) { |
aoqi@1 | 2876 | __ set(off, L6); |
aoqi@1 | 2877 | conversion_off = L6; |
aoqi@1 | 2878 | } |
aoqi@1 | 2879 | __ stx(src.first()->as_Register(), SP, conversion_off); |
aoqi@1 | 2880 | __ ld_long(SP, conversion_off, tmp); |
aoqi@1 | 2881 | } |
aoqi@1 | 2882 | |
aoqi@1 | 2883 | if (tmp == L2) { |
aoqi@1 | 2884 | long_move(masm, reg64_to_VMRegPair(L2), dst); |
aoqi@1 | 2885 | } |
aoqi@1 | 2886 | } else { |
aoqi@1 | 2887 | long_move(masm, src, dst); |
aoqi@1 | 2888 | } |
aoqi@1 | 2889 | break; |
aoqi@1 | 2890 | |
aoqi@1 | 2891 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
aoqi@1 | 2892 | |
aoqi@1 | 2893 | default: |
aoqi@1 | 2894 | move32_64(masm, src, dst); |
aoqi@1 | 2895 | } |
aoqi@1 | 2896 | } |
aoqi@1 | 2897 | |
aoqi@1 | 2898 | |
aoqi@1 | 2899 | // If we have any strings we must store any register based arg to the stack |
aoqi@1 | 2900 | // This includes any still live xmm registers too. |
aoqi@1 | 2901 | |
aoqi@1 | 2902 | if (total_strings > 0 ) { |
aoqi@1 | 2903 | |
aoqi@1 | 2904 | // protect all the arg registers |
aoqi@1 | 2905 | __ save_frame(0); |
aoqi@1 | 2906 | __ mov(G2_thread, L7_thread_cache); |
aoqi@1 | 2907 | const Register L2_string_off = L2; |
aoqi@1 | 2908 | |
aoqi@1 | 2909 | // Get first string offset |
aoqi@1 | 2910 | __ set(string_locs * VMRegImpl::stack_slot_size, L2_string_off); |
aoqi@1 | 2911 | |
aoqi@1 | 2912 | for (c_arg = 0 ; c_arg < total_c_args ; c_arg++ ) { |
aoqi@1 | 2913 | if (out_sig_bt[c_arg] == T_ADDRESS) { |
aoqi@1 | 2914 | |
aoqi@1 | 2915 | VMRegPair dst = out_regs[c_arg]; |
aoqi@1 | 2916 | const Register d = dst.first()->is_reg() ? |
aoqi@1 | 2917 | dst.first()->as_Register()->after_save() : noreg; |
aoqi@1 | 2918 | |
aoqi@1 | 2919 | // It's a string the oop and it was already copied to the out arg |
aoqi@1 | 2920 | // position |
aoqi@1 | 2921 | if (d != noreg) { |
aoqi@1 | 2922 | __ mov(d, O0); |
aoqi@1 | 2923 | } else { |
aoqi@1 | 2924 | assert(Assembler::is_simm13(reg2offset(dst.first()) + STACK_BIAS), |
aoqi@1 | 2925 | "must be"); |
aoqi@1 | 2926 | __ ld_ptr(FP, reg2offset(dst.first()) + STACK_BIAS, O0); |
aoqi@1 | 2927 | } |
aoqi@1 | 2928 | Label skip; |
aoqi@1 | 2929 | |
aoqi@1 | 2930 | __ br_null(O0, false, Assembler::pn, skip); |
aoqi@1 | 2931 | __ delayed()->add(FP, L2_string_off, O1); |
aoqi@1 | 2932 | |
aoqi@1 | 2933 | if (d != noreg) { |
aoqi@1 | 2934 | __ mov(O1, d); |
aoqi@1 | 2935 | } else { |
aoqi@1 | 2936 | assert(Assembler::is_simm13(reg2offset(dst.first()) + STACK_BIAS), |
aoqi@1 | 2937 | "must be"); |
aoqi@1 | 2938 | __ st_ptr(O1, FP, reg2offset(dst.first()) + STACK_BIAS); |
aoqi@1 | 2939 | } |
aoqi@1 | 2940 | |
aoqi@1 | 2941 | __ call(CAST_FROM_FN_PTR(address, SharedRuntime::get_utf), |
aoqi@1 | 2942 | relocInfo::runtime_call_type); |
aoqi@1 | 2943 | __ delayed()->add(L2_string_off, max_dtrace_string_size, L2_string_off); |
aoqi@1 | 2944 | |
aoqi@1 | 2945 | __ bind(skip); |
aoqi@1 | 2946 | |
aoqi@1 | 2947 | } |
aoqi@1 | 2948 | |
aoqi@1 | 2949 | } |
aoqi@1 | 2950 | __ mov(L7_thread_cache, G2_thread); |
aoqi@1 | 2951 | __ restore(); |
aoqi@1 | 2952 | |
aoqi@1 | 2953 | } |
aoqi@1 | 2954 | |
aoqi@1 | 2955 | |
aoqi@1 | 2956 | // Ok now we are done. Need to place the nop that dtrace wants in order to |
aoqi@1 | 2957 | // patch in the trap |
aoqi@1 | 2958 | |
aoqi@1 | 2959 | int patch_offset = ((intptr_t)__ pc()) - start; |
aoqi@1 | 2960 | |
aoqi@1 | 2961 | __ nop(); |
aoqi@1 | 2962 | |
aoqi@1 | 2963 | |
aoqi@1 | 2964 | // Return |
aoqi@1 | 2965 | |
aoqi@1 | 2966 | __ ret(); |
aoqi@1 | 2967 | __ delayed()->restore(); |
aoqi@1 | 2968 | |
aoqi@1 | 2969 | __ flush(); |
aoqi@1 | 2970 | |
aoqi@1 | 2971 | nmethod *nm = nmethod::new_dtrace_nmethod( |
aoqi@1 | 2972 | method, masm->code(), vep_offset, patch_offset, frame_complete, |
aoqi@1 | 2973 | stack_slots / VMRegImpl::slots_per_word); |
aoqi@1 | 2974 | return nm; |
aoqi@1 | 2975 | |
aoqi@1 | 2976 | } |
aoqi@1 | 2977 | |
aoqi@1 | 2978 | #endif // HAVE_DTRACE_H |
aoqi@1 | 2979 | |
aoqi@1 | 2980 | // this function returns the adjust size (in number of words) to a c2i adapter |
aoqi@1 | 2981 | // activation for use during deoptimization |
aoqi@1 | 2982 | int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) { |
aoqi@6880 | 2983 | return (callee_locals - callee_parameters) * Interpreter::stackElementWords; |
aoqi@1 | 2984 | } |
aoqi@1 | 2985 | |
aoqi@1 | 2986 | // "Top of Stack" slots that may be unused by the calling convention but must |
aoqi@1 | 2987 | // otherwise be preserved. |
aoqi@1 | 2988 | // On Intel these are not necessary and the value can be zero. |
aoqi@1 | 2989 | // On Sparc this describes the words reserved for storing a register window |
aoqi@1 | 2990 | // when an interrupt occurs. |
aoqi@1 | 2991 | uint SharedRuntime::out_preserve_stack_slots() { |
aoqi@6880 | 2992 | return 0; |
aoqi@1 | 2993 | } |
aoqi@1 | 2994 | |
aoqi@1 | 2995 | //------------------------------generate_deopt_blob---------------------------- |
aoqi@1 | 2996 | // Ought to generate an ideal graph & compile, but here's some SPARC ASM |
aoqi@1 | 2997 | // instead. |
aoqi@1 | 2998 | void SharedRuntime::generate_deopt_blob() { |
aoqi@1 | 2999 | // allocate space for the code |
aoqi@1 | 3000 | ResourceMark rm; |
aoqi@1 | 3001 | // setup code generation tools |
aoqi@1 | 3002 | //CodeBuffer buffer ("deopt_blob", 4000, 2048); |
aoqi@1 | 3003 | CodeBuffer buffer ("deopt_blob", 8000, 2048);//aoqi FIXME for debug |
aoqi@1 | 3004 | MacroAssembler* masm = new MacroAssembler( & buffer); |
aoqi@1 | 3005 | int frame_size_in_words; |
aoqi@1 | 3006 | OopMap* map = NULL; |
aoqi@1 | 3007 | // Account for the extra args we place on the stack |
aoqi@1 | 3008 | // by the time we call fetch_unroll_info |
aoqi@1 | 3009 | const int additional_words = 2; // deopt kind, thread |
aoqi@1 | 3010 | |
aoqi@1 | 3011 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@1 | 3012 | |
aoqi@1 | 3013 | address start = __ pc(); |
aoqi@1 | 3014 | Label cont; |
aoqi@1 | 3015 | // we use S3 for DeOpt reason register |
aoqi@1 | 3016 | Register reason = S3; |
aoqi@1 | 3017 | // use S6 for thread register |
aoqi@1 | 3018 | Register thread = TREG; |
aoqi@1 | 3019 | // use S7 for fetch_unroll_info returned UnrollBlock |
aoqi@1 | 3020 | Register unroll = S7; |
aoqi@1 | 3021 | // Prolog for non exception case! |
aoqi@1 | 3022 | // Correct the return address we were given. |
aoqi@6880 | 3023 | //FIXME, return address is on the tos or Ra? |
fujie@375 | 3024 | __ addi(RA, RA, - (NativeCall::return_address_offset_long)); |
aoqi@1 | 3025 | // Save everything in sight. |
aoqi@1 | 3026 | map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
aoqi@1 | 3027 | // Normal deoptimization |
aoqi@1 | 3028 | __ move(reason, Deoptimization::Unpack_deopt); |
aoqi@1 | 3029 | __ b(cont); |
aoqi@1 | 3030 | __ delayed()->nop(); |
aoqi@1 | 3031 | |
aoqi@1 | 3032 | int reexecute_offset = __ pc() - start; |
aoqi@1 | 3033 | |
aoqi@6880 | 3034 | // Reexecute case |
aoqi@6880 | 3035 | // return address is the pc describes what bci to do re-execute at |
aoqi@6880 | 3036 | |
aoqi@6880 | 3037 | // No need to update map as each call to save_live_registers will produce identical oopmap |
