|
1 /* |
|
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. |
|
3 * Copyright (c) 2015, 2016, Loongson Technology. All rights reserved. |
|
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
5 * |
|
6 * This code is free software; you can redistribute it and/or modify it |
|
7 * under the terms of the GNU General Public License version 2 only, as |
|
8 * published by the Free Software Foundation. |
|
9 * |
|
10 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
13 * version 2 for more details (a copy is included in the LICENSE file that |
|
14 * accompanied this code). |
|
15 * |
|
16 * You should have received a copy of the GNU General Public License version |
|
17 * 2 along with this work; if not, write to the Free Software Foundation, |
|
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
19 * |
|
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
21 * or visit www.oracle.com if you need additional information or have any |
|
22 * questions. |
|
23 * |
|
24 */ |
|
25 |
|
26 #include "precompiled.hpp" |
|
27 #include "asm/macroAssembler.hpp" |
|
28 #include "asm/macroAssembler.inline.hpp" |
|
29 #include "code/debugInfoRec.hpp" |
|
30 #include "code/icBuffer.hpp" |
|
31 #include "code/vtableStubs.hpp" |
|
32 #include "interpreter/interpreter.hpp" |
|
33 #include "oops/compiledICHolder.hpp" |
|
34 #include "prims/jvmtiRedefineClassesTrace.hpp" |
|
35 #include "runtime/sharedRuntime.hpp" |
|
36 #include "runtime/vframeArray.hpp" |
|
37 #include "vmreg_mips.inline.hpp" |
|
38 #ifdef COMPILER1 |
|
39 #include "c1/c1_Runtime1.hpp" |
|
40 #endif |
|
41 #ifdef COMPILER2 |
|
42 #include "opto/runtime.hpp" |
|
43 #endif |
|
44 |
|
45 #define __ masm-> |
|
46 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; |
|
47 |
|
48 class RegisterSaver { |
|
49 enum { FPU_regs_live = 32 }; |
|
50 // Capture info about frame layout |
|
51 enum layout { |
|
52 #define DEF_LAYOUT_OFFS(regname) regname ## _off, regname ## H_off, |
|
53 DEF_LAYOUT_OFFS(for_16_bytes_aligned) |
|
54 DEF_LAYOUT_OFFS(fpr0) |
|
55 DEF_LAYOUT_OFFS(fpr1) |
|
56 DEF_LAYOUT_OFFS(fpr2) |
|
57 DEF_LAYOUT_OFFS(fpr3) |
|
58 DEF_LAYOUT_OFFS(fpr4) |
|
59 DEF_LAYOUT_OFFS(fpr5) |
|
60 DEF_LAYOUT_OFFS(fpr6) |
|
61 DEF_LAYOUT_OFFS(fpr7) |
|
62 DEF_LAYOUT_OFFS(fpr8) |
|
63 DEF_LAYOUT_OFFS(fpr9) |
|
64 DEF_LAYOUT_OFFS(fpr10) |
|
65 DEF_LAYOUT_OFFS(fpr11) |
|
66 DEF_LAYOUT_OFFS(fpr12) |
|
67 DEF_LAYOUT_OFFS(fpr13) |
|
68 DEF_LAYOUT_OFFS(fpr14) |
|
69 DEF_LAYOUT_OFFS(fpr15) |
|
70 DEF_LAYOUT_OFFS(fpr16) |
|
71 DEF_LAYOUT_OFFS(fpr17) |
|
72 DEF_LAYOUT_OFFS(fpr18) |
|
73 DEF_LAYOUT_OFFS(fpr19) |
|
74 DEF_LAYOUT_OFFS(fpr20) |
|
75 DEF_LAYOUT_OFFS(fpr21) |
|
76 DEF_LAYOUT_OFFS(fpr22) |
|
77 DEF_LAYOUT_OFFS(fpr23) |
|
78 DEF_LAYOUT_OFFS(fpr24) |
|
79 DEF_LAYOUT_OFFS(fpr25) |
|
80 DEF_LAYOUT_OFFS(fpr26) |
|
81 DEF_LAYOUT_OFFS(fpr27) |
|
82 DEF_LAYOUT_OFFS(fpr28) |
|
83 DEF_LAYOUT_OFFS(fpr29) |
|
84 DEF_LAYOUT_OFFS(fpr30) |
|
85 DEF_LAYOUT_OFFS(fpr31) |
|
86 |
|
87 DEF_LAYOUT_OFFS(v0) |
|
88 DEF_LAYOUT_OFFS(v1) |
|
89 DEF_LAYOUT_OFFS(a0) |
|
90 DEF_LAYOUT_OFFS(a1) |
|
91 DEF_LAYOUT_OFFS(a2) |
|
92 DEF_LAYOUT_OFFS(a3) |
|
93 DEF_LAYOUT_OFFS(a4) |
|
94 DEF_LAYOUT_OFFS(a5) |
|
95 DEF_LAYOUT_OFFS(a6) |
|
96 DEF_LAYOUT_OFFS(a7) |
|
97 DEF_LAYOUT_OFFS(t0) |
|
98 DEF_LAYOUT_OFFS(t1) |
|
99 DEF_LAYOUT_OFFS(t2) |
|
100 DEF_LAYOUT_OFFS(t3) |
|
101 DEF_LAYOUT_OFFS(s0) |
|
102 DEF_LAYOUT_OFFS(s1) |
|
103 DEF_LAYOUT_OFFS(s2) |
|
104 DEF_LAYOUT_OFFS(s3) |
|
105 DEF_LAYOUT_OFFS(s4) |
|
106 DEF_LAYOUT_OFFS(s5) |
|
107 DEF_LAYOUT_OFFS(s6) |
|
108 DEF_LAYOUT_OFFS(s7) |
|
109 DEF_LAYOUT_OFFS(t8) |
|
110 DEF_LAYOUT_OFFS(t9) |
|
111 |
|
112 DEF_LAYOUT_OFFS(gp) |
|
113 DEF_LAYOUT_OFFS(fp) |
|
114 DEF_LAYOUT_OFFS(return) |
|
115 /* |
|
116 fpr0_off, fpr1_off, |
|
117 fpr2_off, fpr3_off, |
|
118 fpr4_off, fpr5_off, |
|
119 fpr6_off, fpr7_off, |
|
120 fpr8_off, fpr9_off, |
|
121 fpr10_off, fpr11_off, |
|
122 fpr12_off, fpr13_off, |
|
123 fpr14_off, fpr15_off, |
|
124 fpr16_off, fpr17_off, |
|
125 fpr18_off, fpr19_off, |
|
126 fpr20_off, fpr21_off, |
|
127 fpr22_off, fpr23_off, |
|
128 fpr24_off, fpr25_off, |
|
129 fpr26_off, fpr27_off, |
|
130 fpr28_off, fpr29_off, |
|
131 fpr30_off, fpr31_off, |
|
132 |
|
133 v0_off, v1_off, |
|
134 a0_off, a1_off, |
|
135 a2_off, a3_off, |
|
136 a4_off, a5_off, |
|
137 a6_off, a7_off, |
|
138 t0_off, t1_off, t2_off, t3_off, |
|
139 s0_off, s1_off, s2_off, s3_off, s4_off, s5_off, s6_off, s7_off, |
|
140 t8_off, t9_off, |
|
141 |
|
142 gp_off, fp_off, |
|
143 return_off, |
|
144 */ |
|
145 reg_save_size |
|
146 }; |
|
147 |
|
148 public: |
|
149 |
|
150 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors =false ); |
|
151 static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false); |
|
152 //FIXME, I have no idea which register to use |
|
153 static int raOffset(void) { return return_off / 2; } |
|
154 //Rmethod |
|
155 static int methodOffset(void) { return s3_off / 2; } |
|
156 |
|
157 static int v0Offset(void) { return v0_off / 2; } |
|
158 static int v1Offset(void) { return v1_off / 2; } |
|
159 |
|
160 static int fpResultOffset(void) { return fpr0_off / 2; } |
|
161 |
|
162 // During deoptimization only the result register need to be restored |
|
163 // all the other values have already been extracted. |
|
164 |
|
165 static void restore_result_registers(MacroAssembler* masm); |
|
166 }; |
|
167 |
|
168 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors ) { |
|
169 |
|
170 /* |
|
171 int frame_words = reg_save_size + additional_frame_words; |
|
172 int frame_size_in_bytes = frame_words * wordSize; |
|
173 *total_frame_words = frame_words; |
|
174 */ |
|
175 // Always make the frame size 16-byte aligned |
|
176 int frame_size_in_bytes = round_to(additional_frame_words*wordSize + |
|
177 reg_save_size*BytesPerInt, 16); |
|
178 // OopMap frame size is in compiler stack slots (jint's) not bytes or words |
|
179 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt; |
|
180 // The caller will allocate additional_frame_words |
|
181 int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt; |
|
182 // CodeBlob frame size is in words. |
|
183 int frame_size_in_words = frame_size_in_bytes / wordSize; |
|
184 *total_frame_words = frame_size_in_words; |
|
185 |
|
186 // save registers, fpu state, and flags |
|
187 // We assume caller has already has return address slot on the stack |
|
188 // We push epb twice in this sequence because we want the real ebp |
|
189 // to be under the return like a normal enter and we want to use pushad |
|
190 // We push by hand instead of pusing push |
|
191 |
|
192 __ daddiu(SP, SP, - reg_save_size * jintSize); |
|
193 |
|
194 __ sdc1(F0, SP, fpr0_off * jintSize); __ sdc1(F1, SP, fpr1_off * jintSize); |
|
195 __ sdc1(F2, SP, fpr2_off * jintSize); __ sdc1(F3, SP, fpr3_off * jintSize); |
|
196 __ sdc1(F4, SP, fpr4_off * jintSize); __ sdc1(F5, SP, fpr5_off * jintSize); |
|
197 __ sdc1(F6, SP, fpr6_off * jintSize); __ sdc1(F7, SP, fpr7_off * jintSize); |
|
198 __ sdc1(F8, SP, fpr8_off * jintSize); __ sdc1(F9, SP, fpr9_off * jintSize); |
|
199 __ sdc1(F10, SP, fpr10_off * jintSize); __ sdc1(F11, SP, fpr11_off * jintSize); |
|
200 __ sdc1(F12, SP, fpr12_off * jintSize); __ sdc1(F13, SP, fpr13_off * jintSize); |
|
201 __ sdc1(F14, SP, fpr14_off * jintSize); __ sdc1(F15, SP, fpr15_off * jintSize); |
|
202 __ sdc1(F16, SP, fpr16_off * jintSize); __ sdc1(F17, SP, fpr17_off * jintSize); |
|
203 __ sdc1(F18, SP, fpr18_off * jintSize); __ sdc1(F19, SP, fpr19_off * jintSize); |
|
204 __ sdc1(F20, SP, fpr20_off * jintSize); __ sdc1(F21, SP, fpr21_off * jintSize); |
|
205 __ sdc1(F22, SP, fpr22_off * jintSize); __ sdc1(F23, SP, fpr23_off * jintSize); |
|
206 __ sdc1(F24, SP, fpr24_off * jintSize); __ sdc1(F25, SP, fpr25_off * jintSize); |
|
207 __ sdc1(F26, SP, fpr26_off * jintSize); __ sdc1(F27, SP, fpr27_off * jintSize); |
|
208 __ sdc1(F28, SP, fpr28_off * jintSize); __ sdc1(F29, SP, fpr29_off * jintSize); |
|
209 __ sdc1(F30, SP, fpr30_off * jintSize); __ sdc1(F31, SP, fpr31_off * jintSize); |
|
210 __ sd(V0, SP, v0_off * jintSize); __ sd(V1, SP, v1_off * jintSize); |
|
211 __ sd(A0, SP, a0_off * jintSize); __ sd(A1, SP, a1_off * jintSize); |
|
212 __ sd(A2, SP, a2_off * jintSize); __ sd(A3, SP, a3_off * jintSize); |
|
213 __ sd(A4, SP, a4_off * jintSize); __ sd(A5, SP, a5_off * jintSize); |
|
214 __ sd(A6, SP, a6_off * jintSize); __ sd(A7, SP, a7_off * jintSize); |
|
215 __ sd(T0, SP, t0_off * jintSize); |
|
216 __ sd(T1, SP, t1_off * jintSize); |
|
217 __ sd(T2, SP, t2_off * jintSize); |
|
218 __ sd(T3, SP, t3_off * jintSize); |
|
219 __ sd(S0, SP, s0_off * jintSize); |
|
220 __ sd(S1, SP, s1_off * jintSize); |
|
221 __ sd(S2, SP, s2_off * jintSize); |
|
222 __ sd(S3, SP, s3_off * jintSize); |
|
223 __ sd(S4, SP, s4_off * jintSize); |
|
224 __ sd(S5, SP, s5_off * jintSize); |
|
225 __ sd(S6, SP, s6_off * jintSize); |
|
226 __ sd(S7, SP, s7_off * jintSize); |
|
227 |
|
228 __ sd(T8, SP, t8_off * jintSize); |
|
229 __ sd(T9, SP, t9_off * jintSize); |
|
230 |
|
231 __ sd(GP, SP, gp_off * jintSize); |
|
232 __ sd(FP, SP, fp_off * jintSize); |
|
233 __ sd(RA, SP, return_off * jintSize); |
|
234 __ daddi(FP, SP, fp_off * jintSize); |
|
235 |
|
236 OopMapSet *oop_maps = new OopMapSet(); |
|
237 //OopMap* map = new OopMap( frame_words, 0 ); |
|
238 OopMap* map = new OopMap( frame_size_in_slots, 0 ); |
|
239 |
|
240 |
|
241 //#define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words) |
|
242 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_slots) |
|
243 map->set_callee_saved(STACK_OFFSET( v0_off), V0->as_VMReg()); |
|
244 map->set_callee_saved(STACK_OFFSET( v1_off), V1->as_VMReg()); |
|
245 map->set_callee_saved(STACK_OFFSET( a0_off), A0->as_VMReg()); |
|
246 map->set_callee_saved(STACK_OFFSET( a1_off), A1->as_VMReg()); |
|
247 map->set_callee_saved(STACK_OFFSET( a2_off), A2->as_VMReg()); |
|
248 map->set_callee_saved(STACK_OFFSET( a3_off), A3->as_VMReg()); |
|
249 map->set_callee_saved(STACK_OFFSET( a4_off), A4->as_VMReg()); |
|
250 map->set_callee_saved(STACK_OFFSET( a5_off), A5->as_VMReg()); |
|
251 map->set_callee_saved(STACK_OFFSET( a6_off), A6->as_VMReg()); |
|
252 map->set_callee_saved(STACK_OFFSET( a7_off), A7->as_VMReg()); |
|
253 map->set_callee_saved(STACK_OFFSET( t0_off), T0->as_VMReg()); |
|
254 map->set_callee_saved(STACK_OFFSET( t1_off), T1->as_VMReg()); |
|
255 map->set_callee_saved(STACK_OFFSET( t2_off), T2->as_VMReg()); |
|
256 map->set_callee_saved(STACK_OFFSET( t3_off), T3->as_VMReg()); |
|
257 map->set_callee_saved(STACK_OFFSET( s0_off), S0->as_VMReg()); |
|
258 map->set_callee_saved(STACK_OFFSET( s1_off), S1->as_VMReg()); |
|
259 map->set_callee_saved(STACK_OFFSET( s2_off), S2->as_VMReg()); |
|
260 map->set_callee_saved(STACK_OFFSET( s3_off), S3->as_VMReg()); |
|
261 map->set_callee_saved(STACK_OFFSET( s4_off), S4->as_VMReg()); |
|
262 map->set_callee_saved(STACK_OFFSET( s5_off), S5->as_VMReg()); |
|
263 map->set_callee_saved(STACK_OFFSET( s6_off), S6->as_VMReg()); |
|
264 map->set_callee_saved(STACK_OFFSET( s7_off), S7->as_VMReg()); |
|
265 map->set_callee_saved(STACK_OFFSET( t8_off), T8->as_VMReg()); |
|
266 map->set_callee_saved(STACK_OFFSET( t9_off), T9->as_VMReg()); |
|
267 map->set_callee_saved(STACK_OFFSET( gp_off), GP->as_VMReg()); |
|
268 map->set_callee_saved(STACK_OFFSET( fp_off), FP->as_VMReg()); |
|
269 map->set_callee_saved(STACK_OFFSET( return_off), RA->as_VMReg()); |
|
270 |
|
271 map->set_callee_saved(STACK_OFFSET( fpr0_off), F0->as_VMReg()); |
|
272 map->set_callee_saved(STACK_OFFSET( fpr1_off), F1->as_VMReg()); |
|
273 map->set_callee_saved(STACK_OFFSET( fpr2_off), F2->as_VMReg()); |
|
274 map->set_callee_saved(STACK_OFFSET( fpr3_off), F3->as_VMReg()); |
|
275 map->set_callee_saved(STACK_OFFSET( fpr4_off), F4->as_VMReg()); |
|
276 map->set_callee_saved(STACK_OFFSET( fpr5_off), F5->as_VMReg()); |
|
277 map->set_callee_saved(STACK_OFFSET( fpr6_off), F6->as_VMReg()); |
|
278 map->set_callee_saved(STACK_OFFSET( fpr7_off), F7->as_VMReg()); |
|
279 map->set_callee_saved(STACK_OFFSET( fpr8_off), F8->as_VMReg()); |
|
280 map->set_callee_saved(STACK_OFFSET( fpr9_off), F9->as_VMReg()); |
|
281 map->set_callee_saved(STACK_OFFSET( fpr10_off), F10->as_VMReg()); |
|
282 map->set_callee_saved(STACK_OFFSET( fpr11_off), F11->as_VMReg()); |
|
283 map->set_callee_saved(STACK_OFFSET( fpr12_off), F12->as_VMReg()); |
|
284 map->set_callee_saved(STACK_OFFSET( fpr13_off), F13->as_VMReg()); |
|
285 map->set_callee_saved(STACK_OFFSET( fpr14_off), F14->as_VMReg()); |
|
286 map->set_callee_saved(STACK_OFFSET( fpr15_off), F15->as_VMReg()); |
|
287 map->set_callee_saved(STACK_OFFSET( fpr16_off), F16->as_VMReg()); |
|
288 map->set_callee_saved(STACK_OFFSET( fpr17_off), F17->as_VMReg()); |
|
289 map->set_callee_saved(STACK_OFFSET( fpr18_off), F18->as_VMReg()); |
|
290 map->set_callee_saved(STACK_OFFSET( fpr19_off), F19->as_VMReg()); |
|
291 map->set_callee_saved(STACK_OFFSET( fpr20_off), F20->as_VMReg()); |
|
292 map->set_callee_saved(STACK_OFFSET( fpr21_off), F21->as_VMReg()); |
|
293 map->set_callee_saved(STACK_OFFSET( fpr22_off), F22->as_VMReg()); |
|
294 map->set_callee_saved(STACK_OFFSET( fpr23_off), F23->as_VMReg()); |
|
295 map->set_callee_saved(STACK_OFFSET( fpr24_off), F24->as_VMReg()); |
|
296 map->set_callee_saved(STACK_OFFSET( fpr25_off), F25->as_VMReg()); |
|
297 map->set_callee_saved(STACK_OFFSET( fpr26_off), F26->as_VMReg()); |
|
298 map->set_callee_saved(STACK_OFFSET( fpr27_off), F27->as_VMReg()); |
|
299 map->set_callee_saved(STACK_OFFSET( fpr28_off), F28->as_VMReg()); |
|
300 map->set_callee_saved(STACK_OFFSET( fpr29_off), F29->as_VMReg()); |
|
301 map->set_callee_saved(STACK_OFFSET( fpr30_off), F30->as_VMReg()); |
|
302 map->set_callee_saved(STACK_OFFSET( fpr31_off), F31->as_VMReg()); |
|
303 |
|
304 /* |
|
305 if (true) { |
|
306 map->set_callee_saved(STACK_OFFSET( v0H_off), V0->as_VMReg()->next()); |
|
307 map->set_callee_saved(STACK_OFFSET( v1H_off), V1->as_VMReg()->next()); |
|
308 map->set_callee_saved(STACK_OFFSET( a0H_off), A0->as_VMReg()->next()); |
|
309 map->set_callee_saved(STACK_OFFSET( a1H_off), A1->as_VMReg()->next()); |
|
310 map->set_callee_saved(STACK_OFFSET( a2H_off), A2->as_VMReg()->next()); |
|
311 map->set_callee_saved(STACK_OFFSET( a3H_off), A3->as_VMReg()->next()); |
|
312 map->set_callee_saved(STACK_OFFSET( a4H_off), A4->as_VMReg()->next()); |
|
313 map->set_callee_saved(STACK_OFFSET( a5H_off), A5->as_VMReg()->next()); |
|
314 map->set_callee_saved(STACK_OFFSET( a6H_off), A6->as_VMReg()->next()); |
|
315 map->set_callee_saved(STACK_OFFSET( a7H_off), A7->as_VMReg()->next()); |
|
316 map->set_callee_saved(STACK_OFFSET( t0H_off), T0->as_VMReg()->next()); |
|
317 map->set_callee_saved(STACK_OFFSET( t1H_off), T1->as_VMReg()->next()); |
|
318 map->set_callee_saved(STACK_OFFSET( t2H_off), T2->as_VMReg()->next()); |
|
319 map->set_callee_saved(STACK_OFFSET( t3H_off), T3->as_VMReg()->next()); |
|
320 map->set_callee_saved(STACK_OFFSET( s0H_off), S0->as_VMReg()->next()); |
|
321 map->set_callee_saved(STACK_OFFSET( s1H_off), S1->as_VMReg()->next()); |
|
322 map->set_callee_saved(STACK_OFFSET( s2H_off), S2->as_VMReg()->next()); |
|
323 map->set_callee_saved(STACK_OFFSET( s3H_off), S3->as_VMReg()->next()); |
|
324 map->set_callee_saved(STACK_OFFSET( s4H_off), S4->as_VMReg()->next()); |
|
325 map->set_callee_saved(STACK_OFFSET( s5H_off), S5->as_VMReg()->next()); |
|
326 map->set_callee_saved(STACK_OFFSET( s6H_off), S6->as_VMReg()->next()); |
|
327 map->set_callee_saved(STACK_OFFSET( s7H_off), S7->as_VMReg()->next()); |
|
328 map->set_callee_saved(STACK_OFFSET( t8H_off), T8->as_VMReg()->next()); |
|
329 map->set_callee_saved(STACK_OFFSET( t9H_off), T9->as_VMReg()->next()); |
|
330 map->set_callee_saved(STACK_OFFSET( gpH_off), GP->as_VMReg()->next()); |
|
331 map->set_callee_saved(STACK_OFFSET( fpH_off), FP->as_VMReg()->next()); |
|
332 map->set_callee_saved(STACK_OFFSET( returnH_off), RA->as_VMReg()->next()); |
|
333 |
|
334 map->set_callee_saved(STACK_OFFSET( fpr0H_off), F0->as_VMReg()->next()); |
|
335 map->set_callee_saved(STACK_OFFSET( fpr2H_off), F2->as_VMReg()->next()); |
|
336 map->set_callee_saved(STACK_OFFSET( fpr4H_off), F4->as_VMReg()->next()); |
|
337 map->set_callee_saved(STACK_OFFSET( fpr6H_off), F6->as_VMReg()->next()); |
|
338 map->set_callee_saved(STACK_OFFSET( fpr8H_off), F8->as_VMReg()->next()); |
|
339 map->set_callee_saved(STACK_OFFSET( fpr10H_off), F10->as_VMReg()->next()); |
|
340 map->set_callee_saved(STACK_OFFSET( fpr12H_off), F12->as_VMReg()->next()); |
|
341 map->set_callee_saved(STACK_OFFSET( fpr14H_off), F14->as_VMReg()->next()); |
|
342 map->set_callee_saved(STACK_OFFSET( fpr16H_off), F16->as_VMReg()->next()); |
|
343 map->set_callee_saved(STACK_OFFSET( fpr18H_off), F18->as_VMReg()->next()); |
|
344 map->set_callee_saved(STACK_OFFSET( fpr20H_off), F20->as_VMReg()->next()); |
|
345 map->set_callee_saved(STACK_OFFSET( fpr22H_off), F22->as_VMReg()->next()); |
|
346 map->set_callee_saved(STACK_OFFSET( fpr24H_off), F24->as_VMReg()->next()); |
|
347 map->set_callee_saved(STACK_OFFSET( fpr26H_off), F26->as_VMReg()->next()); |
|
348 map->set_callee_saved(STACK_OFFSET( fpr28H_off), F28->as_VMReg()->next()); |
|
349 map->set_callee_saved(STACK_OFFSET( fpr30H_off), F30->as_VMReg()->next()); |
|
350 } |
|
351 */ |
|
352 #undef STACK_OFFSET |
|
353 return map; |
|
354 } |
|
355 |
|
356 |
|
357 // Pop the current frame and restore all the registers that we |
|
358 // saved. |
|
359 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) { |
|
360 __ ldc1(F0, SP, fpr0_off * jintSize); __ ldc1(F1, SP, fpr1_off * jintSize); |
|
361 __ ldc1(F2, SP, fpr2_off * jintSize); __ ldc1(F3, SP, fpr3_off * jintSize); |
|
362 __ ldc1(F4, SP, fpr4_off * jintSize); __ ldc1(F5, SP, fpr5_off * jintSize); |
|
363 __ ldc1(F6, SP, fpr6_off * jintSize); __ ldc1(F7, SP, fpr7_off * jintSize); |
|
364 __ ldc1(F8, SP, fpr8_off * jintSize); __ ldc1(F9, SP, fpr9_off * jintSize); |
|
365 __ ldc1(F10, SP, fpr10_off * jintSize); __ ldc1(F11, SP, fpr11_off * jintSize); |
|
366 __ ldc1(F12, SP, fpr12_off * jintSize); __ ldc1(F13, SP, fpr13_off * jintSize); |
|
367 __ ldc1(F14, SP, fpr14_off * jintSize); __ ldc1(F15, SP, fpr15_off * jintSize); |
|
368 __ ldc1(F16, SP, fpr16_off * jintSize); __ ldc1(F17, SP, fpr17_off * jintSize); |
|
369 __ ldc1(F18, SP, fpr18_off * jintSize); __ ldc1(F19, SP, fpr19_off * jintSize); |
|
370 __ ldc1(F20, SP, fpr20_off * jintSize); __ ldc1(F21, SP, fpr21_off * jintSize); |
|
371 __ ldc1(F22, SP, fpr22_off * jintSize); __ ldc1(F23, SP, fpr23_off * jintSize); |
|
372 __ ldc1(F24, SP, fpr24_off * jintSize); __ ldc1(F25, SP, fpr25_off * jintSize); |
|
373 __ ldc1(F26, SP, fpr26_off * jintSize); __ ldc1(F27, SP, fpr27_off * jintSize); |
|
374 __ ldc1(F28, SP, fpr28_off * jintSize); __ ldc1(F29, SP, fpr29_off * jintSize); |
|
375 __ ldc1(F30, SP, fpr30_off * jintSize); __ ldc1(F31, SP, fpr31_off * jintSize); |
|
376 |
|
377 __ ld(V0, SP, v0_off * jintSize); __ ld(V1, SP, v1_off * jintSize); |
|
378 __ ld(A0, SP, a0_off * jintSize); __ ld(A1, SP, a1_off * jintSize); |
|
379 __ ld(A2, SP, a2_off * jintSize); __ ld(A3, SP, a3_off * jintSize); |
|
380 __ ld(A4, SP, a4_off * jintSize); __ ld(A5, SP, a5_off * jintSize); |
|
381 __ ld(A6, SP, a6_off * jintSize); __ ld(A7, SP, a7_off * jintSize); |
|
382 __ ld(T0, SP, t0_off * jintSize); |
|
383 __ ld(T1, SP, t1_off * jintSize); |
|
384 __ ld(T2, SP, t2_off * jintSize); |
|
385 __ ld(T3, SP, t3_off * jintSize); |
|
386 __ ld(S0, SP, s0_off * jintSize); |
|
387 __ ld(S1, SP, s1_off * jintSize); |
|
388 __ ld(S2, SP, s2_off * jintSize); |
|
389 __ ld(S3, SP, s3_off * jintSize); |
|
390 __ ld(S4, SP, s4_off * jintSize); |
|
391 __ ld(S5, SP, s5_off * jintSize); |
|
392 __ ld(S6, SP, s6_off * jintSize); |
|
393 __ ld(S7, SP, s7_off * jintSize); |
|
394 |
|
395 __ ld(T8, SP, t8_off * jintSize); |
|
396 __ ld(T9, SP, t9_off * jintSize); |
|
397 |
|
398 __ ld(GP, SP, gp_off * jintSize); |
|
399 __ ld(FP, SP, fp_off * jintSize); |
|
400 __ ld(RA, SP, return_off * jintSize); |
|
401 |
|
402 __ addiu(SP, SP, reg_save_size * jintSize); |
|
403 } |
|
404 |
|
405 // Pop the current frame and restore the registers that might be holding |
|
406 // a result. |
|
407 // FIXME, if the result is float? |
|
408 void RegisterSaver::restore_result_registers(MacroAssembler* masm) { |
|
409 // Just restore result register. Only used by deoptimization. By |
|
410 // now any callee save register that needs to be restore to a c2 |
|
411 // caller of the deoptee has been extracted into the vframeArray |
|
412 // and will be stuffed into the c2i adapter we create for later |
|
413 // restoration so only result registers need to be restored here. |
|
414 // |
|
415 __ ld(V0, SP, v0_off * jintSize); |
|
416 __ ld(V1, SP, v1_off * jintSize); |
|
417 __ addiu(SP, SP, return_off * jintSize); |
|
418 } |
|
419 |
|
420 // Is vector's size (in bytes) bigger than a size saved by default? |
|
421 // 16 bytes XMM registers are saved by default using fxsave/fxrstor instructions. |
|
422 bool SharedRuntime::is_wide_vector(int size) { |
|
423 return size > 16; |
|
424 } |
|
425 |
|
426 // The java_calling_convention describes stack locations as ideal slots on |
|
427 // a frame with no abi restrictions. Since we must observe abi restrictions |
|
428 // (like the placement of the register window) the slots must be biased by |
|
429 // the following value. |
|
430 |
|
431 static int reg2offset_in(VMReg r) { |
|
432 // Account for saved ebp and return address |
|
433 // This should really be in_preserve_stack_slots |
|
434 return (r->reg2stack() + 2 * VMRegImpl::slots_per_word) * VMRegImpl::stack_slot_size; // + 2 * VMRegImpl::stack_slot_size); |
|
435 } |
|
436 |
|
437 static int reg2offset_out(VMReg r) { |
|
438 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; |
|
439 } |
|
440 |
|
441 // --------------------------------------------------------------------------- |
|
442 // Read the array of BasicTypes from a signature, and compute where the |
|
443 // arguments should go. Values in the VMRegPair regs array refer to 4-byte |
|
444 // quantities. Values less than SharedInfo::stack0 are registers, those above |
|
445 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer |
|
446 // as framesizes are fixed. |
|
447 // VMRegImpl::stack0 refers to the first slot 0(sp). |
|
448 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register |
|
449 // up to RegisterImpl::number_of_registers) are the 32-bit |
|
450 // integer registers. |
|
451 |
|
452 // Pass first five oop/int args in registers T0, A0 - A3. |
|
453 // Pass float/double/long args in stack. |
|
454 // Doubles have precedence, so if you pass a mix of floats and doubles |
|
455 // the doubles will grab the registers before the floats will. |
|
456 |
|
457 // Note: the INPUTS in sig_bt are in units of Java argument words, which are |
|
458 // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit |
|
459 // units regardless of build. Of course for i486 there is no 64 bit build |
|
460 |
|
461 |
|
462 // --------------------------------------------------------------------------- |
|
463 // The compiled Java calling convention. |
|
464 // Pass first five oop/int args in registers T0, A0 - A3. |
|
465 // Pass float/double/long args in stack. |
|
466 // Doubles have precedence, so if you pass a mix of floats and doubles |
|
467 // the doubles will grab the registers before the floats will. |
|
468 |
|
469 int SharedRuntime::java_calling_convention(const BasicType *sig_bt, |
|
470 VMRegPair *regs, |
|
471 int total_args_passed, |
|
472 int is_outgoing) { |
|
473 //#define aoqi_test |
|
474 #ifdef aoqi_test |
|
475 tty->print_cr(" SharedRuntime::%s :%d, total_args_passed: %d", __func__, __LINE__, total_args_passed); |
|
476 #endif |
|
477 |
|
478 // Create the mapping between argument positions and |
|
479 // registers. |
|
480 //static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = { |
|
481 static const Register INT_ArgReg[Argument::n_register_parameters + 1] = { |
|
482 T0, A0, A1, A2, A3, A4, A5, A6, A7 |
|
483 }; |
|
484 //static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = { |
|
485 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters] = { |
|
486 F12, F13, F14, F15, F16, F17, F18, F19 |
|
487 }; |
|
488 |
|
489 |
|
490 uint args = 0; |
|
491 uint stk_args = 0; // inc by 2 each time |
|
492 |
|
493 for (int i = 0; i < total_args_passed; i++) { |
|
494 switch (sig_bt[i]) { |
|
495 case T_VOID: |
|
496 // halves of T_LONG or T_DOUBLE |
|
497 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
|
498 regs[i].set_bad(); |
|
499 break; |
|
500 case T_BOOLEAN: |
|
501 case T_CHAR: |
|
502 case T_BYTE: |
|
503 case T_SHORT: |
|
504 case T_INT: |
|
505 if (args < Argument::n_register_parameters) { |
|
506 regs[i].set1(INT_ArgReg[args++]->as_VMReg()); |
|
507 } else { |
|
508 regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
|
509 stk_args += 2; |
|
510 } |
|
511 break; |
|
512 case T_LONG: |
|
513 assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
|
514 // fall through |
|
515 case T_OBJECT: |
|
516 case T_ARRAY: |
|
517 case T_ADDRESS: |
|
518 if (args < Argument::n_register_parameters) { |
|
519 regs[i].set2(INT_ArgReg[args++]->as_VMReg()); |
|
520 } else { |
|
521 regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
|
522 stk_args += 2; |
|
523 } |
|
524 break; |
|
525 case T_FLOAT: |
|
526 if (args < Argument::n_float_register_parameters) { |
|
527 regs[i].set1(FP_ArgReg[args++]->as_VMReg()); |
|
528 } else { |
|
529 regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
|
530 stk_args += 2; |
|
531 } |
|
532 break; |
|
533 case T_DOUBLE: |
|
534 assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
|
535 if (args < Argument::n_float_register_parameters) { |
|
536 regs[i].set2(FP_ArgReg[args++]->as_VMReg()); |
|
537 } else { |
|
538 regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
|
539 stk_args += 2; |
|
540 } |
|
541 break; |
|
542 default: |
|
543 ShouldNotReachHere(); |
|
544 break; |
|
545 } |
|
546 #ifdef aoqi_test |
|
547 tty->print_cr(" SharedRuntime::%s :%d, sig_bt[%d]: %d, reg[%d]:%d|%d, stk_args:%d", __func__, __LINE__, i, sig_bt[i], i, regs[i].first(), regs[i].second(), stk_args); |
|
548 #endif |
|
549 } |
|
550 |
|
551 return round_to(stk_args, 2); |
|
552 /* |
|
553 // Starting stack position for args on stack |
|
554 uint stack = 0; |
|
555 |
|
556 // Pass first five oop/int args in registers T0, A0 - A3. |
|
557 uint reg_arg0 = 9999; |
|
558 uint reg_arg1 = 9999; |
|
559 uint reg_arg2 = 9999; |
|
560 uint reg_arg3 = 9999; |
|
561 uint reg_arg4 = 9999; |
|
562 |
|
563 |
|
564 // Pass doubles & longs &float ligned on the stack. First count stack slots for doubles |
|
565 int i; |
|
566 for( i = 0; i < total_args_passed; i++) { |
|
567 if( sig_bt[i] == T_DOUBLE || sig_bt[i] == T_LONG ) { |
|
568 stack += 2; |
|
569 } |
|
570 } |
|
571 int dstack = 0; // Separate counter for placing doubles |
|
572 for( i = 0; i < total_args_passed; i++) { |
|
573 // From the type and the argument number (count) compute the location |
|
574 switch( sig_bt[i] ) { |
|
575 case T_SHORT: |
|
576 case T_CHAR: |
|
577 case T_BYTE: |
|
578 case T_BOOLEAN: |
|
579 case T_INT: |
|
580 case T_ARRAY: |
|
581 case T_OBJECT: |
|
582 case T_ADDRESS: |
|
583 if( reg_arg0 == 9999 ) { |
|
584 reg_arg0 = i; |
|
585 regs[i].set1(T0->as_VMReg()); |
|
586 } else if( reg_arg1 == 9999 ) { |
|
587 reg_arg1 = i; |
|
588 regs[i].set1(A0->as_VMReg()); |
|
589 } else if( reg_arg2 == 9999 ) { |
|
590 reg_arg2 = i; |
|
591 regs[i].set1(A1->as_VMReg()); |
|
592 }else if( reg_arg3 == 9999 ) { |
|
593 reg_arg3 = i; |
|
594 regs[i].set1(A2->as_VMReg()); |
|
595 }else if( reg_arg4 == 9999 ) { |
|
596 reg_arg4 = i; |
|
597 regs[i].set1(A3->as_VMReg()); |
|
598 } else { |
|
599 regs[i].set1(VMRegImpl::stack2reg(stack++)); |
|
600 } |
|
601 break; |
|
602 case T_FLOAT: |
|
603 regs[i].set1(VMRegImpl::stack2reg(stack++)); |
|
604 break; |
|
605 case T_LONG: |
|
606 assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
|
607 regs[i].set2(VMRegImpl::stack2reg(dstack)); |
|
608 dstack += 2; |
|
609 break; |
|
610 case T_DOUBLE: |
|
611 assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
|
612 regs[i].set2(VMRegImpl::stack2reg(dstack)); |
|
613 dstack += 2; |
|
614 break; |
|
615 case T_VOID: regs[i].set_bad(); break; |
|
616 break; |
|
617 default: |
|
618 ShouldNotReachHere(); |
|
619 break; |
|
620 } |
|
621 } |
|
622 // return value can be odd number of VMRegImpl stack slots make multiple of 2 |
|
623 return round_to(stack, 2); |
|
624 */ |
|
625 } |
|
626 |
|
627 // Helper class mostly to avoid passing masm everywhere, and handle store |
|
628 // displacement overflow logic for LP64 |
|
629 class AdapterGenerator { |
|
630 MacroAssembler *masm; |
|
631 #ifdef _LP64 |
|
632 Register Rdisp; |
|
633 void set_Rdisp(Register r) { Rdisp = r; } |
|
634 #endif // _LP64 |
|
635 |
|
636 void patch_callers_callsite(); |
|
637 // void tag_c2i_arg(frame::Tag t, Register base, int st_off, Register scratch); |
|
638 |
|
639 // base+st_off points to top of argument |
|
640 int arg_offset(const int st_off) { return st_off; } |
|
641 int next_arg_offset(const int st_off) { |
|
642 return st_off - Interpreter::stackElementSize; |
|
643 } |
|
644 |
|
645 #ifdef _LP64 |
|
646 // On _LP64 argument slot values are loaded first into a register |
|
647 // because they might not fit into displacement. |
|
648 Register arg_slot(const int st_off); |
|
649 Register next_arg_slot(const int st_off); |
|
650 #else |
|
651 int arg_slot(const int st_off) { return arg_offset(st_off); } |
|
652 int next_arg_slot(const int st_off) { return next_arg_offset(st_off); } |
|
653 #endif // _LP64 |
|
654 |
|
655 // Stores long into offset pointed to by base |
|
656 void store_c2i_long(Register r, Register base, |
|
657 const int st_off, bool is_stack); |
|
658 void store_c2i_object(Register r, Register base, |
|
659 const int st_off); |
|
660 void store_c2i_int(Register r, Register base, |
|
661 const int st_off); |
|
662 void store_c2i_double(VMReg r_2, |
|
663 VMReg r_1, Register base, const int st_off); |
|
664 void store_c2i_float(FloatRegister f, Register base, |
|
665 const int st_off); |
|
666 |
|
667 public: |
|
668 //void tag_stack(const BasicType sig, int st_off); |
|
669 void gen_c2i_adapter(int total_args_passed, |
|
670 // VMReg max_arg, |
|
671 int comp_args_on_stack, // VMRegStackSlots |
|
672 const BasicType *sig_bt, |
|
673 const VMRegPair *regs, |
|
674 Label& skip_fixup); |
|
675 void gen_i2c_adapter(int total_args_passed, |
|
676 // VMReg max_arg, |
|
677 int comp_args_on_stack, // VMRegStackSlots |
|
678 const BasicType *sig_bt, |
|
679 const VMRegPair *regs); |
|
680 |
|
681 AdapterGenerator(MacroAssembler *_masm) : masm(_masm) {} |
|
682 }; |
|
683 |
|
684 |
|
685 // Patch the callers callsite with entry to compiled code if it exists. |
|
686 void AdapterGenerator::patch_callers_callsite() { |
|
687 Label L; |
|
688 //FIXME , what is stored in eax? |
|
689 //__ verify_oop(ebx); |
|
690 __ verify_oop(Rmethod); |
|
691 // __ cmpl(Address(ebx, in_bytes(Method::code_offset())), NULL_WORD); |
|
692 __ ld_ptr(AT, Rmethod, in_bytes(Method::code_offset())); |
|
693 //__ jcc(Assembler::equal, L); |
|
694 __ beq(AT,R0,L); |
|
695 __ delayed()->nop(); |
|
696 // Schedule the branch target address early. |
|
697 // Call into the VM to patch the caller, then jump to compiled callee |
|
698 // eax isn't live so capture return address while we easily can |
|
699 // __ movl(eax, Address(esp, 0)); |
|
700 // __ lw(T5,SP,0); |
|
701 __ move(V0, RA); |
|
702 |
|
703 __ pushad(); |
|
704 //jerome_for_debug |
|
705 // __ pushad(); |
|
706 // __ pushfd(); |
|
707 #ifdef COMPILER2 |
|
708 // C2 may leave the stack dirty if not in SSE2+ mode |
|
709 __ empty_FPU_stack(); |
|
710 #endif /* COMPILER2 */ |