aoqi@1 | 3038 | (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
aoqi@6880 | 3039 | __ move(reason, Deoptimization::Unpack_reexecute); |
aoqi@1 | 3040 | __ b(cont); |
aoqi@1 | 3041 | __ delayed()->nop(); |
aoqi@1 | 3042 | |
aoqi@1 | 3043 | int exception_offset = __ pc() - start; |
aoqi@1 | 3044 | // Prolog for exception case |
aoqi@1 | 3045 | |
aoqi@9228 | 3046 | // all registers are dead at this entry point, except for V0 and |
aoqi@9228 | 3047 | // V1 which contain the exception oop and exception pc |
aoqi@1 | 3048 | // respectively. Set them in TLS and fall thru to the |
aoqi@1 | 3049 | // unpack_with_exception_in_tls entry point. |
aoqi@6880 | 3050 | |
aoqi@1 | 3051 | __ get_thread(thread); |
aoqi@6880 | 3052 | __ st_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
aoqi@1 | 3053 | __ st_ptr(V0, thread, in_bytes(JavaThread::exception_oop_offset())); |
aoqi@1 | 3054 | int exception_in_tls_offset = __ pc() - start; |
aoqi@1 | 3055 | // new implementation because exception oop is now passed in JavaThread |
aoqi@1 | 3056 | |
aoqi@1 | 3057 | // Prolog for exception case |
aoqi@1 | 3058 | // All registers must be preserved because they might be used by LinearScan |
aoqi@1 | 3059 | // Exceptiop oop and throwing PC are passed in JavaThread |
aoqi@1 | 3060 | // tos: stack at point of call to method that threw the exception (i.e. only |
aoqi@1 | 3061 | // args are on the stack, no return address) |
aoqi@1 | 3062 | |
aoqi@6880 | 3063 | // Return address will be patched later with the throwing pc. The correct value is not |
aoqi@1 | 3064 | // available now because loading it from memory would destroy registers. |
aoqi@6880 | 3065 | // Save everything in sight. |
aoqi@1 | 3066 | // No need to update map as each call to save_live_registers will produce identical oopmap |
fujie@375 | 3067 | __ addi(RA, RA, - (NativeCall::return_address_offset_long)); |
aoqi@1 | 3068 | (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
aoqi@1 | 3069 | |
aoqi@1 | 3070 | // Now it is safe to overwrite any register |
aoqi@1 | 3071 | // store the correct deoptimization type |
aoqi@1 | 3072 | __ move(reason, Deoptimization::Unpack_exception); |
aoqi@6880 | 3073 | // load throwing pc from JavaThread and patch it as the return address |
aoqi@1 | 3074 | // of the current frame. Then clear the field in JavaThread |
aoqi@1 | 3075 | __ get_thread(thread); |
aoqi@1 | 3076 | __ ld_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
aoqi@1 | 3077 | __ st_ptr(V1, SP, RegisterSaver::raOffset() * wordSize); //save ra |
aoqi@1 | 3078 | __ st_ptr(R0, thread, in_bytes(JavaThread::exception_pc_offset())); |
aoqi@1 | 3079 | |
aoqi@1 | 3080 | |
aoqi@1 | 3081 | #ifdef ASSERT |
aoqi@1 | 3082 | // verify that there is really an exception oop in JavaThread |
aoqi@1 | 3083 | __ ld_ptr(AT, thread, in_bytes(JavaThread::exception_oop_offset())); |
aoqi@1 | 3084 | __ verify_oop(AT); |
aoqi@1 | 3085 | // verify that there is no pending exception |
aoqi@1 | 3086 | Label no_pending_exception; |
aoqi@1 | 3087 | __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@6880 | 3088 | __ beq(AT, R0, no_pending_exception); |
aoqi@6880 | 3089 | __ delayed()->nop(); |
aoqi@1 | 3090 | __ stop("must not have pending exception here"); |
aoqi@1 | 3091 | __ bind(no_pending_exception); |
aoqi@1 | 3092 | #endif |
aoqi@1 | 3093 | __ bind(cont); |
aoqi@1 | 3094 | // Compiled code leaves the floating point stack dirty, empty it. |
aoqi@1 | 3095 | __ empty_FPU_stack(); |
aoqi@1 | 3096 | |
aoqi@1 | 3097 | |
aoqi@1 | 3098 | // Call C code. Need thread and this frame, but NOT official VM entry |
aoqi@6880 | 3099 | // crud. We cannot block on this call, no GC can happen. |
aoqi@1 | 3100 | #ifndef OPT_THREAD |
aoqi@1 | 3101 | __ get_thread(thread); |
aoqi@1 | 3102 | #endif |
aoqi@1 | 3103 | |
aoqi@1 | 3104 | __ move(A0, thread); |
aoqi@1 | 3105 | __ addi(SP, SP, -additional_words * wordSize); |
aoqi@1 | 3106 | |
aoqi@1 | 3107 | __ set_last_Java_frame(NOREG, NOREG, NULL); |
aoqi@1 | 3108 | |
aoqi@1 | 3109 | // Call fetch_unroll_info(). Need thread and this frame, but NOT official VM entry - cannot block on |
aoqi@1 | 3110 | // this call, no GC can happen. Call should capture return values. |
aoqi@1 | 3111 | |
aoqi@6880 | 3112 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 3113 | { |
fujie@373 | 3114 | intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + 28; |
fujie@368 | 3115 | __ patchable_set48(AT, save_pc); |
aoqi@1 | 3116 | } |
aoqi@1 | 3117 | __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
aoqi@1 | 3118 | |
aoqi@1 | 3119 | __ call((address)Deoptimization::fetch_unroll_info); |
aoqi@1 | 3120 | //__ call(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info), relocInfo::runtime_call_type); |
aoqi@1 | 3121 | __ delayed()->nop(); |
aoqi@1 | 3122 | oop_maps->add_gc_map(__ pc() - start, map); |
aoqi@1 | 3123 | __ addiu(SP, SP, additional_words * wordSize); |
aoqi@1 | 3124 | __ get_thread(thread); |
fujie@9171 | 3125 | __ reset_last_Java_frame(false); |
aoqi@1 | 3126 | |
aoqi@1 | 3127 | // Load UnrollBlock into S7 |
aoqi@1 | 3128 | __ move(unroll, V0); |
aoqi@1 | 3129 | |
aoqi@1 | 3130 | |
aoqi@1 | 3131 | // Move the unpack kind to a safe place in the UnrollBlock because |
aoqi@1 | 3132 | // we are very short of registers |
aoqi@1 | 3133 | |
aoqi@1 | 3134 | Address unpack_kind(unroll, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes()); |
aoqi@1 | 3135 | __ sw(reason, unpack_kind); |
aoqi@1 | 3136 | // save the unpack_kind value |
aoqi@1 | 3137 | // Retrieve the possible live values (return values) |
aoqi@1 | 3138 | // All callee save registers representing jvm state |
aoqi@1 | 3139 | // are now in the vframeArray. |
aoqi@1 | 3140 | |
aoqi@1 | 3141 | Label noException; |
aoqi@1 | 3142 | __ move(AT, Deoptimization::Unpack_exception); |
aoqi@1 | 3143 | __ bne(AT, reason, noException);// Was exception pending? |
aoqi@1 | 3144 | __ delayed()->nop(); |
aoqi@1 | 3145 | __ ld_ptr(V0, thread, in_bytes(JavaThread::exception_oop_offset())); |
aoqi@1 | 3146 | __ ld_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
aoqi@1 | 3147 | __ st_ptr(R0, thread, in_bytes(JavaThread::exception_pc_offset())); |
aoqi@1 | 3148 | __ st_ptr(R0, thread, in_bytes(JavaThread::exception_oop_offset())); |
aoqi@6880 | 3149 | |
aoqi@1 | 3150 | __ verify_oop(V0); |
aoqi@1 | 3151 | |
aoqi@1 | 3152 | // Overwrite the result registers with the exception results. |
aoqi@6880 | 3153 | __ st_ptr(V0, SP, RegisterSaver::v0Offset()*wordSize); |
aoqi@1 | 3154 | __ st_ptr(V1, SP, RegisterSaver::v1Offset()*wordSize); |
aoqi@6880 | 3155 | |
aoqi@1 | 3156 | __ bind(noException); |
aoqi@1 | 3157 | |
aoqi@1 | 3158 | |
aoqi@1 | 3159 | // Stack is back to only having register save data on the stack. |
aoqi@1 | 3160 | // Now restore the result registers. Everything else is either dead or captured |
aoqi@1 | 3161 | // in the vframeArray. |
aoqi@1 | 3162 | |
aoqi@1 | 3163 | RegisterSaver::restore_result_registers(masm); |
aoqi@1 | 3164 | // All of the register save area has been popped of the stack. Only the |
aoqi@1 | 3165 | // return address remains. |
aoqi@6880 | 3166 | // Pop all the frames we must move/replace. |
aoqi@1 | 3167 | // Frame picture (youngest to oldest) |