|
711 |
|
712 // VM needs caller's callsite |
|
713 // __ pushl(eax); |
|
714 |
|
715 // VM needs target method |
|
716 // __ pushl(ebx); |
|
717 // __ push(Rmethod); |
|
718 // __ verify_oop(ebx); |
|
719 |
|
720 __ move(A0, Rmethod); |
|
721 __ move(A1, V0); |
|
722 // __ addi(SP, SP, -8); |
|
723 //we should preserve the return address |
|
724 __ verify_oop(Rmethod); |
|
725 __ move(S0, SP); |
|
726 __ move(AT, -(StackAlignmentInBytes)); // align the stack |
|
727 __ andr(SP, SP, AT); |
|
728 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite), |
|
729 relocInfo::runtime_call_type); |
|
730 //__ addl(esp, 2*wordSize); |
|
731 |
|
732 __ delayed()->nop(); |
|
733 // __ addi(SP, SP, 8); |
|
734 // __ popfd(); |
|
735 __ move(SP, S0); |
|
736 __ popad(); |
|
737 __ bind(L); |
|
738 } |
|
739 /* |
|
740 void AdapterGenerator::tag_c2i_arg(frame::Tag t, Register base, int st_off, |
|
741 Register scratch) { |
|
742 Unimplemented(); |
|
743 }*/ |
|
744 |
|
745 #ifdef _LP64 |
|
746 Register AdapterGenerator::arg_slot(const int st_off) { |
|
747 Unimplemented(); |
|
748 } |
|
749 |
|
750 Register AdapterGenerator::next_arg_slot(const int st_off){ |
|
751 Unimplemented(); |
|
752 } |
|
753 #endif // _LP64 |
|
754 |
|
755 // Stores long into offset pointed to by base |
|
756 void AdapterGenerator::store_c2i_long(Register r, Register base, |
|
757 const int st_off, bool is_stack) { |
|
758 Unimplemented(); |
|
759 } |
|
760 |
|
761 void AdapterGenerator::store_c2i_object(Register r, Register base, |
|
762 const int st_off) { |
|
763 Unimplemented(); |
|
764 } |
|
765 |
|
766 void AdapterGenerator::store_c2i_int(Register r, Register base, |
|
767 const int st_off) { |
|
768 Unimplemented(); |
|
769 } |
|
770 |
|
771 // Stores into offset pointed to by base |
|
772 void AdapterGenerator::store_c2i_double(VMReg r_2, |
|
773 VMReg r_1, Register base, const int st_off) { |
|
774 Unimplemented(); |
|
775 } |
|
776 |
|
777 void AdapterGenerator::store_c2i_float(FloatRegister f, Register base, |
|
778 const int st_off) { |
|
779 Unimplemented(); |
|
780 } |
|
781 /* |
|
782 void AdapterGenerator::tag_stack(const BasicType sig, int st_off) { |
|
783 if (TaggedStackInterpreter) { |
|
784 int tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(0); |
|
785 if (sig == T_OBJECT || sig == T_ARRAY) { |
|
786 // __ movl(Address(esp, tag_offset), frame::TagReference); |
|
787 // __ addi(AT,R0, frame::TagReference); |
|
788 |
|
789 __ move(AT, frame::TagReference); |
|
790 __ sw (AT, SP, tag_offset); |
|
791 } else if (sig == T_LONG || sig == T_DOUBLE) { |
|
792 int next_tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(1); |
|
793 // __ movl(Address(esp, next_tag_offset), frame::TagValue); |
|
794 // __ addi(AT,R0, frame::TagValue); |
|
795 __ move(AT, frame::TagValue); |
|
796 __ sw (AT, SP, next_tag_offset); |
|
797 //__ movl(Address(esp, tag_offset), frame::TagValue); |
|
798 // __ addi(AT,R0, frame::TagValue); |
|
799 __ move(AT, frame::TagValue); |
|
800 __ sw (AT, SP, tag_offset); |
|
801 |
|
802 } else { |
|
803 // __ movl(Address(esp, tag_offset), frame::TagValue); |
|
804 //__ addi(AT,R0, frame::TagValue); |
|
805 __ move(AT, frame::TagValue); |
|
806 __ sw (AT, SP, tag_offset); |
|
807 |
|
808 } |
|
809 } |
|
810 }*/ |
|
811 |
|
812 void AdapterGenerator::gen_c2i_adapter( |
|
813 int total_args_passed, |
|
814 // VMReg max_arg, |
|
815 int comp_args_on_stack, // VMRegStackSlots |
|
816 const BasicType *sig_bt, |
|
817 const VMRegPair *regs, |
|
818 Label& skip_fixup) { |
|
819 |
|
820 // Before we get into the guts of the C2I adapter, see if we should be here |
|
821 // at all. We've come from compiled code and are attempting to jump to the |
|
822 // interpreter, which means the caller made a static call to get here |
|
823 // (vcalls always get a compiled target if there is one). Check for a |
|
824 // compiled target. If there is one, we need to patch the caller's call. |
|
825 // However we will run interpreted if we come thru here. The next pass |
|
826 // thru the call site will run compiled. If we ran compiled here then |
|
827 // we can (theorectically) do endless i2c->c2i->i2c transitions during |
|
828 // deopt/uncommon trap cycles. If we always go interpreted here then |
|
829 // we can have at most one and don't need to play any tricks to keep |
|
830 // from endlessly growing the stack. |
|
831 // |
|
832 // Actually if we detected that we had an i2c->c2i transition here we |
|
833 // ought to be able to reset the world back to the state of the interpreted |
|
834 // call and not bother building another interpreter arg area. We don't |
|
835 // do that at this point. |
|
836 |
|
837 patch_callers_callsite(); |
|
838 |
|
839 __ bind(skip_fixup); |
|
840 |
|
841 #ifdef COMPILER2 |
|
842 __ empty_FPU_stack(); |
|
843 #endif /* COMPILER2 */ |
|
844 //this is for native ? |
|
845 // Since all args are passed on the stack, total_args_passed * interpreter_ |
|
846 // stack_element_size is the |
|
847 // space we need. |
|
848 int extraspace = total_args_passed * Interpreter::stackElementSize; |
|
849 |
|
850 // stack is aligned, keep it that way |
|
851 extraspace = round_to(extraspace, 2*wordSize); |
|
852 |
|
853 // Get return address |
|
854 // __ popl(eax); |
|
855 //__ pop(T4); |
|
856 __ move(V0, RA); |
|
857 // set senderSP value |
|
858 // __ movl(esi, esp); |
|
859 //refer to interpreter_mips.cpp:generate_asm_entry |
|
860 __ move(Rsender, SP); |
|
861 //__ subl(esp, extraspace); |
|
862 __ addi(SP, SP, -extraspace); |
|
863 |
|
864 // Now write the args into the outgoing interpreter space |
|
865 for (int i = 0; i < total_args_passed; i++) { |
|
866 if (sig_bt[i] == T_VOID) { |
|
867 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), |
|
868 "missing half"); |
|
869 continue; |
|
870 } |
|
871 |
|
872 // st_off points to lowest address on stack. |
|
873 int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize; |
|
874 #ifdef aoqi_test |
|
875 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d, st_off:%d", __func__, __LINE__, i, sig_bt[i], total_args_passed, st_off); |
|
876 #endif |
|
877 // Say 4 args: |
|
878 // i st_off |
|
879 // 0 12 T_LONG |
|
880 // 1 8 T_VOID |
|
881 // 2 4 T_OBJECT |
|
882 // 3 0 T_BOOL |
|
883 VMReg r_1 = regs[i].first(); |
|
884 VMReg r_2 = regs[i].second(); |
|
885 if (!r_1->is_valid()) { |
|
886 assert(!r_2->is_valid(), ""); |
|
887 continue; |
|
888 } |
|
889 |
|
890 if (r_1->is_stack()) { |
|
891 // memory to memory use fpu stack top |
|
892 int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; |
|
893 #ifdef aoqi_test |
|
894 tty->print_cr(" AdapterGenerator::%s :%d, r_1->is_stack, ld_off:%x", __func__, __LINE__, ld_off); |
|
895 #endif |
|
896 |
|
897 if (!r_2->is_valid()) { |
|
898 #ifdef aoqi_test |
|
899 tty->print_cr(" AdapterGenerator::%s :%d, !r_2->is_valid, ld_off:%x", __func__, __LINE__, ld_off); |
|
900 #endif |
|
901 __ ld_ptr(AT, SP, ld_off); |
|
902 __ st_ptr(AT, SP, st_off); |
|
903 //tag_stack(sig_bt[i], st_off); |
|
904 } else { |
|
905 #ifdef aoqi_test |
|
906 tty->print_cr(" AdapterGenerator::%s :%d, r_2->is_valid, ld_off:%x", __func__, __LINE__, ld_off); |
|
907 #endif |
|
908 |
|
909 // ld_off == LSW, ld_off+VMRegImpl::stack_slot_size == MSW |
|
910 // st_off == MSW, st_off-wordSize == LSW |
|
911 |
|
912 int next_off = st_off - Interpreter::stackElementSize; |
|
913 /* |
|
914 __ lw(AT, SP, ld_off); |
|
915 __ sw(AT, SP, next_off); |
|
916 __ lw(AT, SP, ld_off + wordSize); |
|
917 __ sw(AT, SP, st_off); |
|
918 */ |
|
919 __ ld_ptr(AT, SP, ld_off); |
|
920 __ st_ptr(AT, SP, st_off); |
|
921 |
|
922 /* Ref to is_Register condition */ |
|
923 if(sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) |
|
924 __ st_ptr(AT,SP,st_off - 8); |
|
925 //tag_stack(sig_bt[i], next_off); |
|
926 } |
|
927 } else if (r_1->is_Register()) { |
|
928 Register r = r_1->as_Register(); |
|
929 if (!r_2->is_valid()) { |
|
930 #ifdef aoqi_test |
|
931 tty->print_cr(" AdapterGenerator::%s :%d, r_1->is_Register, !r_2->is_valid, st_off: %lx", __func__, __LINE__, st_off); |
|
932 #endif |
|
933 // __ movl(Address(esp, st_off), r); |
|
934 __ sd(r,SP, st_off); //aoqi_test FIXME |
|
935 //tag_stack(sig_bt[i], st_off); |
|
936 } else { |
|
937 #ifdef aoqi_test |
|
938 tty->print_cr(" AdapterGenerator::%s :%d, r_1->is_Register, r_2->is_valid, st_off: %lx", __func__, __LINE__, st_off); |
|
939 #endif |
|
940 //FIXME, mips will not enter here |
|
941 // long/double in gpr |
|
942 __ sd(r,SP, st_off); //aoqi_test FIXME |
|
943 /* Jin: In [java/util/zip/ZipFile.java] |
|
944 |
|
945 private static native long open(String name, int mode, long lastModified); |
|
946 private static native int getTotal(long jzfile); |
|
947 * |
|
948 * We need to transfer T_LONG paramenters from a compiled method to a native method. |
|
949 * It's a complex process: |
|
950 * |
|
951 * Caller -> lir_static_call -> gen_resolve_stub |
|
952 -> -- resolve_static_call_C |
|
953 `- gen_c2i_adapter() [*] |
|
954 | |
|
955 `- AdapterHandlerLibrary::get_create_apapter_index |
|
956 -> generate_native_entry |
|
957 -> InterpreterRuntime::SignatureHandlerGenerator::pass_long [**] |
|
958 |
|
959 * In [**], T_Long parameter is stored in stack as: |
|
960 |
|
961 (high) |
|
962 | | |
|
963 ----------- |
|
964 | 8 bytes | |
|
965 | (void) | |
|
966 ----------- |
|
967 | 8 bytes | |
|
968 | (long) | |
|
969 ----------- |
|
970 | | |
|
971 (low) |
|
972 * |
|
973 * However, the sequence is reversed here: |
|
974 * |
|
975 (high) |
|
976 | | |
|
977 ----------- |
|
978 | 8 bytes | |
|
979 | (long) | |
|
980 ----------- |
|
981 | 8 bytes | |
|
982 | (void) | |
|
983 ----------- |
|
984 | | |
|
985 (low) |
|
986 * |
|
987 * So I stored another 8 bytes in the T_VOID slot. It then can be accessed from generate_native_entry(). |
|
988 */ |
|
989 if (sig_bt[i] == T_LONG) |
|
990 __ sd(r,SP, st_off - 8); |
|
991 // ShouldNotReachHere(); |
|
992 // int next_off = st_off - Interpreter::stackElementSize; |
|
993 // __ sw(r_2->as_Register(),SP, st_off); |
|
994 // __ sw(r,SP, next_off); |
|
995 // tag_stack(masm, sig_bt[i], next_off); |
|
996 } |
|
997 } else if (r_1->is_FloatRegister()) { |
|
998 assert(sig_bt[i] == T_FLOAT || sig_bt[i] == T_DOUBLE, "Must be a float register"); |
|
999 |
|
1000 FloatRegister fr = r_1->as_FloatRegister(); |
|
1001 if (sig_bt[i] == T_FLOAT) |
|
1002 __ swc1(fr,SP, st_off); |
|
1003 else |
|
1004 { |
|
1005 __ sdc1(fr,SP, st_off); |
|
1006 __ sdc1(fr,SP, st_off - 8); /* T_DOUBLE needs two slots */ |
|
1007 } |
|
1008 } |
|
1009 } |
|
1010 |
|
1011 // Schedule the branch target address early. |
|
1012 __ ld_ptr(AT, Rmethod,in_bytes(Method::interpreter_entry_offset()) ); |
|
1013 // And repush original return address |
|
1014 __ move(RA, V0); |
|
1015 __ jr (AT); |
|
1016 __ delayed()->nop(); |
|
1017 } |
|
1018 |
|
1019 void AdapterGenerator::gen_i2c_adapter( |
|
1020 int total_args_passed, |
|
1021 // VMReg max_arg, |
|
1022 int comp_args_on_stack, // VMRegStackSlots |
|
1023 const BasicType *sig_bt, |
|
1024 const VMRegPair *regs) { |
|
1025 |
|
1026 // Generate an I2C adapter: adjust the I-frame to make space for the C-frame |
|
1027 // layout. Lesp was saved by the calling I-frame and will be restored on |
|
1028 // return. Meanwhile, outgoing arg space is all owned by the callee |
|
1029 // C-frame, so we can mangle it at will. After adjusting the frame size, |
|
1030 // hoist register arguments and repack other args according to the compiled |
|
1031 // code convention. Finally, end in a jump to the compiled code. The entry |
|
1032 // point address is the start of the buffer. |
|
1033 |
|
1034 // We will only enter here from an interpreted frame and never from after |
|
1035 // passing thru a c2i. Azul allowed this but we do not. If we lose the |
|
1036 // race and use a c2i we will remain interpreted for the race loser(s). |
|
1037 // This removes all sorts of headaches on the mips side and also eliminates |
|
1038 // the possibility of having c2i -> i2c -> c2i -> ... endless transitions. |
|
1039 |
|
1040 |
|
1041 __ move(T9, SP); |
|
1042 |
|
1043 // Cut-out for having no stack args. Since up to 2 int/oop args are passed |
|
1044 // in registers, we will occasionally have no stack args. |
|
1045 int comp_words_on_stack = 0; |
|
1046 if (comp_args_on_stack) { |
|
1047 // Sig words on the stack are greater-than VMRegImpl::stack0. Those in |
|
1048 // registers are below. By subtracting stack0, we either get a negative |
|
1049 // number (all values in registers) or the maximum stack slot accessed. |
|
1050 // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg); |
|
1051 // Convert 4-byte stack slots to words. |
|
1052 // did mips need round? FIXME aoqi |
|
1053 comp_words_on_stack = round_to(comp_args_on_stack*4, wordSize)>>LogBytesPerWord; |
|
1054 // Round up to miminum stack alignment, in wordSize |
|
1055 comp_words_on_stack = round_to(comp_words_on_stack, 2); |
|
1056 __ daddi(SP, SP, -comp_words_on_stack * wordSize); |
|
1057 } |
|
1058 |
|
1059 // Align the outgoing SP |
|
1060 __ move(AT, -(StackAlignmentInBytes)); |
|
1061 __ andr(SP, SP, AT); |
|
1062 // push the return address on the stack (note that pushing, rather |
|
1063 // than storing it, yields the correct frame alignment for the callee) |
|
1064 // Put saved SP in another register |
|
1065 // const Register saved_sp = eax; |
|
1066 const Register saved_sp = V0; |
|
1067 __ move(saved_sp, T9); |
|
1068 |
|
1069 |
|
1070 // Will jump to the compiled code just as if compiled code was doing it. |
|
1071 // Pre-load the register-jump target early, to schedule it better. |
|
1072 __ ld(T9, Rmethod, in_bytes(Method::from_compiled_offset())); |
|
1073 |
|
1074 // Now generate the shuffle code. Pick up all register args and move the |
|
1075 // rest through the floating point stack top. |
|
1076 for (int i = 0; i < total_args_passed; i++) { |
|
1077 if (sig_bt[i] == T_VOID) { |
|
1078 // Longs and doubles are passed in native word order, but misaligned |
|
1079 // in the 32-bit build. |
|
1080 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
|
1081 continue; |
|
1082 } |
|
1083 |
|
1084 // Pick up 0, 1 or 2 words from SP+offset. |
|
1085 |
|
1086 //FIXME. aoqi. just delete the assert |
|
1087 //assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), "scrambled load targets?"); |
|
1088 // Load in argument order going down. |
|
1089 int ld_off = (total_args_passed -1 - i)*Interpreter::stackElementSize; |
|
1090 // Point to interpreter value (vs. tag) |
|
1091 int next_off = ld_off - Interpreter::stackElementSize; |
|
1092 // |
|
1093 // |
|
1094 // |
|
1095 VMReg r_1 = regs[i].first(); |
|
1096 VMReg r_2 = regs[i].second(); |
|
1097 if (!r_1->is_valid()) { |
|
1098 assert(!r_2->is_valid(), ""); |
|
1099 continue; |
|
1100 } |
|
1101 #ifdef aoqi_test |
|
1102 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d, ld_off:%d, next_off: %d", __func__, __LINE__, i, sig_bt[i], total_args_passed, ld_off, next_off); |
|
1103 #endif |
|
1104 if (r_1->is_stack()) { |
|
1105 // Convert stack slot to an SP offset (+ wordSize to |
|
1106 // account for return address ) |
|
1107 //NOTICE HERE!!!! I sub a wordSize here |
|
1108 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size; |
|
1109 //+ wordSize; |
|
1110 |
|
1111 // We can use esi as a temp here because compiled code doesn't |
|
1112 // need esi as an input |
|
1113 // and if we end up going thru a c2i because of a miss a reasonable |
|
1114 // value of esi |
|
1115 // we be generated. |
|
1116 if (!r_2->is_valid()) { |
|
1117 #ifdef aoqi_test |
|
1118 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d r_1->is_stack() !r_2->is_valid(), st_off:%d", __func__, __LINE__, i, sig_bt[i], total_args_passed, st_off); |
|
1119 #endif |
|
1120 __ ld(AT, saved_sp, ld_off); |
|
1121 __ sd(AT, SP, st_off); |
|
1122 } else { |
|
1123 #ifdef aoqi_test |
|