aoqi@1 | 3168 | // 1: self-frame (no frame link) |
aoqi@1 | 3169 | // 2: deopting frame (no frame link) |
aoqi@6880 | 3170 | // 3: caller of deopting frame (could be compiled/interpreted). |
aoqi@1 | 3171 | // |
aoqi@1 | 3172 | // Note: by leaving the return address of self-frame on the stack |
aoqi@1 | 3173 | // and using the size of frame 2 to adjust the stack |
aoqi@1 | 3174 | // when we are done the return to frame 3 will still be on the stack. |
aoqi@1 | 3175 | |
aoqi@1 | 3176 | // register for the sender's sp |
aoqi@1 | 3177 | Register sender_sp = Rsender; |
aoqi@1 | 3178 | // register for frame pcs |
aoqi@1 | 3179 | Register pcs = T0; |
aoqi@1 | 3180 | // register for frame sizes |
aoqi@1 | 3181 | Register sizes = T1; |
aoqi@1 | 3182 | // register for frame count |
aoqi@1 | 3183 | Register count = T3; |
aoqi@6880 | 3184 | |
aoqi@1 | 3185 | // Pop deoptimized frame |
aoqi@1 | 3186 | __ lw(AT, unroll, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()); |
aoqi@1 | 3187 | __ add(SP, SP, AT); |
aoqi@1 | 3188 | // sp should be pointing at the return address to the caller (3) |
aoqi@6880 | 3189 | |
aoqi@1 | 3190 | // Load array of frame pcs into pcs |
aoqi@1 | 3191 | __ ld_ptr(pcs, unroll, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()); |
aoqi@1 | 3192 | __ addi(SP, SP, wordSize); // trash the old pc |
aoqi@1 | 3193 | // Load array of frame sizes into T6 |
aoqi@1 | 3194 | __ ld_ptr(sizes, unroll, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()); |
aoqi@1 | 3195 | |
aoqi@6880 | 3196 | |
aoqi@1 | 3197 | |
aoqi@1 | 3198 | // Load count of frams into T3 |
aoqi@1 | 3199 | __ lw(count, unroll, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()); |
aoqi@1 | 3200 | // Pick up the initial fp we should save |
aoqi@1 | 3201 | __ ld(FP, unroll, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()); |
aoqi@1 | 3202 | // Now adjust the caller's stack to make up for the extra locals |
aoqi@1 | 3203 | // but record the original sp so that we can save it in the skeletal interpreter |
aoqi@1 | 3204 | // frame and the stack walking of interpreter_sender will get the unextended sp |
aoqi@1 | 3205 | // value and not the "real" sp value. |
aoqi@1 | 3206 | __ move(sender_sp, SP); |
aoqi@1 | 3207 | __ lw(AT, unroll, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()); |
aoqi@1 | 3208 | __ sub(SP, SP, AT); |
aoqi@1 | 3209 | |
aoqi@1 | 3210 | // Push interpreter frames in a loop |
aoqi@9459 | 3211 | // |
aoqi@9459 | 3212 | //Loop: |
aoqi@9459 | 3213 | // 0x000000555bd82d18: lw t2, 0x0(t1) ; lw sizes[i] <--- error lw->ld |
aoqi@9459 | 3214 | // 0x000000555bd82d1c: ld at, 0x0(t0) ; ld pcs[i] |
aoqi@9459 | 3215 | // 0x000000555bd82d20: daddi t2, t2, 0xfffffff0 ; t2 -= 16 |
aoqi@9459 | 3216 | // 0x000000555bd82d24: daddi sp, sp, 0xfffffff0 |
aoqi@9459 | 3217 | // 0x000000555bd82d28: sd fp, 0x0(sp) ; push fp |
aoqi@9459 | 3218 | // 0x000000555bd82d2c: sd at, 0x8(sp) ; push at |
aoqi@9459 | 3219 | // 0x000000555bd82d30: dadd fp, sp, zero ; fp <- sp |
aoqi@9459 | 3220 | // 0x000000555bd82d34: dsub sp, sp, t2 ; sp -= t2 |
aoqi@9459 | 3221 | // 0x000000555bd82d38: sd zero, 0xfffffff0(fp) ; __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
aoqi@9459 | 3222 | // 0x000000555bd82d3c: sd s4, 0xfffffff8(fp) ; __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize); |
aoqi@9459 | 3223 | // 0x000000555bd82d40: dadd s4, sp, zero ; move(sender_sp, SP); |
aoqi@9459 | 3224 | // 0x000000555bd82d44: daddi t3, t3, 0xffffffff ; count -- |
aoqi@9459 | 3225 | // 0x000000555bd82d48: daddi t1, t1, 0x4 ; sizes += 4 |
aoqi@9459 | 3226 | // 0x000000555bd82d4c: bne t3, zero, 0x000000555bd82d18 |
aoqi@9459 | 3227 | // 0x000000555bd82d50: daddi t0, t0, 0x4 ; <--- error t0 += 8 |
aoqi@9459 | 3228 | // |
aoqi@9459 | 3229 | // pcs[0] = frame_pcs[0] = deopt_sender.raw_pc(); regex.split |
aoqi@1 | 3230 | Label loop; |
aoqi@1 | 3231 | __ bind(loop); |
aoqi@6880 | 3232 | __ ld(T2, sizes, 0); // Load frame size |
aoqi@6880 | 3233 | __ ld_ptr(AT, pcs, 0); // save return address |
aoqi@9459 | 3234 | __ addi(T2, T2, -2*wordSize); // we'll push pc and fp, by hand |
aoqi@6880 | 3235 | __ push2(AT, FP); |
aoqi@1 | 3236 | __ move(FP, SP); |
aoqi@6880 | 3237 | __ sub(SP, SP, T2); // Prolog! |
aoqi@1 | 3238 | // This value is corrected by layout_activation_impl |
aoqi@6880 | 3239 | __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
aoqi@1 | 3240 | __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize);// Make it walkable |
aoqi@6880 | 3241 | __ move(sender_sp, SP); // pass to next frame |
aoqi@6880 | 3242 | __ addi(count, count, -1); // decrement counter |
aoqi@6880 | 3243 | __ addi(sizes, sizes, wordSize); // Bump array pointer (sizes) |
aoqi@1 | 3244 | __ bne(count, R0, loop); |
aoqi@6880 | 3245 | __ delayed()->addi(pcs, pcs, wordSize); // Bump array pointer (pcs) |
aoqi@6880 | 3246 | __ ld(AT, pcs, 0); // frame_pcs[number_of_frames] = Interpreter::deopt_entry(vtos, 0); |
aoqi@1 | 3247 | // Re-push self-frame |
aoqi@6880 | 3248 | __ push2(AT, FP); |
aoqi@1 | 3249 | __ move(FP, SP); |
aoqi@6880 | 3250 | __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
aoqi@6880 | 3251 | __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize); |
aoqi@1 | 3252 | __ addi(SP, SP, -(frame_size_in_words - 2 - additional_words) * wordSize); |
aoqi@1 | 3253 | |
aoqi@1 | 3254 | // Restore frame locals after moving the frame |
aoqi@1 | 3255 | __ sd(V0, SP, RegisterSaver::v0Offset() * wordSize); |
aoqi@1 | 3256 | __ sd(V1, SP, RegisterSaver::v1Offset() * wordSize); |
aoqi@1 | 3257 | __ sdc1(F0, SP, RegisterSaver::fpResultOffset()* wordSize);// Pop float stack and store in local |
aoqi@1 | 3258 | __ sdc1(F1, SP, (RegisterSaver::fpResultOffset() + 1) * wordSize); |
aoqi@1 | 3259 | |
aoqi@6880 | 3260 | |
aoqi@1 | 3261 | // Call unpack_frames(). Need thread and this frame, but NOT official VM entry - cannot block on |
aoqi@1 | 3262 | // this call, no GC can happen. |
aoqi@6880 | 3263 | __ move(A1, reason); // exec_mode |
aoqi@1 | 3264 | __ get_thread(thread); |
aoqi@6880 | 3265 | __ move(A0, thread); // thread |
aoqi@1 | 3266 | __ addi(SP, SP, (-additional_words) *wordSize); |
aoqi@1 | 3267 | |
aoqi@1 | 3268 | // set last_Java_sp, last_Java_fp |
aoqi@1 | 3269 | __ set_last_Java_frame(NOREG, FP, NULL); |
aoqi@1 | 3270 | |
aoqi@1 | 3271 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@1 | 3272 | __ andr(SP, SP, AT); // Fix stack alignment as required by ABI |
aoqi@1 | 3273 | |
aoqi@6880 | 3274 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 3275 | { |
fujie@373 | 3276 | intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + 28; |
fujie@368 | 3277 | __ patchable_set48(AT, save_pc); |
aoqi@1 | 3278 | } |
aoqi@1 | 3279 | __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
aoqi@6880 | 3280 | |
aoqi@1 | 3281 | __ call(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames), relocInfo::runtime_call_type); |
aoqi@1 | 3282 | __ delayed()->nop(); |
aoqi@1 | 3283 | // Revert SP alignment after call since we're going to do some SP relative addressing below |
aoqi@1 | 3284 | __ ld(SP, thread, in_bytes(JavaThread::last_Java_sp_offset())); |