1124 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d r_1->is_stack() r_2->is_valid(), st_off:%d", __func__, __LINE__, i, sig_bt[i], total_args_passed, st_off); |
|
1125 #endif |
|
1126 // Interpreter local[n] == MSW, local[n+1] == LSW however locals |
|
1127 // are accessed as negative so LSW is at LOW address |
|
1128 |
|
1129 // ld_off is MSW so get LSW |
|
1130 // st_off is LSW (i.e. reg.first()) |
|
1131 /* |
|
1132 __ ld(AT, saved_sp, next_off); |
|
1133 __ sd(AT, SP, st_off); |
|
1134 __ ld(AT, saved_sp, ld_off); |
|
1135 __ sd(AT, SP, st_off + wordSize); |
|
1136 */ |
|
1137 |
|
1138 /* 2012/4/9 Jin |
|
1139 * [./org/eclipse/swt/graphics/GC.java] |
|
1140 * void drawImageXRender(Image srcImage, int srcX, int srcY, int srcWidth, int srcHeight, |
|
1141 int destX, int destY, int destWidth, int destHeight, |
|
1142 boolean simple, |
|
1143 int imgWidth, int imgHeight, |
|
1144 long maskPixmap, <-- Pass T_LONG in stack |
|
1145 int maskType); |
|
1146 * Before this modification, Eclipse displays icons with solid black background. |
|
1147 */ |
|
1148 __ ld(AT, saved_sp, ld_off); |
|
1149 if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) |
|
1150 __ ld(AT, saved_sp, ld_off - 8); |
|
1151 __ sd(AT, SP, st_off); |
|
1152 //__ ld(AT, saved_sp, next_off); |
|
1153 //__ sd(AT, SP, st_off + wordSize); |
|
1154 } |
|
1155 } else if (r_1->is_Register()) { // Register argument |
|
1156 Register r = r_1->as_Register(); |
|
1157 // assert(r != eax, "must be different"); |
|
1158 if (r_2->is_valid()) { |
|
1159 #ifdef aoqi_test |
|
1160 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d r_1->is_Register() r_2->is_valid()", __func__, __LINE__, i, sig_bt[i], total_args_passed); |
|
1161 #endif |
|
1162 // assert(r_2->as_Register() != eax, "need another temporary register"); |
|
1163 // Remember r_1 is low address (and LSB on mips) |
|
1164 // So r_2 gets loaded from high address regardless of the platform |
|
1165 //aoqi |
|
1166 assert(r_2->as_Register() == r_1->as_Register(), ""); |
|
1167 //__ ld(r_2->as_Register(), saved_sp, ld_off); |
|
1168 //__ ld(r, saved_sp, next_off); |
|
1169 __ ld(r, saved_sp, ld_off); |
|
1170 |
|
1171 /* Jin: |
|
1172 * |
|
1173 * For T_LONG type, the real layout is as below: |
|
1174 |
|
1175 (high) |
|
1176 | | |
|
1177 ----------- |
|
1178 | 8 bytes | |
|
1179 | (void) | |
|
1180 ----------- |
|
1181 | 8 bytes | |
|
1182 | (long) | |
|
1183 ----------- |
|
1184 | | |
|
1185 (low) |
|
1186 * |
|
1187 * We should load the low-8 bytes. |
|
1188 */ |
|
1189 if (sig_bt[i] == T_LONG) |
|
1190 __ ld(r, saved_sp, ld_off - 8); |
|
1191 } else { |
|
1192 #ifdef aoqi_test |
|
1193 tty->print_cr(" AdapterGenerator::%s :%d, sig_bt[%d]:%d, total_args_passed:%d r_1->is_Register() !r_2->is_valid()", __func__, __LINE__, i, sig_bt[i], total_args_passed); |
|
1194 #endif |
|
1195 __ lw(r, saved_sp, ld_off); |
|
1196 } |
|
1197 } else if (r_1->is_FloatRegister()) { // Float Register |
|
1198 assert(sig_bt[i] == T_FLOAT || sig_bt[i] == T_DOUBLE, "Must be a float register"); |
|
1199 |
|
1200 FloatRegister fr = r_1->as_FloatRegister(); |
|
1201 if (sig_bt[i] == T_FLOAT) |
|
1202 __ lwc1(fr, saved_sp, ld_off); |
|
1203 else |
|
1204 { |
|
1205 __ ldc1(fr, saved_sp, ld_off); |
|
1206 __ ldc1(fr, saved_sp, ld_off - 8); |
|
1207 } |
|
1208 } |
|
1209 } |
|
1210 |
|
1211 // 6243940 We might end up in handle_wrong_method if |
|
1212 // the callee is deoptimized as we race thru here. If that |
|
1213 // happens we don't want to take a safepoint because the |
|
1214 // caller frame will look interpreted and arguments are now |
|
1215 // "compiled" so it is much better to make this transition |
|
1216 // invisible to the stack walking code. Unfortunately if |
|
1217 // we try and find the callee by normal means a safepoint |
|
1218 // is possible. So we stash the desired callee in the thread |
|
1219 // and the vm will find there should this case occur. |
|
1220 __ get_thread(T8); |
|
1221 __ sd(Rmethod, T8, in_bytes(JavaThread::callee_target_offset())); |
|
1222 |
|
1223 // move methodOop to eax in case we end up in an c2i adapter. |
|
1224 // the c2i adapters expect methodOop in eax (c2) because c2's |
|
1225 // resolve stubs return the result (the method) in eax. |
|
1226 // I'd love to fix this. |
|
1227 __ move(V0, Rmethod); |
|
1228 __ jr(T9); |
|
1229 __ delayed()->nop(); |
|
1230 } |
|
1231 |
|
1232 // --------------------------------------------------------------- |
|
1233 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, |
|
1234 int total_args_passed, |
|
1235 // VMReg max_arg, |
|
1236 int comp_args_on_stack, // VMRegStackSlots |
|
1237 const BasicType *sig_bt, |
|
1238 const VMRegPair *regs, |
|
1239 AdapterFingerPrint* fingerprint) { |
|
1240 address i2c_entry = __ pc(); |
|
1241 |
|
1242 AdapterGenerator agen(masm); |
|
1243 |
|
1244 agen.gen_i2c_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs); |
|
1245 |
|
1246 |
|
1247 // ------------------------------------------------------------------------- |
|
1248 // Generate a C2I adapter. On entry we know G5 holds the methodOop. The |
|
1249 // args start out packed in the compiled layout. They need to be unpacked |
|
1250 // into the interpreter layout. This will almost always require some stack |
|
1251 // space. We grow the current (compiled) stack, then repack the args. We |
|
1252 // finally end in a jump to the generic interpreter entry point. On exit |
|
1253 // from the interpreter, the interpreter will restore our SP (lest the |
|
1254 // compiled code, which relys solely on SP and not FP, get sick). |
|
1255 |
|
1256 address c2i_unverified_entry = __ pc(); |
|
1257 Label skip_fixup; |
|
1258 { |
|
1259 Register holder = T1; |
|
1260 Register receiver = T0; |
|
1261 Register temp = T8; |
|
1262 address ic_miss = SharedRuntime::get_ic_miss_stub(); |
|
1263 |
|
1264 Label missed; |
|
1265 |
|
1266 __ verify_oop(holder); |
|
1267 // __ movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes())); |
|
1268 //__ ld_ptr(temp, receiver, oopDesc::klass_offset_in_bytes()); |
|
1269 //add for compressedoops |
|
1270 __ load_klass(temp, receiver); |
|
1271 __ verify_oop(temp); |
|
1272 |
|
1273 // __ cmpl(temp, Address(holder, CompiledICHolder::holder_klass_offset())); |
|
1274 __ ld_ptr(AT, holder, CompiledICHolder::holder_klass_offset()); |
|
1275 //__ movl(ebx, Address(holder, CompiledICHolder::holder_method_offset())); |
|
1276 __ ld_ptr(Rmethod, holder, CompiledICHolder::holder_method_offset()); |
|
1277 //__ jcc(Assembler::notEqual, missed); |
|
1278 __ bne(AT, temp, missed); |
|
1279 __ delayed()->nop(); |
|
1280 // Method might have been compiled since the call site was patched to |
|
1281 // interpreted if that is the case treat it as a miss so we can get |
|
1282 // the call site corrected. |
|
1283 //__ cmpl(Address(ebx, in_bytes(Method::code_offset())), NULL_WORD); |
|
1284 //__ jcc(Assembler::equal, skip_fixup); |
|
1285 __ ld_ptr(AT, Rmethod, in_bytes(Method::code_offset())); |
|
1286 __ beq(AT, R0, skip_fixup); |
|
1287 __ delayed()->nop(); |
|
1288 __ bind(missed); |
|
1289 // __ move(AT, (int)&jerome7); |
|
1290 // __ sw(RA, AT, 0); |
|
1291 |
|
1292 __ jmp(ic_miss, relocInfo::runtime_call_type); |
|
1293 __ delayed()->nop(); |
|
1294 } |
|
1295 |
|
1296 address c2i_entry = __ pc(); |
|
1297 |
|
1298 agen.gen_c2i_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); |
|
1299 |
|
1300 __ flush(); |
|
1301 return AdapterHandlerLibrary::new_entry(fingerprint,i2c_entry, c2i_entry, c2i_unverified_entry); |
|
1302 |
|
1303 } |
|
1304 /* |
|
1305 // Helper function for native calling conventions |
|
1306 static VMReg int_stk_helper( int i ) { |
|
1307 // Bias any stack based VMReg we get by ignoring the window area |
|
1308 // but not the register parameter save area. |
|
1309 // |
|
1310 // This is strange for the following reasons. We'd normally expect |
|
1311 // the calling convention to return an VMReg for a stack slot |
|
1312 // completely ignoring any abi reserved area. C2 thinks of that |
|
1313 // abi area as only out_preserve_stack_slots. This does not include |
|
1314 // the area allocated by the C abi to store down integer arguments |
|
1315 // because the java calling convention does not use it. So |
|
1316 // since c2 assumes that there are only out_preserve_stack_slots |
|
1317 // to bias the optoregs (which impacts VMRegs) when actually referencing any actual stack |
|
1318 // location the c calling convention must add in this bias amount |
|
1319 // to make up for the fact that the out_preserve_stack_slots is |
|
1320 // insufficient for C calls. What a mess. I sure hope those 6 |
|
1321 // stack words were worth it on every java call! |
|
1322 |
|
1323 // Another way of cleaning this up would be for out_preserve_stack_slots |
|
1324 // to take a parameter to say whether it was C or java calling conventions. |
|
1325 // Then things might look a little better (but not much). |
|
1326 |
|
1327 int mem_parm_offset = i - SPARC_ARGS_IN_REGS_NUM; |
|
1328 if( mem_parm_offset < 0 ) { |
|
1329 return as_oRegister(i)->as_VMReg(); |
|
1330 } else { |
|
1331 int actual_offset = (mem_parm_offset + frame::memory_parameter_word_sp_offset) * VMRegImpl::slots_per_word; |
|
1332 // Now return a biased offset that will be correct when out_preserve_slots is added back in |
|
1333 return VMRegImpl::stack2reg(actual_offset - SharedRuntime::out_preserve_stack_slots()); |
|
1334 } |
|
1335 } |
|
1336 */ |
|
1337 |
|
1338 |
|
1339 int SharedRuntime::c_calling_convention(const BasicType *sig_bt, |
|
1340 VMRegPair *regs, |
|
1341 VMRegPair *regs2, |
|
1342 int total_args_passed) { |
|
1343 assert(regs2 == NULL, "not needed on MIPS"); |
|
1344 #ifdef aoqi_test |
|
1345 tty->print_cr(" SharedRuntime::%s :%d total_args_passed:%d", __func__, __LINE__, total_args_passed); |
|
1346 #endif |
|
1347 // Return the number of VMReg stack_slots needed for the args. |
|
1348 // This value does not include an abi space (like register window |
|
1349 // save area). |
|
1350 |
|
1351 // The native convention is V8 if !LP64 |
|
1352 // The LP64 convention is the V9 convention which is slightly more sane. |
|
1353 |
|
1354 // We return the amount of VMReg stack slots we need to reserve for all |
|
1355 // the arguments NOT counting out_preserve_stack_slots. Since we always |
|
1356 // have space for storing at least 6 registers to memory we start with that. |
|
1357 // See int_stk_helper for a further discussion. |
|
1358 // We return the amount of VMRegImpl stack slots we need to reserve for all |
|
1359 // the arguments NOT counting out_preserve_stack_slots. |
|
1360 static const Register INT_ArgReg[Argument::n_register_parameters] = { |
|
1361 A0, A1, A2, A3, A4, A5, A6, A7 |
|
1362 }; |
|
1363 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters] = { |
|
1364 F12, F13, F14, F15, F16, F17, F18, F19 |
|
1365 }; |
|
1366 uint args = 0; |
|
1367 uint stk_args = 0; // inc by 2 each time |
|
1368 |
|
1369 /* Example: |
|
1370 --- n java.lang.UNIXProcess::forkAndExec |
|
1371 private native int forkAndExec(byte[] prog, |
|
1372 byte[] argBlock, int argc, |
|
1373 byte[] envBlock, int envc, |
|
1374 byte[] dir, |
|
1375 boolean redirectErrorStream, |
|
1376 FileDescriptor stdin_fd, |
|
1377 FileDescriptor stdout_fd, |
|
1378 FileDescriptor stderr_fd) |
|
1379 JNIEXPORT jint JNICALL |
|
1380 Java_java_lang_UNIXProcess_forkAndExec(JNIEnv *env, |
|
1381 jobject process, |
|
1382 jbyteArray prog, |
|
1383 jbyteArray argBlock, jint argc, |
|
1384 jbyteArray envBlock, jint envc, |
|
1385 jbyteArray dir, |
|
1386 jboolean redirectErrorStream, |
|
1387 jobject stdin_fd, |
|
1388 jobject stdout_fd, |
|
1389 jobject stderr_fd) |
|
1390 |
|
1391 ::c_calling_convention |
|
1392 0: // env <-- a0 |
|
1393 1: L // klass/obj <-- t0 => a1 |
|
1394 2: [ // prog[] <-- a0 => a2 |
|
1395 3: [ // argBlock[] <-- a1 => a3 |
|
1396 4: I // argc |
|
1397 5: [ // envBlock[] <-- a3 => a5 |
|
1398 6: I // envc |
|
1399 7: [ // dir[] <-- a5 => a7 |
|
1400 8: Z // redirectErrorStream a6 => sp[0] |
|
1401 9: L // stdin a7 => sp[8] |
|
1402 10: L // stdout fp[16] => sp[16] |
|
1403 11: L // stderr fp[24] => sp[24] |
|
1404 */ |
|
1405 for (int i = 0; i < total_args_passed; i++) { |
|
1406 switch (sig_bt[i]) { |
|
1407 case T_VOID: // Halves of longs and doubles |
|
1408 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
|
1409 regs[i].set_bad(); |
|
1410 break; |
|
1411 case T_BOOLEAN: |
|
1412 case T_CHAR: |
|
1413 case T_BYTE: |
|
1414 case T_SHORT: |
|
1415 case T_INT: |
|
1416 if (args < Argument::n_register_parameters) { |
|
1417 regs[i].set1(INT_ArgReg[args++]->as_VMReg()); |
|
1418 } else { |
|
1419 regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
|
1420 stk_args += 2; |
|
1421 } |
|
1422 break; |
|
1423 case T_LONG: |
|
1424 assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
|
1425 // fall through |
|
1426 case T_OBJECT: |
|
1427 case T_ARRAY: |
|
1428 case T_ADDRESS: |
|
1429 case T_METADATA: |
|
1430 if (args < Argument::n_register_parameters) { |
|
1431 regs[i].set2(INT_ArgReg[args++]->as_VMReg()); |
|
1432 } else { |
|
1433 regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
|
1434 stk_args += 2; |
|
1435 } |
|
1436 break; |
|
1437 case T_FLOAT: |
|
1438 if (args < Argument::n_float_register_parameters) { |
|
1439 regs[i].set1(FP_ArgReg[args++]->as_VMReg()); |
|
1440 } else { |
|
1441 regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
|
1442 stk_args += 2; |
|
1443 } |
|
1444 break; |
|
1445 case T_DOUBLE: |
|
1446 assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
|
1447 if (args < Argument::n_float_register_parameters) { |
|
1448 regs[i].set2(FP_ArgReg[args++]->as_VMReg()); |
|
1449 } else { |
|
1450 regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
|
1451 stk_args += 2; |
|
1452 } |
|
1453 break; |
|
1454 default: |
|
1455 ShouldNotReachHere(); |
|
1456 break; |
|
1457 } |
|
1458 } |
|
1459 |
|
1460 return round_to(stk_args, 2); |
|
1461 } |
|
1462 /* |
|
1463 int SharedRuntime::c_calling_convention_jni(const BasicType *sig_bt, |
|
1464 VMRegPair *regs, |
|
1465 int total_args_passed) { |
|
1466 // We return the amount of VMRegImpl stack slots we need to reserve for all |
|
1467 // the arguments NOT counting out_preserve_stack_slots. |
|
1468 bool unalign = 0; |
|
1469 uint stack = 0; // All arguments on stack |
|
1470 #ifdef aoqi_test |
|
1471 tty->print_cr(" SharedRuntime::%s :%d total_args_passed:%d", __func__, __LINE__, total_args_passed); |
|
1472 #endif |
|
1473 |
|
1474 for( int i = 0; i < total_args_passed; i++) { |
|
1475 // From the type and the argument number (count) compute the location |
|
1476 switch( sig_bt[i] ) { |
|
1477 case T_BOOLEAN: |
|
1478 case T_CHAR: |
|
1479 case T_FLOAT: |
|
1480 case T_BYTE: |
|
1481 case T_SHORT: |
|
1482 case T_INT: |
|
1483 case T_OBJECT: |
|
1484 case T_ARRAY: |
|
1485 case T_ADDRESS: |
|
1486 regs[i].set1(VMRegImpl::stack2reg(stack++)); |
|
1487 unalign = !unalign; |
|
1488 break; |
|
1489 case T_LONG: |
|
1490 case T_DOUBLE: // The stack numbering is reversed from Java |
|
1491 // Since C arguments do not get reversed, the ordering for |
|
1492 // doubles on the stack must be opposite the Java convention |
|
1493 assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
|
1494 if(unalign){ |
|
1495 stack += 1; |
|
1496 unalign = ! unalign; |
|
1497 } |
|
1498 regs[i].set2(VMRegImpl::stack2reg(stack)); |
|
1499 stack += 2; |
|
1500 break; |
|
1501 case T_VOID: regs[i].set_bad(); break; |
|
1502 default: |
|
1503 ShouldNotReachHere(); |
|
1504 break; |
|
1505 } |
|
1506 } |
|
1507 return stack; |
|
1508 } |
|
1509 */ |
|
1510 |
|
1511 // --------------------------------------------------------------------------- |
|
1512 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
|
1513 // We always ignore the frame_slots arg and just use the space just below frame pointer |
|
1514 // which by this time is free to use |
|
1515 switch (ret_type) { |