aoqi@1 | 3285 | // Set an oopmap for the call site |
aoqi@1 | 3286 | oop_maps->add_gc_map(__ offset(), new OopMap( frame_size_in_words , 0)); |
aoqi@1 | 3287 | |
aoqi@1 | 3288 | __ push(V0); |
aoqi@6880 | 3289 | |
aoqi@1 | 3290 | __ get_thread(thread); |
fujie@9171 | 3291 | __ reset_last_Java_frame(true); |
aoqi@1 | 3292 | |
aoqi@1 | 3293 | // Collect return values |
aoqi@1 | 3294 | __ ld(V0, SP, (RegisterSaver::v0Offset() + additional_words +1) * wordSize); |
aoqi@1 | 3295 | __ ld(V1, SP, (RegisterSaver::v1Offset() + additional_words +1) * wordSize); |
aoqi@1 | 3296 | __ ldc1(F0, SP, RegisterSaver::fpResultOffset()* wordSize);// Pop float stack and store in local |
aoqi@1 | 3297 | __ ldc1(F1, SP, (RegisterSaver::fpResultOffset() + 1) * wordSize); |
aoqi@6880 | 3298 | //FIXME, |
aoqi@1 | 3299 | // Clear floating point stack before returning to interpreter |
aoqi@1 | 3300 | __ empty_FPU_stack(); |
aoqi@1 | 3301 | //FIXME, we should consider about float and double |
aoqi@1 | 3302 | // Push a float or double return value if necessary. |
aoqi@1 | 3303 | __ leave(); |
aoqi@1 | 3304 | |
aoqi@1 | 3305 | // Jump to interpreter |
aoqi@1 | 3306 | __ jr(RA); |
aoqi@1 | 3307 | __ delayed()->nop(); |
aoqi@1 | 3308 | |
aoqi@1 | 3309 | masm->flush(); |
aoqi@1 | 3310 | _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); |
aoqi@1 | 3311 | _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); |
aoqi@1 | 3312 | } |
aoqi@1 | 3313 | |
aoqi@1 | 3314 | #ifdef COMPILER2 |
aoqi@1 | 3315 | |
aoqi@1 | 3316 | //------------------------------generate_uncommon_trap_blob-------------------- |
aoqi@1 | 3317 | // Ought to generate an ideal graph & compile, but here's some SPARC ASM |
aoqi@1 | 3318 | // instead. |
aoqi@1 | 3319 | void SharedRuntime::generate_uncommon_trap_blob() { |
aoqi@1 | 3320 | // allocate space for the code |
aoqi@1 | 3321 | ResourceMark rm; |
aoqi@1 | 3322 | // setup code generation tools |
aoqi@6880 | 3323 | CodeBuffer buffer ("uncommon_trap_blob", 512*80 , 512*40 ); |
aoqi@6880 | 3324 | MacroAssembler* masm = new MacroAssembler(&buffer); |
aoqi@1 | 3325 | |
aoqi@1 | 3326 | enum frame_layout { |
aoqi@6880 | 3327 | s0_off, s0_off2, |
aoqi@6880 | 3328 | s1_off, s1_off2, |
aoqi@6880 | 3329 | s2_off, s2_off2, |
aoqi@6880 | 3330 | s3_off, s3_off2, |
aoqi@6880 | 3331 | s4_off, s4_off2, |
aoqi@6880 | 3332 | s5_off, s5_off2, |
aoqi@6880 | 3333 | s6_off, s6_off2, |
aoqi@6880 | 3334 | s7_off, s7_off2, |
aoqi@6880 | 3335 | fp_off, fp_off2, |
aoqi@6880 | 3336 | return_off, return_off2, // slot for return address sp + 9 |
aoqi@1 | 3337 | framesize |
aoqi@1 | 3338 | }; |
aoqi@1 | 3339 | assert(framesize % 4 == 0, "sp not 16-byte aligned"); |
aoqi@1 | 3340 | |
aoqi@1 | 3341 | address start = __ pc(); |
aoqi@1 | 3342 | |
aoqi@1 | 3343 | // Push self-frame. |
aoqi@1 | 3344 | __ daddiu(SP, SP, -framesize * BytesPerInt); |
aoqi@1 | 3345 | |
aoqi@1 | 3346 | __ sd(RA, SP, return_off * BytesPerInt); |
aoqi@1 | 3347 | __ sd(FP, SP, fp_off * BytesPerInt); |
aoqi@1 | 3348 | |
aoqi@6880 | 3349 | // Save callee saved registers. None for UseSSE=0, |
aoqi@1 | 3350 | // floats-only for UseSSE=1, and doubles for UseSSE=2. |
aoqi@1 | 3351 | __ sd(S0, SP, s0_off * BytesPerInt); |
aoqi@1 | 3352 | __ sd(S1, SP, s1_off * BytesPerInt); |
aoqi@1 | 3353 | __ sd(S2, SP, s2_off * BytesPerInt); |
aoqi@1 | 3354 | __ sd(S3, SP, s3_off * BytesPerInt); |
aoqi@1 | 3355 | __ sd(S4, SP, s4_off * BytesPerInt); |
aoqi@1 | 3356 | __ sd(S5, SP, s5_off * BytesPerInt); |
aoqi@1 | 3357 | __ sd(S6, SP, s6_off * BytesPerInt); |
aoqi@1 | 3358 | __ sd(S7, SP, s7_off * BytesPerInt); |
aoqi@1 | 3359 | |
aoqi@1 | 3360 | __ daddi(FP, SP, fp_off * BytesPerInt); |
aoqi@1 | 3361 | |
aoqi@1 | 3362 | // Clear the floating point exception stack |
aoqi@1 | 3363 | __ empty_FPU_stack(); |
aoqi@1 | 3364 | |
aoqi@1 | 3365 | Register thread = TREG; |
aoqi@1 | 3366 | |
aoqi@1 | 3367 | #ifndef OPT_THREAD |
aoqi@1 | 3368 | __ get_thread(thread); |
aoqi@1 | 3369 | #endif |
aoqi@1 | 3370 | // set last_Java_sp |
aoqi@1 | 3371 | __ set_last_Java_frame(NOREG, FP, NULL); |
aoqi@6880 | 3372 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 3373 | { |
fujie@373 | 3374 | long save_pc = (long)__ pc() + 52; |
fujie@368 | 3375 | __ patchable_set48(AT, (long)save_pc); |
aoqi@1 | 3376 | __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
aoqi@1 | 3377 | } |
aoqi@1 | 3378 | // Call C code. Need thread but NOT official VM entry |
aoqi@1 | 3379 | // crud. We cannot block on this call, no GC can happen. Call should |
aoqi@1 | 3380 | // capture callee-saved registers as well as return values. |
aoqi@1 | 3381 | __ move(A0, thread); |
aoqi@1 | 3382 | // argument already in T0 |
aoqi@1 | 3383 | __ move(A1, T0); |
fujie@386 | 3384 | __ patchable_call((address)Deoptimization::uncommon_trap); |
aoqi@1 | 3385 | |
aoqi@1 | 3386 | // Set an oopmap for the call site |
aoqi@1 | 3387 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@1 | 3388 | OopMap* map = new OopMap( framesize, 0 ); |
aoqi@1 | 3389 | |
aoqi@6880 | 3390 | map->set_callee_saved( VMRegImpl::stack2reg(s0_off ), S0->as_VMReg() ); |
aoqi@1 | 3391 | map->set_callee_saved( VMRegImpl::stack2reg(s1_off ), S1->as_VMReg() ); |
aoqi@1 | 3392 | map->set_callee_saved( VMRegImpl::stack2reg(s2_off ), S2->as_VMReg() ); |
aoqi@1 | 3393 | map->set_callee_saved( VMRegImpl::stack2reg(s3_off ), S3->as_VMReg() ); |
aoqi@1 | 3394 | map->set_callee_saved( VMRegImpl::stack2reg(s4_off ), S4->as_VMReg() ); |
aoqi@1 | 3395 | map->set_callee_saved( VMRegImpl::stack2reg(s5_off ), S5->as_VMReg() ); |
aoqi@1 | 3396 | map->set_callee_saved( VMRegImpl::stack2reg(s6_off ), S6->as_VMReg() ); |
aoqi@1 | 3397 | map->set_callee_saved( VMRegImpl::stack2reg(s7_off ), S7->as_VMReg() ); |
aoqi@1 | 3398 | |
aoqi@1 | 3399 | //oop_maps->add_gc_map( __ offset(), true, map); |
aoqi@6880 | 3400 | oop_maps->add_gc_map( __ offset(), map); |
aoqi@1 | 3401 | |
aoqi@1 | 3402 | #ifndef OPT_THREAD |
aoqi@1 | 3403 | __ get_thread(thread); |
aoqi@1 | 3404 | #endif |
fujie@9171 | 3405 | __ reset_last_Java_frame(false); |
aoqi@1 | 3406 | |
aoqi@1 | 3407 | // Load UnrollBlock into S7 |
aoqi@1 | 3408 | Register unroll = S7; |
aoqi@1 | 3409 | __ move(unroll, V0); |
aoqi@1 | 3410 | |
aoqi@6880 | 3411 | // Pop all the frames we must move/replace. |
aoqi@6880 | 3412 | // |
aoqi@1 | 3413 | // Frame picture (youngest to oldest) |
aoqi@1 | 3414 | // 1: self-frame (no frame link) |
aoqi@1 | 3415 | // 2: deopting frame (no frame link) |
aoqi@6880 | 3416 | // 3: possible-i2c-adapter-frame |
aoqi@1 | 3417 | // 4: caller of deopting frame (could be compiled/interpreted. If interpreted we will create an |
aoqi@1 | 3418 | // and c2i here) |
aoqi@1 | 3419 | |
aoqi@1 | 3420 | __ daddiu(SP, SP, framesize * BytesPerInt); |
aoqi@1 | 3421 | |
aoqi@1 | 3422 | // Pop deoptimized frame |
aoqi@1 | 3423 | __ lw(AT, unroll, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()); |
aoqi@1 | 3424 | __ dadd(SP, SP, AT); |
aoqi@1 | 3425 | |
aoqi@1 | 3426 | // register for frame pcs |
aoqi@1 | 3427 | Register pcs = T8; |
aoqi@1 | 3428 | // register for frame sizes |
aoqi@1 | 3429 | Register sizes = T9; |
aoqi@1 | 3430 | // register for frame count |