|
1516 case T_FLOAT: |
|
1517 __ swc1(FSF, FP, -wordSize); |
|
1518 break; |
|
1519 case T_DOUBLE: |
|
1520 __ sdc1(FSF, FP, -wordSize ); |
|
1521 break; |
|
1522 case T_VOID: break; |
|
1523 case T_LONG: |
|
1524 __ sd(V0, FP, -wordSize); |
|
1525 break; |
|
1526 case T_OBJECT: |
|
1527 case T_ARRAY: |
|
1528 __ sd(V0, FP, -wordSize); |
|
1529 break; |
|
1530 default: { |
|
1531 __ sw(V0, FP, -wordSize); |
|
1532 } |
|
1533 } |
|
1534 } |
|
1535 |
|
1536 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
|
1537 // We always ignore the frame_slots arg and just use the space just below frame pointer |
|
1538 // which by this time is free to use |
|
1539 switch (ret_type) { |
|
1540 case T_FLOAT: |
|
1541 __ lwc1(FSF, FP, -wordSize); |
|
1542 break; |
|
1543 case T_DOUBLE: |
|
1544 __ ldc1(FSF, FP, -wordSize ); |
|
1545 break; |
|
1546 case T_LONG: |
|
1547 __ ld(V0, FP, -wordSize); |
|
1548 break; |
|
1549 case T_VOID: break; |
|
1550 case T_OBJECT: |
|
1551 case T_ARRAY: |
|
1552 __ ld(V0, FP, -wordSize); |
|
1553 break; |
|
1554 default: { |
|
1555 __ lw(V0, FP, -wordSize); |
|
1556 } |
|
1557 } |
|
1558 } |
|
1559 |
|
1560 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
|
1561 for ( int i = first_arg ; i < arg_count ; i++ ) { |
|
1562 if (args[i].first()->is_Register()) { |
|
1563 __ push(args[i].first()->as_Register()); |
|
1564 } else if (args[i].first()->is_FloatRegister()) { |
|
1565 __ push(args[i].first()->as_FloatRegister()); |
|
1566 } |
|
1567 } |
|
1568 } |
|
1569 |
|
1570 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
|
1571 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) { |
|
1572 if (args[i].first()->is_Register()) { |
|
1573 __ pop(args[i].first()->as_Register()); |
|
1574 } else if (args[i].first()->is_FloatRegister()) { |
|
1575 __ pop(args[i].first()->as_FloatRegister()); |
|
1576 } |
|
1577 } |
|
1578 } |
|
1579 |
|
1580 // A simple move of integer like type |
|
1581 static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1582 if (src.first()->is_stack()) { |
|
1583 if (dst.first()->is_stack()) { |
|
1584 // stack to stack |
|
1585 __ lw(AT, FP, reg2offset_in(src.first())); |
|
1586 __ sd(AT,SP, reg2offset_out(dst.first())); |
|
1587 } else { |
|
1588 // stack to reg |
|
1589 //__ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); |
|
1590 __ lw(dst.first()->as_Register(), FP, reg2offset_in(src.first())); |
|
1591 } |
|
1592 } else if (dst.first()->is_stack()) { |
|
1593 // reg to stack |
|
1594 __ sd(src.first()->as_Register(), SP, reg2offset_out(dst.first())); |
|
1595 } else { |
|
1596 //__ mov(src.first()->as_Register(), dst.first()->as_Register()); |
|
1597 if (dst.first() != src.first()){ |
|
1598 __ move(dst.first()->as_Register(), src.first()->as_Register()); // fujie error:dst.first() |
|
1599 } |
|
1600 } |
|
1601 } |
|
1602 /* |
|
1603 // On 64 bit we will store integer like items to the stack as |
|
1604 // 64 bits items (sparc abi) even though java would only store |
|
1605 // 32bits for a parameter. On 32bit it will simply be 32 bits |
|
1606 // So this routine will do 32->32 on 32bit and 32->64 on 64bit |
|
1607 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1608 if (src.first()->is_stack()) { |
|
1609 if (dst.first()->is_stack()) { |
|
1610 // stack to stack |
|
1611 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); |
|
1612 __ st_ptr(L5, SP, reg2offset(dst.first()) + STACK_BIAS); |
|
1613 } else { |
|
1614 // stack to reg |
|
1615 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); |
|
1616 } |
|
1617 } else if (dst.first()->is_stack()) { |
|
1618 // reg to stack |
|
1619 __ st_ptr(src.first()->as_Register(), SP, reg2offset(dst.first()) + STACK_BIAS); |
|
1620 } else { |
|
1621 __ mov(src.first()->as_Register(), dst.first()->as_Register()); |
|
1622 } |
|
1623 } |
|
1624 */ |
|
1625 |
|
1626 // An oop arg. Must pass a handle not the oop itself |
|
1627 static void object_move(MacroAssembler* masm, |
|
1628 OopMap* map, |
|
1629 int oop_handle_offset, |
|
1630 int framesize_in_slots, |
|
1631 VMRegPair src, |
|
1632 VMRegPair dst, |
|
1633 bool is_receiver, |
|
1634 int* receiver_offset) { |
|
1635 |
|
1636 // must pass a handle. First figure out the location we use as a handle |
|
1637 |
|
1638 //FIXME, for mips, dst can be register |
|
1639 if (src.first()->is_stack()) { |
|
1640 // Oop is already on the stack as an argument |
|
1641 Register rHandle = V0; |
|
1642 Label nil; |
|
1643 //__ xorl(rHandle, rHandle); |
|
1644 __ xorr(rHandle, rHandle, rHandle); |
|
1645 //__ cmpl(Address(ebp, reg2offset_in(src.first())), NULL_WORD); |
|
1646 __ ld(AT, FP, reg2offset_in(src.first())); |
|
1647 //__ jcc(Assembler::equal, nil); |
|
1648 __ beq(AT,R0, nil); |
|
1649 __ delayed()->nop(); |
|
1650 // __ leal(rHandle, Address(ebp, reg2offset_in(src.first()))); |
|
1651 __ lea(rHandle, Address(FP, reg2offset_in(src.first()))); |
|
1652 __ bind(nil); |
|
1653 //__ movl(Address(esp, reg2offset_out(dst.first())), rHandle); |
|
1654 if(dst.first()->is_stack())__ sd( rHandle, SP, reg2offset_out(dst.first())); |
|
1655 else __ move( (dst.first())->as_Register(),rHandle); |
|
1656 //if dst is register |
|
1657 //FIXME, do mips need out preserve stack slots? |
|
1658 int offset_in_older_frame = src.first()->reg2stack() |
|
1659 + SharedRuntime::out_preserve_stack_slots(); |
|
1660 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); |
|
1661 if (is_receiver) { |
|
1662 *receiver_offset = (offset_in_older_frame |
|
1663 + framesize_in_slots) * VMRegImpl::stack_slot_size; |
|
1664 } |
|
1665 } else { |
|
1666 // Oop is in an a register we must store it to the space we reserve |
|
1667 // on the stack for oop_handles |
|
1668 const Register rOop = src.first()->as_Register(); |
|
1669 assert( (rOop->encoding() >= A0->encoding()) && (rOop->encoding() <= T0->encoding()),"wrong register"); |
|
1670 // const Register rHandle = eax; |
|
1671 const Register rHandle = V0; |
|
1672 //Important: refer to java_calling_convertion |
|
1673 int oop_slot = (rOop->encoding() - A0->encoding()) * VMRegImpl::slots_per_word + oop_handle_offset; |
|
1674 int offset = oop_slot*VMRegImpl::stack_slot_size; |
|
1675 Label skip; |
|
1676 // __ movl(Address(esp, offset), rOop); |
|
1677 __ sd( rOop , SP, offset ); |
|
1678 map->set_oop(VMRegImpl::stack2reg(oop_slot)); |
|
1679 // __ xorl(rHandle, rHandle); |
|
1680 __ xorr( rHandle, rHandle, rHandle); |
|
1681 //__ cmpl(rOop, NULL_WORD); |
|
1682 // __ jcc(Assembler::equal, skip); |
|
1683 __ beq(rOop, R0, skip); |
|
1684 __ delayed()->nop(); |
|
1685 // __ leal(rHandle, Address(esp, offset)); |
|
1686 __ lea(rHandle, Address(SP, offset)); |
|
1687 __ bind(skip); |
|
1688 // Store the handle parameter |
|
1689 //__ movl(Address(esp, reg2offset_out(dst.first())), rHandle); |
|
1690 if(dst.first()->is_stack())__ sd( rHandle, SP, reg2offset_out(dst.first())); |
|
1691 else __ move((dst.first())->as_Register(), rHandle); |
|
1692 //if dst is register |
|
1693 |
|
1694 if (is_receiver) { |
|
1695 *receiver_offset = offset; |
|
1696 } |
|
1697 } |
|
1698 } |
|
1699 |
|
1700 // A float arg may have to do float reg int reg conversion |
|
1701 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1702 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); |
|
1703 |
|
1704 if (src.first()->is_stack()) { |
|
1705 if(dst.first()->is_stack()){ |
|
1706 // __ movl(eax, Address(ebp, reg2offset_in(src.first()))); |
|
1707 __ lwc1(F12 , FP, reg2offset_in(src.first())); |
|
1708 // __ movl(Address(esp, reg2offset_out(dst.first())), eax); |
|
1709 __ swc1(F12 ,SP, reg2offset_out(dst.first())); |
|
1710 } |
|
1711 else |
|
1712 __ lwc1( dst.first()->as_FloatRegister(), FP, reg2offset_in(src.first())); |
|
1713 } else { |
|
1714 // reg to stack |
|
1715 // __ movss(Address(esp, reg2offset_out(dst.first())), |
|
1716 // src.first()->as_XMMRegister()); |
|
1717 // __ movl(Address(esp, reg2offset_out(dst.first())), eax); |
|
1718 if(dst.first()->is_stack()) |
|
1719 __ swc1( src.first()->as_FloatRegister(),SP, reg2offset_out(dst.first())); |
|
1720 else |
|
1721 __ mov_s( dst.first()->as_FloatRegister(), src.first()->as_FloatRegister()); |
|
1722 } |
|
1723 } |
|
1724 /* |
|
1725 static void split_long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1726 VMRegPair src_lo(src.first()); |
|
1727 VMRegPair src_hi(src.second()); |
|
1728 VMRegPair dst_lo(dst.first()); |
|
1729 VMRegPair dst_hi(dst.second()); |
|
1730 simple_move32(masm, src_lo, dst_lo); |
|
1731 simple_move32(masm, src_hi, dst_hi); |
|
1732 } |
|
1733 */ |
|
1734 // A long move |
|
1735 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1736 |
|
1737 // The only legal possibility for a long_move VMRegPair is: |
|
1738 // 1: two stack slots (possibly unaligned) |
|
1739 // as neither the java or C calling convention will use registers |
|
1740 // for longs. |
|
1741 |
|
1742 if (src.first()->is_stack()) { |
|
1743 assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack"); |
|
1744 // __ movl(eax, Address(ebp, reg2offset_in(src.first()))); |
|
1745 if( dst.first()->is_stack()){ |
|
1746 __ ld(AT, FP, reg2offset_in(src.first())); |
|
1747 // __ movl(ebx, address(ebp, reg2offset_in(src.second()))); |
|
1748 //__ lw(V0, FP, reg2offset_in(src.second())); |
|
1749 // __ movl(address(esp, reg2offset_out(dst.first())), eax); |
|
1750 __ sd(AT, SP, reg2offset_out(dst.first())); |
|
1751 // __ movl(address(esp, reg2offset_out(dst.second())), ebx); |
|
1752 //__ sw(V0, SP, reg2offset_out(dst.second())); |
|
1753 } else{ |
|
1754 __ ld( (dst.first())->as_Register() , FP, reg2offset_in(src.first())); |
|
1755 //__ lw( (dst.second())->as_Register(), FP, reg2offset_in(src.second())); |
|
1756 } |
|
1757 } else { |
|
1758 if( dst.first()->is_stack()){ |
|
1759 __ sd( (src.first())->as_Register(), SP, reg2offset_out(dst.first())); |
|
1760 //__ sw( (src.second())->as_Register(), SP, reg2offset_out(dst.second())); |
|
1761 } else{ |
|
1762 __ move( (dst.first())->as_Register() , (src.first())->as_Register()); |
|
1763 //__ move( (dst.second())->as_Register(), (src.second())->as_Register()); |
|
1764 } |
|
1765 } |
|
1766 } |
|
1767 |
|
1768 // A double move |
|
1769 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
|
1770 |
|
1771 // The only legal possibilities for a double_move VMRegPair are: |
|
1772 // The painful thing here is that like long_move a VMRegPair might be |
|
1773 |
|
1774 // Because of the calling convention we know that src is either |
|
1775 // 1: a single physical register (xmm registers only) |
|
1776 // 2: two stack slots (possibly unaligned) |
|
1777 // dst can only be a pair of stack slots. |
|
1778 |
|
1779 // assert(dst.first()->is_stack() && (src.first()->is_XMMRegister() || |
|
1780 // src.first()->is_stack()), "bad args"); |
|
1781 // assert(dst.first()->is_stack() || src.first()->is_stack()), "bad args"); |
|
1782 |
|
1783 if (src.first()->is_stack()) { |
|
1784 // source is all stack |
|
1785 // __ movl(eax, Address(ebp, reg2offset_in(src.first()))); |
|
1786 if( dst.first()->is_stack()){ |
|
1787 __ ldc1(F12, FP, reg2offset_in(src.first())); |
|
1788 //__ movl(ebx, Address(ebp, reg2offset_in(src.second()))); |
|
1789 //__ lwc1(F14, FP, reg2offset_in(src.second())); |
|
1790 |
|
1791 // __ movl(Address(esp, reg2offset_out(dst.first())), eax); |
|
1792 __ sdc1(F12, SP, reg2offset_out(dst.first())); |
|
1793 // __ movl(Address(esp, reg2offset_out(dst.second())), ebx); |
|
1794 //__ swc1(F14, SP, reg2offset_out(dst.second())); |
|
1795 } else{ |
|
1796 __ ldc1( (dst.first())->as_FloatRegister(), FP, reg2offset_in(src.first())); |
|
1797 //__ lwc1( (dst.second())->as_FloatRegister(), FP, reg2offset_in(src.second())); |
|
1798 } |
|
1799 |
|
1800 } else { |
|
1801 // reg to stack |
|
1802 // No worries about stack alignment |
|
1803 // __ movsd(Address(esp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); |
|
1804 if( dst.first()->is_stack()){ |
|
1805 __ sdc1( src.first()->as_FloatRegister(),SP, reg2offset_out(dst.first())); |
|
1806 //__ swc1( src.second()->as_FloatRegister(),SP, reg2offset_out(dst.second())); |
|
1807 } |
|
1808 else |
|
1809 __ mov_d( dst.first()->as_FloatRegister(), src.first()->as_FloatRegister()); |
|
1810 //__ mov_s( dst.second()->as_FloatRegister(), src.second()->as_FloatRegister()); |
|
1811 |
|
1812 } |
|
1813 } |
|
1814 |
|
1815 static void verify_oop_args(MacroAssembler* masm, |
|
1816 methodHandle method, |
|
1817 const BasicType* sig_bt, |
|
1818 const VMRegPair* regs) { |
|
1819 Register temp_reg = T9; // not part of any compiled calling seq |
|
1820 if (VerifyOops) { |
|
1821 for (int i = 0; i < method->size_of_parameters(); i++) { |
|
1822 if (sig_bt[i] == T_OBJECT || |
|
1823 sig_bt[i] == T_ARRAY) { |
|
1824 VMReg r = regs[i].first(); |
|
1825 assert(r->is_valid(), "bad oop arg"); |
|
1826 if (r->is_stack()) { |
|
1827 // __ movptr(temp_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1828 __ ld(temp_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1829 __ verify_oop(temp_reg); |
|
1830 } else { |
|
1831 __ verify_oop(r->as_Register()); |
|
1832 } |
|
1833 } |
|
1834 } |
|
1835 } |
|
1836 } |
|
1837 |
|
1838 static void gen_special_dispatch(MacroAssembler* masm, |
|
1839 methodHandle method, |
|
1840 const BasicType* sig_bt, |
|
1841 const VMRegPair* regs) { |
|
1842 verify_oop_args(masm, method, sig_bt, regs); |
|
1843 vmIntrinsics::ID iid = method->intrinsic_id(); |
|
1844 |
|
1845 // Now write the args into the outgoing interpreter space |
|
1846 bool has_receiver = false; |
|
1847 Register receiver_reg = noreg; |
|
1848 int member_arg_pos = -1; |
|
1849 Register member_reg = noreg; |
|
1850 int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid); |
|
1851 if (ref_kind != 0) { |
|
1852 member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument |
|
1853 // member_reg = rbx; // known to be free at this point |
|
1854 member_reg = S3; // known to be free at this point |
|
1855 has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind); |
|
1856 } else if (iid == vmIntrinsics::_invokeBasic) { |
|
1857 has_receiver = true; |
|
1858 } else { |
|
1859 fatal(err_msg_res("unexpected intrinsic id %d", iid)); |
|
1860 } |
|
1861 |
|
1862 if (member_reg != noreg) { |
|
1863 // Load the member_arg into register, if necessary. |
|
1864 SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs); |
|
1865 VMReg r = regs[member_arg_pos].first(); |
|
1866 if (r->is_stack()) { |
|
1867 // __ movptr(member_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1868 __ ld(member_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1869 } else { |
|
1870 // no data motion is needed |
|
1871 member_reg = r->as_Register(); |
|
1872 } |
|
1873 } |
|
1874 |
|
1875 if (has_receiver) { |
|
1876 // Make sure the receiver is loaded into a register. |
|
1877 assert(method->size_of_parameters() > 0, "oob"); |
|
1878 assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object"); |
|
1879 VMReg r = regs[0].first(); |
|
1880 assert(r->is_valid(), "bad receiver arg"); |
|
1881 if (r->is_stack()) { |
|
1882 // Porting note: This assumes that compiled calling conventions always |
|
1883 // pass the receiver oop in a register. If this is not true on some |
|
1884 // platform, pick a temp and load the receiver from stack. |
|
1885 fatal("receiver always in a register"); |
|
1886 // receiver_reg = j_rarg0; // known to be free at this point |
|
1887 receiver_reg = SSR; // known to be free at this point |
|
1888 // __ movptr(receiver_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1889 __ ld(receiver_reg, Address(SP, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize)); |
|
1890 } else { |
|
1891 // no data motion is needed |
|
1892 receiver_reg = r->as_Register(); |
|
1893 } |
|
1894 } |
|
1895 |
|
1896 // Figure out which address we are really jumping to: |
|
1897 MethodHandles::generate_method_handle_dispatch(masm, iid, |
|
1898 receiver_reg, member_reg, /*for_compiler_entry:*/ true); |
|
1899 } |
|
1900 |
|
1901 // --------------------------------------------------------------------------- |
|
1902 // Generate a native wrapper for a given method. The method takes arguments |
|
1903 // in the Java compiled code convention, marshals them to the native |
|
1904 // convention (handlizes oops, etc), transitions to native, makes the call, |
|
1905 // returns to java state (possibly blocking), unhandlizes any result and |
|
1906 // returns. |
|
1907 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler* masm, |
|
1908 methodHandle method, |
|
1909 int compile_id, |
|
1910 BasicType *in_sig_bt, |
|
1911 VMRegPair *in_regs, |
|
1912 BasicType ret_type) { |
|
1913 |
|
1914 if (method->is_method_handle_intrinsic()) { |
|
1915 vmIntrinsics::ID iid = method->intrinsic_id(); |
|
1916 intptr_t start = (intptr_t)__ pc(); |
|
1917 int vep_offset = ((intptr_t)__ pc()) - start; |
|
1918 |
|
1919 gen_special_dispatch(masm, |
|
1920 method, |
|
1921 in_sig_bt, |
|
1922 in_regs); |
|
1923 |
|
1924 int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period |
|
1925 __ flush(); |
|
1926 int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually |
|
1927 return nmethod::new_native_nmethod(method, |
|
1928 compile_id, |
|
1929 masm->code(), |
|
1930 vep_offset, |
|
1931 frame_complete, |
|
1932 stack_slots / VMRegImpl::slots_per_word, |
|
1933 in_ByteSize(-1), |
|
1934 in_ByteSize(-1), |
|
1935 (OopMapSet*)NULL); |
|
1936 } |
|
1937 bool is_critical_native = true; |
|
1938 address native_func = method->critical_native_function(); |
|
1939 if (native_func == NULL) { |
|
1940 native_func = method->native_function(); |
|
1941 is_critical_native = false; |
|
1942 } |
|
1943 assert(native_func != NULL, "must have function"); |
|
1944 |
|
1945 // Native nmethod wrappers never take possesion of the oop arguments. |
|
1946 // So the caller will gc the arguments. The only thing we need an |
|
1947 // oopMap for is if the call is static |
|
1948 // |
|
1949 // An OopMap for lock (and class if static), and one for the VM call itself |
|
1950 OopMapSet *oop_maps = new OopMapSet(); |
|
1951 |
|
1952 // We have received a description of where all the java arg are located |
|
1953 // on entry to the wrapper. We need to convert these args to where |
|
1954 // the jni function will expect them. To figure out where they go |
|
1955 // we convert the java signature to a C signature by inserting |
|
1956 // the hidden arguments as arg[0] and possibly arg[1] (static method) |
|
1957 |
|
1958 const int total_in_args = method->size_of_parameters(); |
|
1959 int total_c_args = total_in_args; |
|
1960 if (!is_critical_native) { |
|
1961 total_c_args += 1; |
|
1962 if (method->is_static()) { |
|
1963 total_c_args++; |
|
1964 } |
|
1965 } else { |
|
1966 for (int i = 0; i < total_in_args; i++) { |
|
1967 if (in_sig_bt[i] == T_ARRAY) { |
|
1968 total_c_args++; |
|
1969 } |
|
1970 } |
|
1971 } |
|
1972 |
|
1973 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); |
|
1974 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); |
|
1975 BasicType* in_elem_bt = NULL; |
|
1976 |
|
1977 int argc = 0; |
|
1978 if (!is_critical_native) { |
|
1979 out_sig_bt[argc++] = T_ADDRESS; |
|
1980 if (method->is_static()) { |
|
1981 out_sig_bt[argc++] = T_OBJECT; |
|
1982 } |
|
1983 |
|
1984 for (int i = 0; i < total_in_args ; i++ ) { |
|
1985 out_sig_bt[argc++] = in_sig_bt[i]; |
|
1986 } |
|
1987 } else { |
|
1988 Thread* THREAD = Thread::current(); |
|
1989 in_elem_bt = NEW_RESOURCE_ARRAY(BasicType, total_in_args); |
|
1990 SignatureStream ss(method->signature()); |
|
1991 for (int i = 0; i < total_in_args ; i++ ) { |
|
1992 if (in_sig_bt[i] == T_ARRAY) { |
|
1993 // Arrays are passed as int, elem* pair |
|
1994 out_sig_bt[argc++] = T_INT; |
|
1995 out_sig_bt[argc++] = T_ADDRESS; |
|
1996 Symbol* atype = ss.as_symbol(CHECK_NULL); |
|
1997 const char* at = atype->as_C_string(); |
|
1998 if (strlen(at) == 2) { |
|
1999 assert(at[0] == '[', "must be"); |
|
2000 switch (at[1]) { |
|
2001 case 'B': in_elem_bt[i] = T_BYTE; break; |
|
2002 case 'C': in_elem_bt[i] = T_CHAR; break; |
|
2003 case 'D': in_elem_bt[i] = T_DOUBLE; break; |
|
2004 case 'F': in_elem_bt[i] = T_FLOAT; break; |
|
2005 case 'I': in_elem_bt[i] = T_INT; break; |
|
2006 case 'J': in_elem_bt[i] = T_LONG; break; |
|
2007 case 'S': in_elem_bt[i] = T_SHORT; break; |
|
2008 case 'Z': in_elem_bt[i] = T_BOOLEAN; break; |
|
2009 default: ShouldNotReachHere(); |
|
2010 } |
|
2011 } |
|
2012 } else { |
|
2013 out_sig_bt[argc++] = in_sig_bt[i]; |
|
2014 in_elem_bt[i] = T_VOID; |
|
2015 } |
|
2016 if (in_sig_bt[i] != T_VOID) { |
|
2017 assert(in_sig_bt[i] == ss.type(), "must match"); |
|
2018 ss.next(); |
|
2019 } |
|
2020 } |
|
2021 } |
|
2022 |
|
2023 // Now figure out where the args must be stored and how much stack space |
|
2024 // they require (neglecting out_preserve_stack_slots but space for storing |
|
2025 // the 1st six register arguments). It's weird see int_stk_helper. |
|
2026 // |
|
2027 int out_arg_slots; |
|
2028 //out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); |
|
2029 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args); |
|
2030 |
|
2031 // Compute framesize for the wrapper. We need to handlize all oops in |
|
2032 // registers. We must create space for them here that is disjoint from |
|
2033 // the windowed save area because we have no control over when we might |
|
2034 // flush the window again and overwrite values that gc has since modified. |
|
2035 // (The live window race) |
|
2036 // |
|
2037 // We always just allocate 6 word for storing down these object. This allow |
|
2038 // us to simply record the base and use the Ireg number to decide which |
|
2039 // slot to use. (Note that the reg number is the inbound number not the |
|
2040 // outbound number). |
|
2041 // We must shuffle args to match the native convention, and include var-args space. |
|
2042 |
|
2043 // Calculate the total number of stack slots we will need. |
|
2044 |
|
2045 // First count the abi requirement plus all of the outgoing args |
|
2046 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
|
2047 |
|
2048 // Now the space for the inbound oop handle area |
|
2049 int total_save_slots = 9 * VMRegImpl::slots_per_word; // 9 arguments passed in registers |
|
2050 if (is_critical_native) { |
|
2051 // Critical natives may have to call out so they need a save area |
|
2052 // for register arguments. |
|
2053 int double_slots = 0; |
|
2054 int single_slots = 0; |
|
2055 for ( int i = 0; i < total_in_args; i++) { |
|
2056 if (in_regs[i].first()->is_Register()) { |
|
2057 const Register reg = in_regs[i].first()->as_Register(); |
|
2058 switch (in_sig_bt[i]) { |
|
2059 case T_BOOLEAN: |
|
2060 case T_BYTE: |
|
2061 case T_SHORT: |
|
2062 case T_CHAR: |
|
2063 case T_INT: single_slots++; break; |
|
2064 case T_ARRAY: // specific to LP64 (7145024) |
|
2065 case T_LONG: double_slots++; break; |
|
2066 default: ShouldNotReachHere(); |
|
2067 } |
|
2068 } else if (in_regs[i].first()->is_FloatRegister()) { |
|
2069 switch (in_sig_bt[i]) { |
|
2070 case T_FLOAT: single_slots++; break; |
|
2071 case T_DOUBLE: double_slots++; break; |
|
2072 default: ShouldNotReachHere(); |
|
2073 } |
|
2074 } |
|
2075 } |
|
2076 total_save_slots = double_slots * 2 + single_slots; |
|
2077 // align the save area |
|
2078 if (double_slots != 0) { |
|
2079 stack_slots = round_to(stack_slots, 2); |
|
2080 } |
|
2081 } |
|
2082 |
|
2083 int oop_handle_offset = stack_slots; |
|
2084 // stack_slots += 9*VMRegImpl::slots_per_word; // T0, A0 ~ A7 |
|
2085 stack_slots += total_save_slots; |
|
2086 |
|
2087 // Now any space we need for handlizing a klass if static method |
|
2088 |
|
2089 int klass_slot_offset = 0; |
|
2090 int klass_offset = -1; |
|
2091 int lock_slot_offset = 0; |
|
2092 bool is_static = false; |
|
2093 //int oop_temp_slot_offset = 0; |
|
2094 |
|
2095 if (method->is_static()) { |
|
2096 klass_slot_offset = stack_slots; |
|
2097 stack_slots += VMRegImpl::slots_per_word; |
|
2098 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; |
|
2099 is_static = true; |
|
2100 } |
|
2101 |
|
2102 // Plus a lock if needed |
|
2103 |
|
2104 if (method->is_synchronized()) { |
|
2105 lock_slot_offset = stack_slots; |
|
2106 stack_slots += VMRegImpl::slots_per_word; |
|
2107 } |
|
2108 |
|
2109 // Now a place to save return value or as a temporary for any gpr -> fpr moves |
|
2110 // + 2 for return address (which we own) and saved ebp |
|
2111 //stack_slots += 2; |
|
2112 stack_slots += 2 + 9 * VMRegImpl::slots_per_word; // (T0, A0, A1, A2, A3, A4, A5, A6, A7) |
|
2113 |
|
2114 // Ok The space we have allocated will look like: |
|
2115 // |
|
2116 // |
|
2117 // FP-> | | |
|
2118 // |---------------------| |
|
2119 // | 2 slots for moves | |
|
2120 // |---------------------| |
|
2121 // | lock box (if sync) | |
|
2122 // |---------------------| <- lock_slot_offset |
|
2123 // | klass (if static) | |
|
2124 // |---------------------| <- klass_slot_offset |
|
2125 // | oopHandle area | |
|
2126 // |---------------------| <- oop_handle_offset |
|
2127 // | outbound memory | |
|
2128 // | based arguments | |
|
2129 // | | |
|
2130 // |---------------------| |
|
2131 // | vararg area | |
|
2132 // |---------------------| |
|
2133 // | | |
|
2134 // SP-> | out_preserved_slots | |
|
2135 // |
|
2136 // |
|
2137 |
|
2138 |
|
2139 // Now compute actual number of stack words we need rounding to make |
|
2140 // stack properly aligned. |
|
2141 stack_slots = round_to(stack_slots, StackAlignmentInSlots); |
|
2142 |
|
2143 int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
|
2144 |
|
2145 intptr_t start = (intptr_t)__ pc(); |
|
2146 |
|
2147 |
|
2148 |
|
2149 // First thing make an ic check to see if we should even be here |
|
2150 address ic_miss = SharedRuntime::get_ic_miss_stub(); |
|
2151 |
|
2152 // We are free to use all registers as temps without saving them and |
|
2153 // restoring them except ebp. ebp is the only callee save register |
|
2154 // as far as the interpreter and the compiler(s) are concerned. |
|
2155 |
|
2156 //refer to register_mips.hpp:IC_Klass |
|
2157 const Register ic_reg = T1; |
|
2158 const Register receiver = T0; |
|
2159 Label hit; |
|
2160 Label exception_pending; |
|
2161 |
|
2162 __ verify_oop(receiver); |
|
2163 //__ lw(AT, receiver, oopDesc::klass_offset_in_bytes()); |
|
2164 //add for compressedoops |
|
2165 __ load_klass(AT, receiver); |
|
2166 __ beq(AT, ic_reg, hit); |
|
2167 __ delayed()->nop(); |
|
2168 __ jmp(ic_miss, relocInfo::runtime_call_type); |
|
2169 __ delayed()->nop(); |
|
2170 // verified entry must be aligned for code patching. |
|
2171 // and the first 5 bytes must be in the same cache line |
|
2172 // if we align at 8 then we will be sure 5 bytes are in the same line |
|
2173 __ align(8); |
|
2174 |
|
2175 __ bind(hit); |
|
2176 |
|
2177 |
|
2178 int vep_offset = ((intptr_t)__ pc()) - start; |
|
2179 #ifdef COMPILER1 |
|
2180 if (InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) { |
|
2181 // Object.hashCode can pull the hashCode from the header word |
|
2182 // instead of doing a full VM transition once it's been computed. |
|
2183 // Since hashCode is usually polymorphic at call sites we can't do |
|
2184 // this optimization at the call site without a lot of work. |
|
2185 Label slowCase; |
|
2186 Register receiver = T0; |
|
2187 Register result = V0; |
|
2188 __ ld ( result, receiver, oopDesc::mark_offset_in_bytes()); |
|
2189 // check if locked |
|
2190 __ andi(AT, result, markOopDesc::unlocked_value); |
|
2191 __ beq(AT, R0, slowCase); |
|
2192 __ delayed()->nop(); |
|
2193 if (UseBiasedLocking) { |
|
2194 // Check if biased and fall through to runtime if so |
|
2195 __ andi (AT, result, markOopDesc::biased_lock_bit_in_place); |
|
2196 __ bne(AT,R0, slowCase); |
|
2197 __ delayed()->nop(); |
|
2198 } |
|
2199 // get hash |
|
2200 __ li(AT, markOopDesc::hash_mask_in_place); |
|
2201 __ andr (AT, result, AT); |
|
2202 // test if hashCode exists |
|
2203 __ beq (AT, R0, slowCase); |
|
2204 __ delayed()->nop(); |
|
2205 __ shr(result, markOopDesc::hash_shift); |
|
2206 __ jr(RA); |
|
2207 __ delayed()->nop(); |
|
2208 __ bind (slowCase); |
|
2209 } |
|
2210 #endif // COMPILER1 |
|
2211 |
|
2212 // The instruction at the verified entry point must be 5 bytes or longer |
|
2213 // because it can be patched on the fly by make_non_entrant. The stack bang |
|
2214 // instruction fits that requirement. |
|
2215 |
|
2216 // Generate stack overflow check |
|
2217 |
|
2218 if (UseStackBanging) { |
|
2219 //this function will modify the value in A0 |
|
2220 __ push(A0); |
|
2221 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
|
2222 __ pop(A0); |
|
2223 } else { |
|
2224 // need a 5 byte instruction to allow MT safe patching to non-entrant |
|
2225 __ nop(); |
|
2226 __ nop(); |
|
2227 __ nop(); |
|
2228 __ nop(); |
|
2229 __ nop(); |
|
2230 } |
|
2231 // Generate a new frame for the wrapper. |
|
2232 // do mips need this ? |
|
2233 #ifndef OPT_THREAD |
|
2234 __ get_thread(TREG); |
|
2235 #endif |
|
2236 //FIXME here |
|
2237 __ st_ptr(SP, TREG, in_bytes(JavaThread::last_Java_sp_offset())); |
|
2238 // -2 because return address is already present and so is saved ebp |
|
2239 __ move(AT, -(StackAlignmentInBytes)); |
|
2240 __ andr(SP, SP, AT); |
|
2241 |
|
2242 __ enter(); |
|
2243 __ addiu(SP, SP, -1 * (stack_size - 2*wordSize)); |
|
2244 |
|
2245 // Frame is now completed as far a size and linkage. |
|
2246 |
|
2247 int frame_complete = ((intptr_t)__ pc()) - start; |
|
2248 |
|
2249 // Calculate the difference between esp and ebp. We need to know it |
|
2250 // after the native call because on windows Java Natives will pop |
|
2251 // the arguments and it is painful to do esp relative addressing |
|
2252 // in a platform independent way. So after the call we switch to |
|
2253 // ebp relative addressing. |
|
2254 //FIXME actually , the fp_adjustment may not be the right, because andr(sp,sp,at)may change |
|
2255 //the SP |
|
2256 int fp_adjustment = stack_size - 2*wordSize; |
|
2257 |
|
2258 #ifdef COMPILER2 |
|
2259 // C2 may leave the stack dirty if not in SSE2+ mode |
|
2260 // if (UseSSE >= 2) { |
|
2261 // __ verify_FPU(0, "c2i transition should have clean FPU stack"); |
|
2262 //} else { |
|
2263 __ empty_FPU_stack(); |
|
2264 //} |
|
2265 #endif /* COMPILER2 */ |
|
2266 |
|
2267 // Compute the ebp offset for any slots used after the jni call |
|
2268 |
|
2269 int lock_slot_ebp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; |
|
2270 // We use edi as a thread pointer because it is callee save and |
|
2271 // if we load it once it is usable thru the entire wrapper |
|
2272 // const Register thread = edi; |
|
2273 const Register thread = TREG; |
|
2274 |
|
2275 // We use esi as the oop handle for the receiver/klass |
|
2276 // It is callee save so it survives the call to native |
|
2277 |
|
2278 // const Register oop_handle_reg = esi; |
|
2279 const Register oop_handle_reg = S4; |
|
2280 if (is_critical_native) { |
|
2281 __ stop("generate_native_wrapper in sharedRuntime <2>"); |
|
2282 //TODO:Fu |
|
2283 /* |
|
2284 check_needs_gc_for_critical_native(masm, stack_slots, total_c_args, total_in_args, |
|
2285 oop_handle_offset, oop_maps, in_regs, in_sig_bt); |
|
2286 */ |
|
2287 } |
|
2288 |
|
2289 #ifndef OPT_THREAD |
|
2290 __ get_thread(thread); |
|
2291 #endif |
|
2292 |
|
2293 // |
|
2294 // We immediately shuffle the arguments so that any vm call we have to |
|
2295 // make from here on out (sync slow path, jvmpi, etc.) we will have |
|
2296 // captured the oops from our caller and have a valid oopMap for |
|
2297 // them. |
|
2298 |
|
2299 // ----------------- |
|
2300 // The Grand Shuffle |
|
2301 // |
|
2302 // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv* |
|
2303 // and, if static, the class mirror instead of a receiver. This pretty much |