aoqi@1 | 3431 | Register count = T3; |
aoqi@1 | 3432 | // register for the sender's sp |
aoqi@1 | 3433 | Register sender_sp = T1; |
aoqi@1 | 3434 | |
aoqi@1 | 3435 | // sp should be pointing at the return address to the caller (4) |
aoqi@9459 | 3436 | // Load array of frame pcs |
aoqi@1 | 3437 | __ ld(pcs, unroll, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()); |
aoqi@1 | 3438 | |
aoqi@9459 | 3439 | // Load array of frame sizes |
aoqi@1 | 3440 | __ ld(sizes, unroll, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()); |
aoqi@1 | 3441 | __ lwu(count, unroll, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()); |
aoqi@1 | 3442 | |
aoqi@1 | 3443 | // Pick up the initial fp we should save |
aoqi@1 | 3444 | __ ld(FP, unroll, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()); |
aoqi@1 | 3445 | // Now adjust the caller's stack to make up for the extra locals |
aoqi@1 | 3446 | // but record the original sp so that we can save it in the skeletal interpreter |
aoqi@1 | 3447 | // frame and the stack walking of interpreter_sender will get the unextended sp |
aoqi@1 | 3448 | // value and not the "real" sp value. |
aoqi@1 | 3449 | |
aoqi@1 | 3450 | __ move(sender_sp, SP); |
aoqi@1 | 3451 | __ lw(AT, unroll, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()); |
aoqi@1 | 3452 | __ dsub(SP, SP, AT); |
aoqi@1 | 3453 | // Push interpreter frames in a loop |
aoqi@1 | 3454 | Label loop; |
aoqi@1 | 3455 | __ bind(loop); |
aoqi@1 | 3456 | __ ld(T2, sizes, 0); // Load frame size |
aoqi@1 | 3457 | __ ld(AT, pcs, 0); // save return address |
aoqi@9459 | 3458 | __ daddi(T2, T2, -2*wordSize); // we'll push pc and fp, by hand |
aoqi@1 | 3459 | __ push2(AT, FP); |
aoqi@1 | 3460 | __ move(FP, SP); |
aoqi@1 | 3461 | __ dsub(SP, SP, T2); // Prolog! |
aoqi@1 | 3462 | // This value is corrected by layout_activation_impl |
aoqi@1 | 3463 | __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
aoqi@1 | 3464 | __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize);// Make it walkable |
aoqi@1 | 3465 | __ move(sender_sp, SP); // pass to next frame |
aoqi@1 | 3466 | __ daddi(count, count, -1); // decrement counter |
aoqi@1 | 3467 | __ daddi(sizes, sizes, wordSize); // Bump array pointer (sizes) |
aoqi@1 | 3468 | __ addi(pcs, pcs, wordSize); // Bump array pointer (pcs) |
aoqi@1 | 3469 | __ bne(count, R0, loop); |
aoqi@1 | 3470 | __ delayed()->nop(); // Bump array pointer (pcs) |
aoqi@1 | 3471 | |
aoqi@1 | 3472 | __ ld(RA, pcs, 0); |
aoqi@1 | 3473 | |
aoqi@1 | 3474 | // Re-push self-frame |
aoqi@1 | 3475 | __ daddi(SP, SP, - 2 * wordSize); // save old & set new FP |
aoqi@1 | 3476 | __ sd(FP, SP, 0 * wordSize); // save final return address |
aoqi@1 | 3477 | __ sd(RA, SP, 1 * wordSize); |
aoqi@6880 | 3478 | __ move(FP, SP); |
aoqi@1 | 3479 | __ daddi(SP, SP, -(framesize / 2 - 2) * wordSize); |
aoqi@1 | 3480 | |
aoqi@1 | 3481 | // set last_Java_sp, last_Java_fp |
aoqi@1 | 3482 | __ set_last_Java_frame(NOREG, FP, NULL); |
aoqi@1 | 3483 | |
aoqi@1 | 3484 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@1 | 3485 | __ andr(SP, SP, AT); // Fix stack alignment as required by ABI |
aoqi@1 | 3486 | |
aoqi@6880 | 3487 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 3488 | { |
fujie@373 | 3489 | long save_pc = (long)__ pc() + 52; |
fujie@368 | 3490 | __ patchable_set48(AT, (long)save_pc); |
aoqi@1 | 3491 | } |
huangjia@9645 | 3492 | __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
aoqi@1 | 3493 | |
aoqi@1 | 3494 | // Call C code. Need thread but NOT official VM entry |
aoqi@1 | 3495 | // crud. We cannot block on this call, no GC can happen. Call should |
aoqi@1 | 3496 | // restore return values to their stack-slots with the new SP. |
aoqi@1 | 3497 | __ move(A0, thread); |
aoqi@1 | 3498 | __ move(A1, Deoptimization::Unpack_uncommon_trap); |
fujie@386 | 3499 | __ patchable_call((address)Deoptimization::unpack_frames); |
aoqi@1 | 3500 | // Set an oopmap for the call site |
huangjia@9645 | 3501 | oop_maps->add_gc_map( __ offset(), new OopMap( framesize, 0 ) ); |
aoqi@1 | 3502 | |
fujie@9171 | 3503 | __ reset_last_Java_frame(true); |
aoqi@1 | 3504 | |
aoqi@1 | 3505 | // Pop self-frame. |
aoqi@1 | 3506 | __ leave(); // Epilog! |
aoqi@1 | 3507 | |
aoqi@1 | 3508 | // Jump to interpreter |
aoqi@1 | 3509 | __ jr(RA); |
aoqi@1 | 3510 | __ delayed()->nop(); |
aoqi@1 | 3511 | // ------------- |
aoqi@1 | 3512 | // make sure all code is generated |
aoqi@1 | 3513 | masm->flush(); |
aoqi@1 | 3514 | |
aoqi@1 | 3515 | _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize / 2); |
aoqi@1 | 3516 | } |
aoqi@1 | 3517 | |
aoqi@1 | 3518 | #endif // COMPILER2 |
aoqi@1 | 3519 | |
aoqi@1 | 3520 | //------------------------------generate_handler_blob------------------- |
aoqi@1 | 3521 | // |
aoqi@1 | 3522 | // Generate a special Compile2Runtime blob that saves all registers, and sets |
aoqi@1 | 3523 | // up an OopMap and calls safepoint code to stop the compiled code for |
aoqi@1 | 3524 | // a safepoint. |
aoqi@1 | 3525 | // |
aoqi@1 | 3526 | // This blob is jumped to (via a breakpoint and the signal handler) from a |
aoqi@6880 | 3527 | // safepoint in compiled code. |
aoqi@6880 | 3528 | |
aoqi@1 | 3529 | SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int pool_type) { |
aoqi@1 | 3530 | |
aoqi@1 | 3531 | // Account for thread arg in our frame |
aoqi@6880 | 3532 | const int additional_words = 0; |
aoqi@1 | 3533 | int frame_size_in_words; |
aoqi@1 | 3534 | |
aoqi@6880 | 3535 | assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
aoqi@1 | 3536 | |
aoqi@1 | 3537 | ResourceMark rm; |
aoqi@1 | 3538 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@1 | 3539 | OopMap* map; |
aoqi@1 | 3540 | |
aoqi@1 | 3541 | // allocate space for the code |
aoqi@6880 | 3542 | // setup code generation tools |
aoqi@1 | 3543 | CodeBuffer buffer ("handler_blob", 2048, 512); |
aoqi@1 | 3544 | MacroAssembler* masm = new MacroAssembler( &buffer); |
aoqi@6880 | 3545 | |
aoqi@6880 | 3546 | const Register thread = TREG; |
aoqi@6880 | 3547 | address start = __ pc(); |
aoqi@6880 | 3548 | address call_pc = NULL; |
aoqi@1 | 3549 | bool cause_return = (pool_type == POLL_AT_RETURN); |
aoqi@1 | 3550 | bool save_vectors = (pool_type == POLL_AT_VECTOR_LOOP); |
aoqi@1 | 3551 | |
aoqi@1 | 3552 | // If cause_return is true we are at a poll_return and there is |
aoqi@1 | 3553 | // the return address in RA to the caller on the nmethod |
aoqi@1 | 3554 | // that is safepoint. We can leave this return in RA and |
aoqi@1 | 3555 | // effectively complete the return and safepoint in the caller. |
aoqi@1 | 3556 | // Otherwise we load exception pc to RA. |
aoqi@1 | 3557 | __ push(thread); |
aoqi@1 | 3558 | #ifndef OPT_THREAD |
aoqi@1 | 3559 | __ get_thread(thread); |
aoqi@1 | 3560 | #endif |
aoqi@1 | 3561 | |
aoqi@1 | 3562 | if(!cause_return) { |
aoqi@1 | 3563 | __ ld_ptr(RA, Address(thread, JavaThread::saved_exception_pc_offset())); |
aoqi@1 | 3564 | } |
aoqi@6880 | 3565 | |
aoqi@1 | 3566 | __ pop(thread); |
aoqi@1 | 3567 | map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, save_vectors); |
aoqi@1 | 3568 | |
aoqi@1 | 3569 | #ifndef OPT_THREAD |
aoqi@1 | 3570 | __ get_thread(thread); |