|
2304 // guarantees that register layout will not match (and mips doesn't use reg |
|
2305 // parms though amd does). Since the native abi doesn't use register args |
|
2306 // and the java conventions does we don't have to worry about collisions. |
|
2307 // All of our moved are reg->stack or stack->stack. |
|
2308 // We ignore the extra arguments during the shuffle and handle them at the |
|
2309 // last moment. The shuffle is described by the two calling convention |
|
2310 // vectors we have in our possession. We simply walk the java vector to |
|
2311 // get the source locations and the c vector to get the destinations. |
|
2312 |
|
2313 int c_arg = method->is_static() ? 2 : 1 ; |
|
2314 |
|
2315 // Record esp-based slot for receiver on stack for non-static methods |
|
2316 int receiver_offset = -1; |
|
2317 |
|
2318 // This is a trick. We double the stack slots so we can claim |
|
2319 // the oops in the caller's frame. Since we are sure to have |
|
2320 // more args than the caller doubling is enough to make |
|
2321 // sure we can capture all the incoming oop args from the |
|
2322 // caller. |
|
2323 // |
|
2324 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); |
|
2325 |
|
2326 // Mark location of rbp (someday) |
|
2327 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp)); |
|
2328 |
|
2329 // Use eax, ebx as temporaries during any memory-memory moves we have to do |
|
2330 // All inbound args are referenced based on rbp and all outbound args via rsp. |
|
2331 |
|
2332 |
|
2333 |
|
2334 #ifdef ASSERT |
|
2335 bool reg_destroyed[RegisterImpl::number_of_registers]; |
|
2336 bool freg_destroyed[FloatRegisterImpl::number_of_registers]; |
|
2337 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { |
|
2338 reg_destroyed[r] = false; |
|
2339 } |
|
2340 for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) { |
|
2341 freg_destroyed[f] = false; |
|
2342 } |
|
2343 |
|
2344 #endif /* ASSERT */ |
|
2345 |
|
2346 // We know that we only have args in at most two integer registers (ecx, edx). So eax, ebx |
|
2347 // Are free to temporaries if we have to do stack to steck moves. |
|
2348 // All inbound args are referenced based on ebp and all outbound args via esp. |
|
2349 |
|
2350 // This may iterate in two different directions depending on the |
|
2351 // kind of native it is. The reason is that for regular JNI natives |
|
2352 // the incoming and outgoing registers are offset upwards and for |
|
2353 // critical natives they are offset down. |
|
2354 GrowableArray<int> arg_order(2 * total_in_args); |
|
2355 VMRegPair tmp_vmreg; |
|
2356 // tmp_vmreg.set1(rbx->as_VMReg()); |
|
2357 tmp_vmreg.set1(T8->as_VMReg()); |
|
2358 |
|
2359 if (!is_critical_native) { |
|
2360 for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) { |
|
2361 arg_order.push(i); |
|
2362 arg_order.push(c_arg); |
|
2363 } |
|
2364 } else { |
|
2365 // Compute a valid move order, using tmp_vmreg to break any cycles |
|
2366 __ stop("generate_native_wrapper in sharedRuntime <2>"); |
|
2367 //TODO:Fu |
|
2368 // ComputeMoveOrder cmo(total_in_args, in_regs, total_c_args, out_regs, in_sig_bt, arg_order, tmp_vmreg); |
|
2369 } |
|
2370 |
|
2371 int temploc = -1; |
|
2372 for (int ai = 0; ai < arg_order.length(); ai += 2) { |
|
2373 int i = arg_order.at(ai); |
|
2374 int c_arg = arg_order.at(ai + 1); |
|
2375 __ block_comment(err_msg("move %d -> %d", i, c_arg)); |
|
2376 if (c_arg == -1) { |
|
2377 assert(is_critical_native, "should only be required for critical natives"); |
|
2378 // This arg needs to be moved to a temporary |
|
2379 __ move(tmp_vmreg.first()->as_Register(), in_regs[i].first()->as_Register()); |
|
2380 in_regs[i] = tmp_vmreg; |
|
2381 temploc = i; |
|
2382 continue; |
|
2383 } else if (i == -1) { |
|
2384 assert(is_critical_native, "should only be required for critical natives"); |
|
2385 // Read from the temporary location |
|
2386 assert(temploc != -1, "must be valid"); |
|
2387 i = temploc; |
|
2388 temploc = -1; |
|
2389 } |
|
2390 #ifdef ASSERT |
|
2391 if (in_regs[i].first()->is_Register()) { |
|
2392 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!"); |
|
2393 } else if (in_regs[i].first()->is_FloatRegister()) { |
|
2394 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!"); |
|
2395 } |
|
2396 if (out_regs[c_arg].first()->is_Register()) { |
|
2397 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; |
|
2398 } else if (out_regs[c_arg].first()->is_FloatRegister()) { |
|
2399 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true; |
|
2400 } |
|
2401 #endif /* ASSERT */ |
|
2402 switch (in_sig_bt[i]) { |
|
2403 case T_ARRAY: |
|
2404 if (is_critical_native) { |
|
2405 __ stop("generate_native_wrapper in sharedRuntime <2>"); |
|
2406 //TODO:Fu |
|
2407 // unpack_array_argument(masm, in_regs[i], in_elem_bt[i], out_regs[c_arg + 1], out_regs[c_arg]); |
|
2408 c_arg++; |
|
2409 #ifdef ASSERT |
|
2410 if (out_regs[c_arg].first()->is_Register()) { |
|
2411 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; |
|
2412 } else if (out_regs[c_arg].first()->is_FloatRegister()) { |
|
2413 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true; |
|
2414 } |
|
2415 #endif |
|
2416 break; |
|
2417 } |
|
2418 case T_OBJECT: |
|
2419 assert(!is_critical_native, "no oop arguments"); |
|
2420 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], |
|
2421 ((i == 0) && (!is_static)), |
|
2422 &receiver_offset); |
|
2423 break; |
|
2424 case T_VOID: |
|
2425 break; |
|
2426 |
|
2427 case T_FLOAT: |
|
2428 float_move(masm, in_regs[i], out_regs[c_arg]); |
|
2429 break; |
|
2430 |
|
2431 case T_DOUBLE: |
|
2432 assert( i + 1 < total_in_args && |
|
2433 in_sig_bt[i + 1] == T_VOID && |
|
2434 out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
|
2435 double_move(masm, in_regs[i], out_regs[c_arg]); |
|
2436 break; |
|
2437 |
|
2438 case T_LONG : |
|
2439 long_move(masm, in_regs[i], out_regs[c_arg]); |
|
2440 break; |
|
2441 |
|
2442 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
|
2443 |
|
2444 default: |
|
2445 // move32_64(masm, in_regs[i], out_regs[c_arg]); |
|
2446 simple_move32(masm, in_regs[i], out_regs[c_arg]); |
|
2447 } |
|
2448 } |
|
2449 |
|
2450 // point c_arg at the first arg that is already loaded in case we |
|
2451 // need to spill before we call out |
|
2452 c_arg = total_c_args - total_in_args; |
|
2453 // Pre-load a static method's oop into esi. Used both by locking code and |
|
2454 // the normal JNI call code. |
|
2455 |
|
2456 __ move(oop_handle_reg, A1); |
|
2457 |
|
2458 if (method->is_static() && !is_critical_native) { |
|
2459 |
|
2460 // load opp into a register |
|
2461 int oop_index = __ oop_recorder()->find_index(JNIHandles::make_local( |
|
2462 (method->method_holder())->java_mirror())); |
|
2463 |
|
2464 |
|
2465 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
|
2466 __ relocate(rspec); |
|
2467 //__ lui(oop_handle_reg, Assembler::split_high((int)JNIHandles::make_local( |
|
2468 // Klass::cast(method->method_holder())->java_mirror()))); |
|
2469 //__ addiu(oop_handle_reg, oop_handle_reg, Assembler::split_low((int) |
|
2470 // JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror()))); |
|
2471 __ li48(oop_handle_reg, (long)JNIHandles::make_local((method->method_holder())->java_mirror())); |
|
2472 // __ verify_oop(oop_handle_reg); |
|
2473 // Now handlize the static class mirror it's known not-null. |
|
2474 __ sd( oop_handle_reg, SP, klass_offset); |
|
2475 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); |
|
2476 |
|
2477 // Now get the handle |
|
2478 __ lea(oop_handle_reg, Address(SP, klass_offset)); |
|
2479 // store the klass handle as second argument |
|
2480 __ move(A1, oop_handle_reg); |
|
2481 // and protect the arg if we must spill |
|
2482 c_arg--; |
|
2483 } |
|
2484 // Change state to native (we save the return address in the thread, since it might not |
|
2485 // be pushed on the stack when we do a a stack traversal). It is enough that the pc() |
|
2486 // points into the right code segment. It does not have to be the correct return pc. |
|
2487 // We use the same pc/oopMap repeatedly when we call out |
|
2488 |
|
2489 intptr_t the_pc = (intptr_t) __ pc(); |
|
2490 |
|
2491 oop_maps->add_gc_map(the_pc - start, map); |
|
2492 |
|
2493 //__ set_last_Java_frame(thread, esp, noreg, (address)the_pc); |
|
2494 __ set_last_Java_frame(SP, noreg, NULL); |
|
2495 __ relocate(relocInfo::internal_pc_type); |
|
2496 { |
|
2497 intptr_t save_pc = (intptr_t)the_pc ; |
|
2498 __ li48(AT, save_pc); |
|
2499 } |
|
2500 __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
|
2501 |
|
2502 |
|
2503 // We have all of the arguments setup at this point. We must not touch any register |
|
2504 // argument registers at this point (what if we save/restore them there are no oop? |
|
2505 { |
|
2506 SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
|
2507 int metadata_index = __ oop_recorder()->find_index(method()); |
|
2508 RelocationHolder rspec = metadata_Relocation::spec(metadata_index); |
|
2509 __ relocate(rspec); |
|
2510 //__ lui(T6, Assembler::split_high((int)JNIHandles::make_local(method()))); |
|
2511 //__ addiu(T6, T6, Assembler::split_low((int)JNIHandles::make_local(method()))); |
|
2512 __ li48(AT, (long)(method())); |
|
2513 |
|
2514 __ call_VM_leaf( |
|
2515 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), |
|
2516 thread, AT); |
|
2517 |
|
2518 } |
|
2519 |
|
2520 // These are register definitions we need for locking/unlocking |
|
2521 // const Register swap_reg = eax; // Must use eax for cmpxchg instruction |
|
2522 // const Register obj_reg = ecx; // Will contain the oop |
|
2523 // const Register lock_reg = edx; // Address of compiler lock object (BasicLock) |
|
2524 //FIXME, I hava no idea which register to use |
|
2525 const Register swap_reg = T8; // Must use eax for cmpxchg instruction |
|
2526 const Register obj_reg = T9; // Will contain the oop |
|
2527 //const Register lock_reg = T6; // Address of compiler lock object (BasicLock) |
|
2528 const Register lock_reg = c_rarg0; // Address of compiler lock object (BasicLock) |
|
2529 |
|
2530 |
|
2531 |
|
2532 Label slow_path_lock; |
|
2533 Label lock_done; |
|
2534 |
|
2535 // Lock a synchronized method |
|
2536 if (method->is_synchronized()) { |
|
2537 assert(!is_critical_native, "unhandled"); |
|
2538 |
|
2539 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); |
|
2540 |
|
2541 // Get the handle (the 2nd argument) |
|
2542 __ move(oop_handle_reg, A1); |
|
2543 |
|
2544 // Get address of the box |
|
2545 __ lea(lock_reg, Address(FP, lock_slot_ebp_offset)); |
|
2546 |
|
2547 // Load the oop from the handle |
|
2548 __ ld(obj_reg, oop_handle_reg, 0); |
|
2549 |
|
2550 if (UseBiasedLocking) { |
|
2551 // Note that oop_handle_reg is trashed during this call |
|
2552 __ biased_locking_enter(lock_reg, obj_reg, swap_reg, A1, |
|
2553 false, lock_done, &slow_path_lock); |
|
2554 } |
|
2555 |
|
2556 // Load immediate 1 into swap_reg %eax |
|
2557 __ move(swap_reg, 1); |
|
2558 |
|
2559 __ ld(AT, obj_reg, 0); |
|
2560 __ orr(swap_reg, swap_reg, AT); |
|
2561 |
|
2562 __ sd( swap_reg, lock_reg, mark_word_offset); |
|
2563 __ cmpxchg(lock_reg, Address(obj_reg, 0), swap_reg); |
|
2564 __ bne(AT, R0, lock_done); |
|
2565 __ delayed()->nop(); |
|
2566 // Test if the oopMark is an obvious stack pointer, i.e., |
|
2567 // 1) (mark & 3) == 0, and |
|
2568 // 2) esp <= mark < mark + os::pagesize() |
|
2569 // These 3 tests can be done by evaluating the following |
|
2570 // expression: ((mark - esp) & (3 - os::vm_page_size())), |
|
2571 // assuming both stack pointer and pagesize have their |
|
2572 // least significant 2 bits clear. |
|
2573 // NOTE: the oopMark is in swap_reg %eax as the result of cmpxchg |
|
2574 |
|
2575 __ dsub(swap_reg, swap_reg,SP); |
|
2576 __ move(AT, 3 - os::vm_page_size()); |
|
2577 __ andr(swap_reg , swap_reg, AT); |
|
2578 // Save the test result, for recursive case, the result is zero |
|
2579 __ sd(swap_reg, lock_reg, mark_word_offset); |
|
2580 //FIXME here, Why notEqual? |
|
2581 __ bne(swap_reg,R0, slow_path_lock); |
|
2582 __ delayed()->nop(); |
|
2583 // Slow path will re-enter here |
|
2584 __ bind(lock_done); |
|
2585 |
|
2586 if (UseBiasedLocking) { |
|
2587 // Re-fetch oop_handle_reg as we trashed it above |
|
2588 __ move(A1, oop_handle_reg); |
|
2589 } |
|
2590 } |
|
2591 |
|
2592 |
|
2593 // Finally just about ready to make the JNI call |
|
2594 |
|
2595 |
|
2596 // get JNIEnv* which is first argument to native |
|
2597 if (!is_critical_native) { |
|
2598 __ addi(A0, thread, in_bytes(JavaThread::jni_environment_offset())); |
|
2599 } |
|
2600 |
|
2601 // Example: Java_java_lang_ref_Finalizer_invokeFinalizeMethod(JNIEnv *env, jclass clazz, jobject ob) |
|
2602 /* Load the second arguments into A1 */ |
|
2603 //__ ld(A1, SP , wordSize ); // klass |
|
2604 |
|
2605 // Now set thread in native |
|
2606 __ addi(AT, R0, _thread_in_native); |
|
2607 __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
|
2608 /* Jin: do the call */ |
|
2609 __ call(method->native_function(), relocInfo::runtime_call_type); |
|
2610 __ delayed()->nop(); |
|
2611 // WARNING - on Windows Java Natives use pascal calling convention and pop the |
|
2612 // arguments off of the stack. We could just re-adjust the stack pointer here |
|
2613 // and continue to do SP relative addressing but we instead switch to FP |
|
2614 // relative addressing. |
|
2615 |
|
2616 // Unpack native results. |
|
2617 switch (ret_type) { |
|
2618 case T_BOOLEAN: __ c2bool(V0); break; |
|
2619 case T_CHAR : __ andi(V0,V0, 0xFFFF); break; |
|
2620 case T_BYTE : __ sign_extend_byte (V0); break; |
|
2621 case T_SHORT : __ sign_extend_short(V0); break; |
|
2622 case T_INT : // nothing to do break; |
|
2623 case T_DOUBLE : |
|
2624 case T_FLOAT : |
|
2625 // Result is in st0 we'll save as needed |
|
2626 break; |
|
2627 case T_ARRAY: // Really a handle |
|
2628 case T_OBJECT: // Really a handle |
|
2629 break; // can't de-handlize until after safepoint check |
|
2630 case T_VOID: break; |
|
2631 case T_LONG: break; |
|
2632 default : ShouldNotReachHere(); |
|
2633 } |
|
2634 // Switch thread to "native transition" state before reading the synchronization state. |
|
2635 // This additional state is necessary because reading and testing the synchronization |
|
2636 // state is not atomic w.r.t. GC, as this scenario demonstrates: |
|
2637 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. |
|
2638 // VM thread changes sync state to synchronizing and suspends threads for GC. |
|
2639 // Thread A is resumed to finish this native method, but doesn't block here since it |
|
2640 // didn't see any synchronization is progress, and escapes. |
|
2641 // __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); |
|
2642 //__ sw(_thread_in_native_trans, thread, JavaThread::thread_state_offset()); |
|
2643 // __ move(AT, (int)_thread_in_native_trans); |
|
2644 __ addi(AT, R0, _thread_in_native_trans); |
|
2645 __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
|
2646 |
|
2647 Label after_transition; |
|
2648 |
|
2649 // check for safepoint operation in progress and/or pending suspend requests |
|
2650 { Label Continue; |
|
2651 //FIXME here, which regiser should we use? |
|
2652 // SafepointSynchronize::_not_synchronized); |
|
2653 __ li(AT, SafepointSynchronize::address_of_state()); |
|
2654 __ lw(A0, AT, 0); |
|
2655 __ addi(AT, A0, -SafepointSynchronize::_not_synchronized); |
|
2656 Label L; |
|
2657 __ bne(AT,R0, L); |
|
2658 __ delayed()->nop(); |
|
2659 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset())); |
|
2660 __ beq(AT, R0, Continue); |
|
2661 __ delayed()->nop(); |
|
2662 __ bind(L); |
|
2663 |
|
2664 // Don't use call_VM as it will see a possible pending exception and forward it |
|
2665 // and never return here preventing us from clearing _last_native_pc down below. |
|
2666 // Also can't use call_VM_leaf either as it will check to see if esi & edi are |