aoqi@1 | 3571 | #endif |
aoqi@1 | 3572 | // The following is basically a call_VM. However, we need the precise |
aoqi@1 | 3573 | // address of the call in order to generate an oopmap. Hence, we do all the |
aoqi@1 | 3574 | // work outselvs. |
aoqi@1 | 3575 | |
aoqi@1 | 3576 | __ move(A0, thread); |
aoqi@1 | 3577 | __ set_last_Java_frame(NOREG, NOREG, NULL); |
aoqi@1 | 3578 | |
aoqi@1 | 3579 | |
aoqi@1 | 3580 | // do the call |
aoqi@1 | 3581 | __ call(call_ptr); |
aoqi@1 | 3582 | __ delayed()->nop(); |
aoqi@1 | 3583 | |
aoqi@1 | 3584 | // Set an oopmap for the call site. This oopmap will map all |
aoqi@1 | 3585 | // oop-registers and debug-info registers as callee-saved. This |
aoqi@1 | 3586 | // will allow deoptimization at this safepoint to find all possible |
aoqi@1 | 3587 | // debug-info recordings, as well as let GC find all oops. |
aoqi@1 | 3588 | oop_maps->add_gc_map(__ offset(), map); |
aoqi@1 | 3589 | |
aoqi@1 | 3590 | Label noException; |
aoqi@1 | 3591 | |
aoqi@1 | 3592 | // Clear last_Java_sp again |
fujie@9171 | 3593 | __ reset_last_Java_frame(false); |
aoqi@1 | 3594 | |
aoqi@1 | 3595 | __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@1 | 3596 | __ beq(AT, R0, noException); |
aoqi@1 | 3597 | __ delayed()->nop(); |
aoqi@1 | 3598 | |
aoqi@1 | 3599 | // Exception pending |
aoqi@1 | 3600 | |
aoqi@1 | 3601 | RegisterSaver::restore_live_registers(masm, save_vectors); |
aoqi@1 | 3602 | //forward_exception_entry need return address on the stack |
aoqi@1 | 3603 | __ push(RA); |
fujie@386 | 3604 | __ patchable_jump((address)StubRoutines::forward_exception_entry()); |
aoqi@1 | 3605 | |
aoqi@1 | 3606 | // No exception case |
aoqi@1 | 3607 | __ bind(noException); |
aoqi@6880 | 3608 | // Normal exit, register restoring and exit |
aoqi@1 | 3609 | RegisterSaver::restore_live_registers(masm, save_vectors); |
aoqi@1 | 3610 | __ jr(RA); |
aoqi@1 | 3611 | __ delayed()->nop(); |
aoqi@6880 | 3612 | |
aoqi@6880 | 3613 | masm->flush(); |
aoqi@1 | 3614 | |
aoqi@1 | 3615 | // Fill-out other meta info |
aoqi@6880 | 3616 | return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); |
aoqi@1 | 3617 | } |
aoqi@1 | 3618 | |
aoqi@1 | 3619 | // |
aoqi@1 | 3620 | // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss |
aoqi@1 | 3621 | // |
aoqi@1 | 3622 | // Generate a stub that calls into vm to find out the proper destination |
aoqi@1 | 3623 | // of a java call. All the argument registers are live at this point |
aoqi@1 | 3624 | // but since this is generic code we don't know what they are and the caller |
aoqi@1 | 3625 | // must do any gc of the args. |
aoqi@1 | 3626 | // |
aoqi@1 | 3627 | RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) { |
aoqi@1 | 3628 | assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
aoqi@1 | 3629 | |
aoqi@1 | 3630 | // allocate space for the code |
aoqi@1 | 3631 | ResourceMark rm; |
aoqi@1 | 3632 | |
aoqi@1 | 3633 | //CodeBuffer buffer(name, 1000, 512); |
aoqi@1 | 3634 | //FIXME. aoqi. code_size |
aoqi@8009 | 3635 | CodeBuffer buffer(name, 2000, 2048); |
aoqi@1 | 3636 | MacroAssembler* masm = new MacroAssembler(&buffer); |
aoqi@1 | 3637 | |
aoqi@1 | 3638 | int frame_size_words; |
aoqi@6880 | 3639 | //we put the thread in A0 |
aoqi@1 | 3640 | |
aoqi@1 | 3641 | OopMapSet *oop_maps = new OopMapSet(); |
aoqi@1 | 3642 | OopMap* map = NULL; |
aoqi@1 | 3643 | |
aoqi@1 | 3644 | int start = __ offset(); |
aoqi@1 | 3645 | map = RegisterSaver::save_live_registers(masm, 0, &frame_size_words); |
aoqi@1 | 3646 | |
aoqi@1 | 3647 | |
aoqi@1 | 3648 | int frame_complete = __ offset(); |
aoqi@1 | 3649 | |
aoqi@1 | 3650 | const Register thread = T8; |
aoqi@1 | 3651 | __ get_thread(thread); |
aoqi@1 | 3652 | |
aoqi@6880 | 3653 | __ move(A0, thread); |
aoqi@1 | 3654 | __ set_last_Java_frame(noreg, FP, NULL); |
aoqi@6880 | 3655 | //align the stack before invoke native |
aoqi@1 | 3656 | __ move(AT, -(StackAlignmentInBytes)); |
aoqi@6880 | 3657 | __ andr(SP, SP, AT); |
aoqi@6880 | 3658 | __ relocate(relocInfo::internal_pc_type); |
aoqi@6880 | 3659 | { |
fujie@373 | 3660 | intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + 24 + 1 * BytesPerInstWord; |
fujie@368 | 3661 | __ patchable_set48(AT, save_pc); |
aoqi@1 | 3662 | } |
aoqi@1 | 3663 | __ sd(AT, thread, in_bytes(JavaThread::last_Java_pc_offset())); |
aoqi@1 | 3664 | |
aoqi@1 | 3665 | __ call(destination); |
aoqi@1 | 3666 | __ delayed()->nop(); |
aoqi@1 | 3667 | |
aoqi@1 | 3668 | // Set an oopmap for the call site. |
aoqi@1 | 3669 | // We need this not only for callee-saved registers, but also for volatile |
aoqi@1 | 3670 | // registers that the compiler might be keeping live across a safepoint. |
aoqi@1 | 3671 | oop_maps->add_gc_map( __ offset() - start, map); |
aoqi@1 | 3672 | // V0 contains the address we are going to jump to assuming no exception got installed |
aoqi@1 | 3673 | __ get_thread(thread); |
aoqi@1 | 3674 | __ ld_ptr(SP, thread, in_bytes(JavaThread::last_Java_sp_offset())); |
aoqi@1 | 3675 | // clear last_Java_sp |
fujie@9171 | 3676 | __ reset_last_Java_frame(true); |
aoqi@1 | 3677 | // check for pending exceptions |
aoqi@1 | 3678 | Label pending; |
aoqi@1 | 3679 | __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@1 | 3680 | __ bne(AT, R0, pending); |
aoqi@6880 | 3681 | __ delayed()->nop(); |
aoqi@6880 | 3682 | // get the returned Method* |
aoqi@6880 | 3683 | //FIXME, do mips need this ? |
aoqi@1 | 3684 | __ get_vm_result_2(Rmethod, thread); // Refer to OpenJDK8 |
aoqi@1 | 3685 | __ st_ptr(Rmethod, SP, RegisterSaver::methodOffset() * wordSize); |
aoqi@1 | 3686 | __ st_ptr(V0, SP, RegisterSaver::v0Offset() * wordSize); |
aoqi@1 | 3687 | RegisterSaver::restore_live_registers(masm); |
aoqi@1 | 3688 | |
aoqi@1 | 3689 | // We are back the the original state on entry and ready to go the callee method. |
aoqi@1 | 3690 | __ jr(V0); |
aoqi@1 | 3691 | __ delayed()->nop(); |
aoqi@1 | 3692 | // Pending exception after the safepoint |
aoqi@1 | 3693 | |
aoqi@1 | 3694 | __ bind(pending); |
aoqi@1 | 3695 | |
aoqi@1 | 3696 | RegisterSaver::restore_live_registers(masm); |
aoqi@1 | 3697 | |
aoqi@1 | 3698 | // exception pending => remove activation and forward to exception handler |
aoqi@6880 | 3699 | //forward_exception_entry need return address on the stack |
aoqi@1 | 3700 | __ push(RA); |
aoqi@1 | 3701 | __ get_thread(thread); |
aoqi@6880 | 3702 | __ st_ptr(R0, thread, in_bytes(JavaThread::vm_result_offset())); |
aoqi@1 | 3703 | __ ld_ptr(V0, thread, in_bytes(Thread::pending_exception_offset())); |
aoqi@1 | 3704 | __ jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); |
zhaixiang@9144 | 3705 | __ delayed()->nop(); |
aoqi@9459 | 3706 | // |
aoqi@1 | 3707 | // make sure all code is generated |
aoqi@6880 | 3708 | masm->flush(); |
aoqi@1 | 3709 | |
aoqi@1 | 3710 | RuntimeStub* tmp= RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true); |
aoqi@1 | 3711 | return tmp; |
aoqi@1 | 3712 | } |
aoqi@1 | 3713 | |
aoqi@1 | 3714 | extern "C" int SpinPause() {return 0;} |
wanghaomin@9639 | 3715 | |
wanghaomin@9639 | 3716 | |
wanghaomin@9639 | 3717 | //------------------------------Montgomery multiplication------------------------ |
wanghaomin@9639 | 3718 | // |
wanghaomin@9639 | 3719 | |
wanghaomin@9639 | 3720 | // Subtract 0:b from carry:a. Return carry. |
wanghaomin@9639 | 3721 | static unsigned long |
wanghaomin@9639 | 3722 | sub(unsigned long a[], unsigned long b[], unsigned long carry, long len) { |
wanghaomin@9639 | 3723 | long borrow = 0, t = 0; |
wanghaomin@9639 | 3724 | unsigned long tmp0, tmp1; |
wanghaomin@9639 | 3725 | __asm__ __volatile__ ( |
wanghaomin@9639 | 3726 | "0: \n" |
wanghaomin@9639 | 3727 | "ld %[tmp0], 0(%[a]) \n" |
wanghaomin@9639 | 3728 | "ld %[tmp1], 0(%[b]) \n" |
wanghaomin@9639 | 3729 | "sltu %[t], %[tmp0], %[borrow] \n" |
wanghaomin@9639 | 3730 | "dsubu %[tmp0], %[tmp0], %[borrow] \n" |
wanghaomin@9639 | 3731 | "sltu %[borrow], %[tmp0], %[tmp1] \n" |
wanghaomin@9639 | 3732 | "or %[borrow], %[borrow], %[t] \n" |
wanghaomin@9639 | 3733 | "dsubu %[tmp0], %[tmp0], %[tmp1] \n" |
wanghaomin@9639 | 3734 | "sd %[tmp0], 0(%[a]) \n" |
wanghaomin@9639 | 3735 | "daddiu %[a], %[a], 8 \n" |
wanghaomin@9639 | 3736 | "daddiu %[b], %[b], 8 \n" |
wanghaomin@9639 | 3737 | "daddiu %[len], %[len], -1 \n" |
wanghaomin@9639 | 3738 | "bgtz %[len], 0b \n" |
wanghaomin@9639 | 3739 | "dsubu %[tmp0], %[carry], %[borrow] \n" |
wanghaomin@9639 | 3740 | : [len]"+r"(len), [tmp0]"=&r"(tmp0), [tmp1]"=&r"(tmp1), [borrow]"+r"(borrow), [a]"+r"(a), [b]"+r"(b), [t]"+r"(t) |
wanghaomin@9639 | 3741 | : [carry]"r"(carry) |
wanghaomin@9639 | 3742 | : "memory" |
wanghaomin@9639 | 3743 | ); |
wanghaomin@9639 | 3744 | return tmp0; |
wanghaomin@9639 | 3745 | } |
wanghaomin@9639 | 3746 | |
wanghaomin@9639 | 3747 | // Multiply (unsigned) Long A by Long B, accumulating the double- |
wanghaomin@9639 | 3748 | // length result into the accumulator formed of t0, t1, and t2. |
wanghaomin@9639 | 3749 | inline void MACC(unsigned long A, unsigned long B, unsigned long &t0, unsigned long &t1, unsigned long &t2) { |
wanghaomin@9639 | 3750 | unsigned long hi, lo, carry = 0, t = 0; |
wanghaomin@9639 | 3751 | __asm__ __volatile__( |
wanghaomin@9639 | 3752 | "dmultu %[A], %[B] \n" |
wanghaomin@9639 | 3753 | "mfhi %[hi] \n" |
wanghaomin@9639 | 3754 | "mflo %[lo] \n" |
wanghaomin@9639 | 3755 | "daddu %[t0], %[t0], %[lo] \n" |
wanghaomin@9639 | 3756 | "sltu %[carry], %[t0], %[lo] \n" |
wanghaomin@9639 | 3757 | "daddu %[t1], %[t1], %[carry] \n" |
wanghaomin@9639 | 3758 | "sltu %[t], %[t1], %[carry] \n" |
wanghaomin@9639 | 3759 | "daddu %[t1], %[t1], %[hi] \n" |
wanghaomin@9639 | 3760 | "sltu %[carry], %[t1], %[hi] \n" |
wanghaomin@9639 | 3761 | "or %[carry], %[carry], %[t] \n" |
wanghaomin@9639 | 3762 | "daddu %[t2], %[t2], %[carry] \n" |
wanghaomin@9639 | 3763 | : [hi]"=&r"(hi), [lo]"=&r"(lo), [t0]"+r"(t0), [t1]"+r"(t1), [t2]"+r"(t2), [carry]"+r"(carry), [t]"+r"(t) |
wanghaomin@9639 | 3764 | : [A]"r"(A), [B]"r"(B) |
wanghaomin@9639 | 3765 | : |
wanghaomin@9639 | 3766 | ); |
wanghaomin@9639 | 3767 | } |
wanghaomin@9639 | 3768 | |
wanghaomin@9639 | 3769 | // As above, but add twice the double-length result into the |
wanghaomin@9639 | 3770 | // accumulator. |
wanghaomin@9639 | 3771 | inline void MACC2(unsigned long A, unsigned long B, unsigned long &t0, unsigned long &t1, unsigned long &t2) { |
wanghaomin@9639 | 3772 | unsigned long hi, lo, carry = 0, t = 0; |
wanghaomin@9639 | 3773 | __asm__ __volatile__( |
wanghaomin@9639 | 3774 | "dmultu %[A], %[B] \n" |
wanghaomin@9639 | 3775 | "mfhi %[hi] \n" |
wanghaomin@9639 | 3776 | "mflo %[lo] \n" |
wanghaomin@9639 | 3777 | "daddu %[t0], %[t0], %[lo] \n" |
wanghaomin@9639 | 3778 | "sltu %[carry], %[t0], %[lo] \n" |
wanghaomin@9639 | 3779 | "daddu %[t1], %[t1], %[carry] \n" |
wanghaomin@9639 | 3780 | "sltu %[t], %[t1], %[carry] \n" |
wanghaomin@9639 | 3781 | "daddu %[t1], %[t1], %[hi] \n" |
wanghaomin@9639 | 3782 | "sltu %[carry], %[t1], %[hi] \n" |
wanghaomin@9639 | 3783 | "or %[carry], %[carry], %[t] \n" |
wanghaomin@9639 | 3784 | "daddu %[t2], %[t2], %[carry] \n" |
wanghaomin@9639 | 3785 | "daddu %[t0], %[t0], %[lo] \n" |
wanghaomin@9639 | 3786 | "sltu %[carry], %[t0], %[lo] \n" |
wanghaomin@9639 | 3787 | "daddu %[t1], %[t1], %[carry] \n" |
wanghaomin@9639 | 3788 | "sltu %[t], %[t1], %[carry] \n" |
wanghaomin@9639 | 3789 | "daddu %[t1], %[t1], %[hi] \n" |
wanghaomin@9639 | 3790 | "sltu %[carry], %[t1], %[hi] \n" |
wanghaomin@9639 | 3791 | "or %[carry], %[carry], %[t] \n" |
wanghaomin@9639 | 3792 | "daddu %[t2], %[t2], %[carry] \n" |
wanghaomin@9639 | 3793 | : [hi]"=&r"(hi), [lo]"=&r"(lo), [t0]"+r"(t0), [t1]"+r"(t1), [t2]"+r"(t2), [carry]"+r"(carry), [t]"+r"(t) |
wanghaomin@9639 | 3794 | : [A]"r"(A), [B]"r"(B) |
wanghaomin@9639 | 3795 | : |
wanghaomin@9639 | 3796 | ); |
wanghaomin@9639 | 3797 | } |
wanghaomin@9639 | 3798 | |
wanghaomin@9639 | 3799 | // Fast Montgomery multiplication. The derivation of the algorithm is |
wanghaomin@9639 | 3800 | // in A Cryptographic Library for the Motorola DSP56000, |
wanghaomin@9639 | 3801 | // Dusse and Kaliski, Proc. EUROCRYPT 90, pp. 230-237. |
wanghaomin@9639 | 3802 | |
wanghaomin@9639 | 3803 | static void __attribute__((noinline)) |
wanghaomin@9639 | 3804 | montgomery_multiply(unsigned long a[], unsigned long b[], unsigned long n[], |
wanghaomin@9639 | 3805 | unsigned long m[], unsigned long inv, int len) { |
wanghaomin@9639 | 3806 | unsigned long t0 = 0, t1 = 0, t2 = 0; // Triple-precision accumulator |
wanghaomin@9639 | 3807 | int i; |
wanghaomin@9639 | 3808 | |
wanghaomin@9639 | 3809 | assert(inv * n[0] == -1UL, "broken inverse in Montgomery multiply"); |
wanghaomin@9639 | 3810 | |
wanghaomin@9639 | 3811 | for (i = 0; i < len; i++) { |
wanghaomin@9639 | 3812 | int j; |
wanghaomin@9639 | 3813 | for (j = 0; j < i; j++) { |
wanghaomin@9639 | 3814 | MACC(a[j], b[i-j], t0, t1, t2); |
wanghaomin@9639 | 3815 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3816 | } |
wanghaomin@9639 | 3817 | MACC(a[i], b[0], t0, t1, t2); |
wanghaomin@9639 | 3818 | m[i] = t0 * inv; |
wanghaomin@9639 | 3819 | MACC(m[i], n[0], t0, t1, t2); |
wanghaomin@9639 | 3820 | |
wanghaomin@9639 | 3821 | assert(t0 == 0, "broken Montgomery multiply"); |
wanghaomin@9639 | 3822 | |
wanghaomin@9639 | 3823 | t0 = t1; t1 = t2; t2 = 0; |
wanghaomin@9639 | 3824 | } |
wanghaomin@9639 | 3825 | |
wanghaomin@9639 | 3826 | for (i = len; i < 2*len; i++) { |
wanghaomin@9639 | 3827 | int j; |
wanghaomin@9639 | 3828 | for (j = i-len+1; j < len; j++) { |
wanghaomin@9639 | 3829 | MACC(a[j], b[i-j], t0, t1, t2); |
wanghaomin@9639 | 3830 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3831 | } |
wanghaomin@9639 | 3832 | m[i-len] = t0; |
wanghaomin@9639 | 3833 | t0 = t1; t1 = t2; t2 = 0; |
wanghaomin@9639 | 3834 | } |
wanghaomin@9639 | 3835 | |
wanghaomin@9639 | 3836 | while (t0) |
wanghaomin@9639 | 3837 | t0 = sub(m, n, t0, len); |
wanghaomin@9639 | 3838 | } |
wanghaomin@9639 | 3839 | |
wanghaomin@9639 | 3840 | // Fast Montgomery squaring. This uses asymptotically 25% fewer |
wanghaomin@9639 | 3841 | // multiplies so it should be up to 25% faster than Montgomery |
wanghaomin@9639 | 3842 | // multiplication. However, its loop control is more complex and it |
wanghaomin@9639 | 3843 | // may actually run slower on some machines. |
wanghaomin@9639 | 3844 | |
wanghaomin@9639 | 3845 | static void __attribute__((noinline)) |
wanghaomin@9639 | 3846 | montgomery_square(unsigned long a[], unsigned long n[], |
wanghaomin@9639 | 3847 | unsigned long m[], unsigned long inv, int len) { |
wanghaomin@9639 | 3848 | unsigned long t0 = 0, t1 = 0, t2 = 0; // Triple-precision accumulator |
wanghaomin@9639 | 3849 | int i; |
wanghaomin@9639 | 3850 | |
wanghaomin@9639 | 3851 | assert(inv * n[0] == -1UL, "broken inverse in Montgomery multiply"); |
wanghaomin@9639 | 3852 | |
wanghaomin@9639 | 3853 | for (i = 0; i < len; i++) { |
wanghaomin@9639 | 3854 | int j; |
wanghaomin@9639 | 3855 | int end = (i+1)/2; |
wanghaomin@9639 | 3856 | for (j = 0; j < end; j++) { |
wanghaomin@9639 | 3857 | MACC2(a[j], a[i-j], t0, t1, t2); |
wanghaomin@9639 | 3858 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3859 | } |
wanghaomin@9639 | 3860 | if ((i & 1) == 0) { |
wanghaomin@9639 | 3861 | MACC(a[j], a[j], t0, t1, t2); |
wanghaomin@9639 | 3862 | } |
wanghaomin@9639 | 3863 | for (; j < i; j++) { |
wanghaomin@9639 | 3864 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3865 | } |
wanghaomin@9639 | 3866 | m[i] = t0 * inv; |
wanghaomin@9639 | 3867 | MACC(m[i], n[0], t0, t1, t2); |
wanghaomin@9639 | 3868 | |
wanghaomin@9639 | 3869 | assert(t0 == 0, "broken Montgomery square"); |
wanghaomin@9639 | 3870 | |
wanghaomin@9639 | 3871 | t0 = t1; t1 = t2; t2 = 0; |
wanghaomin@9639 | 3872 | } |
wanghaomin@9639 | 3873 | |
wanghaomin@9639 | 3874 | for (i = len; i < 2*len; i++) { |
wanghaomin@9639 | 3875 | int start = i-len+1; |
wanghaomin@9639 | 3876 | int end = start + (len - start)/2; |
wanghaomin@9639 | 3877 | int j; |
wanghaomin@9639 | 3878 | for (j = start; j < end; j++) { |
wanghaomin@9639 | 3879 | MACC2(a[j], a[i-j], t0, t1, t2); |
wanghaomin@9639 | 3880 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3881 | } |
wanghaomin@9639 | 3882 | if ((i & 1) == 0) { |
wanghaomin@9639 | 3883 | MACC(a[j], a[j], t0, t1, t2); |
wanghaomin@9639 | 3884 | } |
wanghaomin@9639 | 3885 | for (; j < len; j++) { |
wanghaomin@9639 | 3886 | MACC(m[j], n[i-j], t0, t1, t2); |
wanghaomin@9639 | 3887 | } |
wanghaomin@9639 | 3888 | m[i-len] = t0; |
wanghaomin@9639 | 3889 | t0 = t1; t1 = t2; t2 = 0; |
wanghaomin@9639 | 3890 | } |
wanghaomin@9639 | 3891 | |
wanghaomin@9639 | 3892 | while (t0) |
wanghaomin@9639 | 3893 | t0 = sub(m, n, t0, len); |
wanghaomin@9639 | 3894 | } |
wanghaomin@9639 | 3895 | |
wanghaomin@9639 | 3896 | // Swap words in a longword. |
wanghaomin@9639 | 3897 | static unsigned long swap(unsigned long x) { |
wanghaomin@9639 | 3898 | return (x << 32) | (x >> 32); |
wanghaomin@9639 | 3899 | } |
wanghaomin@9639 | 3900 | |
wanghaomin@9639 | 3901 | // Copy len longwords from s to d, word-swapping as we go. The |
wanghaomin@9639 | 3902 | // destination array is reversed. |
wanghaomin@9639 | 3903 | static void reverse_words(unsigned long *s, unsigned long *d, int len) { |
wanghaomin@9639 | 3904 | d += len; |
wanghaomin@9639 | 3905 | while(len-- > 0) { |
wanghaomin@9639 | 3906 | d--; |
wanghaomin@9639 | 3907 | *d = swap(*s); |
wanghaomin@9639 | 3908 | s++; |
wanghaomin@9639 | 3909 | } |
wanghaomin@9639 | 3910 | } |
wanghaomin@9639 | 3911 | |
wanghaomin@9639 | 3912 | // The threshold at which squaring is advantageous was determined |
wanghaomin@9639 | 3913 | // experimentally on an i7-3930K (Ivy Bridge) CPU @ 3.5GHz. |
wanghaomin@9639 | 3914 | // Doesn't seem to be relevant for MIPS64 so we use the same value. |
wanghaomin@9639 | 3915 | #define MONTGOMERY_SQUARING_THRESHOLD 64 |
wanghaomin@9639 | 3916 | |
wanghaomin@9639 | 3917 | void SharedRuntime::montgomery_multiply(jint *a_ints, jint *b_ints, jint *n_ints, |
wanghaomin@9639 | 3918 | jint len, jlong inv, |
wanghaomin@9639 | 3919 | jint *m_ints) { |
wanghaomin@9639 | 3920 | assert(len % 2 == 0, "array length in montgomery_multiply must be even"); |
wanghaomin@9639 | 3921 | int longwords = len/2; |
wanghaomin@9639 | 3922 | |
wanghaomin@9639 | 3923 | // Make very sure we don't use so much space that the stack might |
wanghaomin@9639 | 3924 | // overflow. 512 jints corresponds to an 16384-bit integer and |
wanghaomin@9639 | 3925 | // will use here a total of 8k bytes of stack space. |
wanghaomin@9639 | 3926 | int total_allocation = longwords * sizeof (unsigned long) * 4; |
wanghaomin@9639 | 3927 | guarantee(total_allocation <= 8192, "must be"); |
wanghaomin@9639 | 3928 | unsigned long *scratch = (unsigned long *)alloca(total_allocation); |
wanghaomin@9639 | 3929 | |
wanghaomin@9639 | 3930 | // Local scratch arrays |
wanghaomin@9639 | 3931 | unsigned long |
wanghaomin@9639 | 3932 | *a = scratch + 0 * longwords, |
wanghaomin@9639 | 3933 | *b = scratch + 1 * longwords, |
wanghaomin@9639 | 3934 | *n = scratch + 2 * longwords, |
wanghaomin@9639 | 3935 | *m = scratch + 3 * longwords; |
wanghaomin@9639 | 3936 | |
wanghaomin@9639 | 3937 | reverse_words((unsigned long *)a_ints, a, longwords); |
wanghaomin@9639 | 3938 | reverse_words((unsigned long *)b_ints, b, longwords); |
wanghaomin@9639 | 3939 | reverse_words((unsigned long *)n_ints, n, longwords); |
wanghaomin@9639 | 3940 | |
wanghaomin@9639 | 3941 | ::montgomery_multiply(a, b, n, m, (unsigned long)inv, longwords); |
wanghaomin@9639 | 3942 | |
wanghaomin@9639 | 3943 | reverse_words(m, (unsigned long *)m_ints, longwords); |
wanghaomin@9639 | 3944 | } |
wanghaomin@9639 | 3945 | |
wanghaomin@9639 | 3946 | void SharedRuntime::montgomery_square(jint *a_ints, jint *n_ints, |
wanghaomin@9639 | 3947 | jint len, jlong inv, |
wanghaomin@9639 | 3948 | jint *m_ints) { |
wanghaomin@9639 | 3949 | assert(len % 2 == 0, "array length in montgomery_square must be even"); |
wanghaomin@9639 | 3950 | int longwords = len/2; |
wanghaomin@9639 | 3951 | |
wanghaomin@9639 | 3952 | // Make very sure we don't use so much space that the stack might |
wanghaomin@9639 | 3953 | // overflow. 512 jints corresponds to an 16384-bit integer and |
wanghaomin@9639 | 3954 | // will use here a total of 6k bytes of stack space. |
wanghaomin@9639 | 3955 | int total_allocation = longwords * sizeof (unsigned long) * 3; |
wanghaomin@9639 | 3956 | guarantee(total_allocation <= 8192, "must be"); |
wanghaomin@9639 | 3957 | unsigned long *scratch = (unsigned long *)alloca(total_allocation); |
wanghaomin@9639 | 3958 | |
wanghaomin@9639 | 3959 | // Local scratch arrays |
wanghaomin@9639 | 3960 | unsigned long |
wanghaomin@9639 | 3961 | *a = scratch + 0 * longwords, |
wanghaomin@9639 | 3962 | *n = scratch + 1 * longwords, |
wanghaomin@9639 | 3963 | *m = scratch + 2 * longwords; |
wanghaomin@9639 | 3964 | |
wanghaomin@9639 | 3965 | reverse_words((unsigned long *)a_ints, a, longwords); |
wanghaomin@9639 | 3966 | reverse_words((unsigned long *)n_ints, n, longwords); |
wanghaomin@9639 | 3967 | |
wanghaomin@9639 | 3968 | if (len >= MONTGOMERY_SQUARING_THRESHOLD) { |
wanghaomin@9639 | 3969 | ::montgomery_square(a, n, m, (unsigned long)inv, longwords); |
wanghaomin@9639 | 3970 | } else { |
wanghaomin@9639 | 3971 | ::montgomery_multiply(a, a, n, m, (unsigned long)inv, longwords); |
wanghaomin@9639 | 3972 | } |
wanghaomin@9639 | 3973 | |
wanghaomin@9639 | 3974 | reverse_words(m, (unsigned long *)m_ints, longwords); |
wanghaomin@9639 | 3975 | } |