|
2667 // preserved and correspond to the bcp/locals pointers. So we do a runtime call |
|
2668 // by hand. |
|
2669 // |
|
2670 save_native_result(masm, ret_type, stack_slots); |
|
2671 __ move (A0, thread); |
|
2672 __ addi(SP,SP, -wordSize); |
|
2673 if (!is_critical_native) { |
|
2674 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::runtime_call_type); |
|
2675 __ delayed()->nop(); |
|
2676 } else { |
|
2677 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans_and_transition), relocInfo::runtime_call_type); |
|
2678 __ delayed()->nop(); |
|
2679 } |
|
2680 // __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::runtime_call_type); |
|
2681 // __ delayed()->nop(); |
|
2682 __ addi(SP,SP, wordSize); |
|
2683 //add for compressedoops |
|
2684 __ reinit_heapbase(); |
|
2685 // Restore any method result value |
|
2686 restore_native_result(masm, ret_type, stack_slots); |
|
2687 |
|
2688 if (is_critical_native) { |
|
2689 // The call above performed the transition to thread_in_Java so |
|
2690 // skip the transition logic below. |
|
2691 __ beq(R0, R0, after_transition); |
|
2692 __ delayed()->nop(); |
|
2693 } |
|
2694 |
|
2695 __ bind(Continue); |
|
2696 } |
|
2697 |
|
2698 // change thread state |
|
2699 __ addi(AT, R0, _thread_in_Java); |
|
2700 __ sw(AT, thread, in_bytes(JavaThread::thread_state_offset())); |
|
2701 __ bind(after_transition); |
|
2702 Label reguard; |
|
2703 Label reguard_done; |
|
2704 __ ld(AT, thread, in_bytes(JavaThread::stack_guard_state_offset())); |
|
2705 __ addi(AT, AT, -JavaThread::stack_guard_yellow_disabled); |
|
2706 __ beq(AT, R0, reguard); |
|
2707 __ delayed()->nop(); |
|
2708 // slow path reguard re-enters here |
|
2709 __ bind(reguard_done); |
|
2710 |
|
2711 // Handle possible exception (will unlock if necessary) |
|
2712 |
|
2713 // native result if any is live |
|
2714 |
|
2715 // Unlock |
|
2716 Label slow_path_unlock; |
|
2717 Label unlock_done; |
|
2718 if (method->is_synchronized()) { |
|
2719 |
|
2720 Label done; |
|
2721 |
|
2722 // Get locked oop from the handle we passed to jni |
|
2723 __ ld( obj_reg, oop_handle_reg, 0); |
|
2724 //FIXME |
|
2725 if (UseBiasedLocking) { |
|
2726 __ biased_locking_exit(obj_reg, T8, done); |
|
2727 |
|
2728 } |
|
2729 |
|
2730 // Simple recursive lock? |
|
2731 |
|
2732 __ ld(AT, FP, lock_slot_ebp_offset); |
|
2733 __ beq(AT, R0, done); |
|
2734 __ delayed()->nop(); |
|
2735 // Must save eax if if it is live now because cmpxchg must use it |
|
2736 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
|
2737 save_native_result(masm, ret_type, stack_slots); |
|
2738 } |
|
2739 |
|
2740 // get old displaced header |
|
2741 __ ld (T8, FP, lock_slot_ebp_offset); |
|
2742 // get address of the stack lock |
|
2743 //FIXME aoqi |
|
2744 //__ addi (T6, FP, lock_slot_ebp_offset); |
|
2745 __ addi (c_rarg0, FP, lock_slot_ebp_offset); |
|
2746 // Atomic swap old header if oop still contains the stack lock |
|
2747 //FIXME aoqi |
|
2748 //__ cmpxchg(T8, Address(obj_reg, 0),T6 ); |
|
2749 __ cmpxchg(T8, Address(obj_reg, 0), c_rarg0); |
|
2750 |
|
2751 __ beq(AT, R0, slow_path_unlock); |
|
2752 __ delayed()->nop(); |
|
2753 // slow path re-enters here |
|
2754 __ bind(unlock_done); |
|
2755 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
|
2756 restore_native_result(masm, ret_type, stack_slots); |
|
2757 } |
|
2758 |
|
2759 __ bind(done); |
|
2760 |
|
2761 } |
|
2762 { |
|
2763 SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
|
2764 // Tell dtrace about this method exit |
|
2765 save_native_result(masm, ret_type, stack_slots); |
|
2766 int metadata_index = __ oop_recorder()->find_index( (method())); |
|
2767 RelocationHolder rspec = metadata_Relocation::spec(metadata_index); |
|
2768 __ relocate(rspec); |
|
2769 //__ lui(T6, Assembler::split_high((int)JNIHandles::make_local(method()))); |
|
2770 //__ addiu(T6, T6, Assembler::split_low((int)JNIHandles::make_local(method()))); |
|
2771 __ li48(AT, (long)(method())); |
|
2772 |
|
2773 __ call_VM_leaf( |
|
2774 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
|
2775 thread, AT); |
|
2776 restore_native_result(masm, ret_type, stack_slots); |
|
2777 } |
|
2778 |
|
2779 // We can finally stop using that last_Java_frame we setup ages ago |
|
2780 |
|
2781 __ reset_last_Java_frame(false, true); |
|
2782 |
|
2783 // Unpack oop result |
|
2784 if (ret_type == T_OBJECT || ret_type == T_ARRAY) { |
|
2785 Label L; |
|
2786 // __ cmpl(eax, NULL_WORD); |
|
2787 // __ jcc(Assembler::equal, L); |
|
2788 __ beq(V0, R0,L ); |
|
2789 __ delayed()->nop(); |
|
2790 // __ movl(eax, Address(eax)); |
|
2791 __ ld(V0, V0, 0); |
|
2792 __ bind(L); |
|
2793 // __ verify_oop(eax); |
|
2794 __ verify_oop(V0); |
|
2795 } |
|
2796 |
|
2797 if (!is_critical_native) { |
|
2798 // reset handle block |
|
2799 __ ld(AT, thread, in_bytes(JavaThread::active_handles_offset())); |
|
2800 __ sw(R0, AT, JNIHandleBlock::top_offset_in_bytes()); |
|
2801 } |
|
2802 |
|
2803 if (!is_critical_native) { |
|
2804 // Any exception pending? |
|
2805 __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
2806 |
|
2807 __ bne(AT, R0, exception_pending); |
|
2808 __ delayed()->nop(); |
|
2809 } |
|
2810 // no exception, we're almost done |
|
2811 |
|
2812 // check that only result value is on FPU stack |
|
2813 __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit"); |
|
2814 |
|
2815 // Fixup floating pointer results so that result looks like a return from a compiled method |
|
2816 /* if (ret_type == T_FLOAT) { |
|
2817 if (UseSSE >= 1) { |
|
2818 // Pop st0 and store as float and reload into xmm register |
|
2819 __ fstp_s(Address(ebp, -4)); |
|
2820 __ movss(xmm0, Address(ebp, -4)); |
|
2821 } |
|
2822 } else if (ret_type == T_DOUBLE) { |
|
2823 if (UseSSE >= 2) { |
|
2824 // Pop st0 and store as double and reload into xmm register |
|
2825 __ fstp_d(Address(ebp, -8)); |
|
2826 __ movsd(xmm0, Address(ebp, -8)); |
|
2827 } |
|
2828 } |
|
2829 */ |
|
2830 // Return |
|
2831 #ifndef OPT_THREAD |
|
2832 __ get_thread(TREG); |
|
2833 #endif |
|
2834 __ ld_ptr(SP, TREG, in_bytes(JavaThread::last_Java_sp_offset())); |
|
2835 __ leave(); |
|
2836 |
|
2837 __ jr(RA); |
|
2838 __ delayed()->nop(); |
|
2839 // Unexpected paths are out of line and go here |
|
2840 /* |
|
2841 if (!is_critical_native) { |
|
2842 // forward the exception |
|
2843 __ bind(exception_pending); |
|
2844 |
|
2845 // and forward the exception |
|
2846 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
|
2847 } |
|
2848 */ |
|
2849 // Slow path locking & unlocking |
|
2850 if (method->is_synchronized()) { |
|
2851 |
|
2852 // BEGIN Slow path lock |
|
2853 |
|
2854 __ bind(slow_path_lock); |
|
2855 |
|
2856 // protect the args we've loaded |
|
2857 save_args(masm, total_c_args, c_arg, out_regs); |
|
2858 |
|
2859 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM |
|
2860 // args are (oop obj, BasicLock* lock, JavaThread* thread) |
|
2861 |
|
2862 __ move(A0, obj_reg); |
|
2863 __ move(A1, lock_reg); |
|
2864 __ move(A2, thread); |
|
2865 __ addi(SP, SP, - 3*wordSize); |
|
2866 |
|
2867 __ move(AT, -(StackAlignmentInBytes)); |
|
2868 __ move(S2, SP); // use S2 as a sender SP holder |
|
2869 __ andr(SP, SP, AT); // align stack as required by ABI |
|
2870 |
|
2871 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), relocInfo::runtime_call_type); |
|
2872 __ delayed()->nop(); |
|
2873 __ move(SP, S2); |
|
2874 __ addi(SP, SP, 3*wordSize); |
|
2875 |
|
2876 restore_args(masm, total_c_args, c_arg, out_regs); |
|
2877 |
|
2878 #ifdef ASSERT |
|
2879 { Label L; |
|
2880 // __ cmpl(Address(thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD); |
|
2881 __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
2882 //__ jcc(Assembler::equal, L); |
|
2883 __ beq(AT, R0, L); |
|
2884 __ delayed()->nop(); |
|
2885 __ stop("no pending exception allowed on exit from monitorenter"); |
|
2886 __ bind(L); |
|
2887 } |
|
2888 #endif |
|
2889 __ b(lock_done); |
|
2890 __ delayed()->nop(); |
|
2891 // END Slow path lock |
|
2892 |
|
2893 // BEGIN Slow path unlock |
|
2894 __ bind(slow_path_unlock); |
|
2895 |
|
2896 // Slow path unlock |
|
2897 |
|
2898 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
|
2899 save_native_result(masm, ret_type, stack_slots); |
|
2900 } |
|
2901 // Save pending exception around call to VM (which contains an EXCEPTION_MARK) |
|
2902 |
|
2903 __ ld(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
2904 __ push(AT); |
|
2905 __ sd(R0, thread, in_bytes(Thread::pending_exception_offset())); |
|
2906 |
|
2907 __ move(AT, -(StackAlignmentInBytes)); |
|
2908 __ move(S2, SP); // use S2 as a sender SP holder |
|
2909 __ andr(SP, SP, AT); // align stack as required by ABI |
|
2910 |
|
2911 // should be a peal |
|
2912 // +wordSize because of the push above |
|
2913 __ addi(A1, FP, lock_slot_ebp_offset); |
|
2914 |
|
2915 __ move(A0, obj_reg); |
|
2916 __ addi(SP,SP, -2*wordSize); |
|
2917 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C), |
|
2918 relocInfo::runtime_call_type); |
|
2919 __ delayed()->nop(); |
|
2920 __ addi(SP,SP, 2*wordSize); |
|
2921 __ move(SP, S2); |
|
2922 //add for compressedoops |
|
2923 __ reinit_heapbase(); |
|
2924 #ifdef ASSERT |
|
2925 { |
|
2926 Label L; |
|
2927 // __ cmpl(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD); |
|
2928 __ lw( AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
2929 //__ jcc(Assembler::equal, L); |
|
2930 __ beq(AT, R0, L); |
|
2931 __ delayed()->nop(); |
|
2932 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); |
|
2933 __ bind(L); |
|
2934 } |
|
2935 #endif /* ASSERT */ |
|
2936 |
|
2937 __ pop(AT); |
|
2938 __ sd(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
2939 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
|
2940 restore_native_result(masm, ret_type, stack_slots); |
|
2941 } |
|
2942 __ b(unlock_done); |
|
2943 __ delayed()->nop(); |
|
2944 // END Slow path unlock |
|
2945 |
|
2946 } |
|
2947 |
|
2948 // SLOW PATH Reguard the stack if needed |
|
2949 |
|
2950 __ bind(reguard); |
|
2951 save_native_result(masm, ret_type, stack_slots); |
|
2952 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), |
|
2953 relocInfo::runtime_call_type); |
|
2954 __ delayed()->nop(); |
|
2955 //add for compressedoops |
|
2956 __ reinit_heapbase(); |
|
2957 restore_native_result(masm, ret_type, stack_slots); |
|
2958 __ b(reguard_done); |
|
2959 __ delayed()->nop(); |
|
2960 |
|
2961 // BEGIN EXCEPTION PROCESSING |
|
2962 if (!is_critical_native) { |
|
2963 // Forward the exception |
|
2964 __ bind(exception_pending); |
|
2965 |
|
2966 // remove possible return value from FPU register stack |
|
2967 __ empty_FPU_stack(); |
|
2968 |
|
2969 // pop our frame |
|
2970 //forward_exception_entry need return address on stack |
|
2971 __ addiu(SP, FP, wordSize); |
|
2972 __ ld(FP, SP, (-1) * wordSize); |
|
2973 |
|
2974 // and forward the exception |
|
2975 __ jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); |
|
2976 __ delayed()->nop(); |
|
2977 } |
|
2978 __ flush(); |
|
2979 |
|
2980 nmethod *nm = nmethod::new_native_nmethod(method, |
|
2981 compile_id, |
|
2982 masm->code(), |
|
2983 vep_offset, |
|
2984 frame_complete, |
|
2985 stack_slots / VMRegImpl::slots_per_word, |
|
2986 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), |
|
2987 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), |
|
2988 oop_maps); |
|
2989 |
|
2990 if (is_critical_native) { |
|
2991 nm->set_lazy_critical_native(true); |
|
2992 } |
|
2993 return nm; |
|
2994 |
|
2995 |
|
2996 } |
|
2997 |
|
2998 #ifdef HAVE_DTRACE_H |
|
2999 // --------------------------------------------------------------------------- |
|
3000 // Generate a dtrace nmethod for a given signature. The method takes arguments |
|
3001 // in the Java compiled code convention, marshals them to the native |
|
3002 // abi and then leaves nops at the position you would expect to call a native |
|
3003 // function. When the probe is enabled the nops are replaced with a trap |
|
3004 // instruction that dtrace inserts and the trace will cause a notification |
|
3005 // to dtrace. |
|
3006 // |
|
3007 // The probes are only able to take primitive types and java/lang/String as |
|
3008 // arguments. No other java types are allowed. Strings are converted to utf8 |
|
3009 // strings so that from dtrace point of view java strings are converted to C |
|
3010 // strings. There is an arbitrary fixed limit on the total space that a method |
|
3011 // can use for converting the strings. (256 chars per string in the signature). |
|
3012 // So any java string larger then this is truncated. |
|
3013 |
|
3014 static int fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 }; |
|
3015 static bool offsets_initialized = false; |
|
3016 |
|
3017 static VMRegPair reg64_to_VMRegPair(Register r) { |
|
3018 VMRegPair ret; |
|
3019 if (wordSize == 8) { |
|
3020 ret.set2(r->as_VMReg()); |
|
3021 } else { |
|
3022 ret.set_pair(r->successor()->as_VMReg(), r->as_VMReg()); |
|
3023 } |
|
3024 return ret; |
|
3025 } |
|
3026 |
|
3027 |
|
3028 nmethod *SharedRuntime::generate_dtrace_nmethod( |
|
3029 MacroAssembler *masm, methodHandle method) { |
|
3030 |
|
3031 |
|
3032 // generate_dtrace_nmethod is guarded by a mutex so we are sure to |
|
3033 // be single threaded in this method. |
|
3034 assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); |
|
3035 |
|
3036 // Fill in the signature array, for the calling-convention call. |
|
3037 int total_args_passed = method->size_of_parameters(); |
|
3038 |
|
3039 BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); |
|
3040 VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); |
|
3041 |
|
3042 // The signature we are going to use for the trap that dtrace will see |
|
3043 // java/lang/String is converted. We drop "this" and any other object |
|
3044 // is converted to NULL. (A one-slot java/lang/Long object reference |
|
3045 // is converted to a two-slot long, which is why we double the allocation). |
|
3046 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); |
|
3047 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); |
|
3048 |
|
3049 int i=0; |
|
3050 int total_strings = 0; |
|
3051 int first_arg_to_pass = 0; |
|
3052 int total_c_args = 0; |
|
3053 |
|
3054 // Skip the receiver as dtrace doesn't want to see it |
|
3055 if( !method->is_static() ) { |
|
3056 in_sig_bt[i++] = T_OBJECT; |
|
3057 first_arg_to_pass = 1; |
|
3058 } |
|
3059 |
|
3060 SignatureStream ss(method->signature()); |
|
3061 for ( ; !ss.at_return_type(); ss.next()) { |
|
3062 BasicType bt = ss.type(); |
|
3063 in_sig_bt[i++] = bt; // Collect remaining bits of signature |
|
3064 out_sig_bt[total_c_args++] = bt; |
|
3065 if( bt == T_OBJECT) { |
|
3066 symbolOop s = ss.as_symbol_or_null(); |
|
3067 if (s == vmSymbols::java_lang_String()) { |
|
3068 total_strings++; |
|
3069 out_sig_bt[total_c_args-1] = T_ADDRESS; |
|
3070 } else if (s == vmSymbols::java_lang_Boolean() || |
|
3071 s == vmSymbols::java_lang_Byte()) { |
|
3072 out_sig_bt[total_c_args-1] = T_BYTE; |
|
3073 } else if (s == vmSymbols::java_lang_Character() || |
|
3074 s == vmSymbols::java_lang_Short()) { |
|
3075 out_sig_bt[total_c_args-1] = T_SHORT; |
|
3076 } else if (s == vmSymbols::java_lang_Integer() || |
|
3077 s == vmSymbols::java_lang_Float()) { |
|
3078 out_sig_bt[total_c_args-1] = T_INT; |
|
3079 } else if (s == vmSymbols::java_lang_Long() || |
|
3080 s == vmSymbols::java_lang_Double()) { |
|
3081 out_sig_bt[total_c_args-1] = T_LONG; |
|
3082 out_sig_bt[total_c_args++] = T_VOID; |
|
3083 } |
|
3084 } else if ( bt == T_LONG || bt == T_DOUBLE ) { |
|
3085 in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots |
|
3086 // We convert double to long |
|
3087 out_sig_bt[total_c_args-1] = T_LONG; |
|
3088 out_sig_bt[total_c_args++] = T_VOID; |
|
3089 } else if ( bt == T_FLOAT) { |
|
3090 // We convert float to int |
|
3091 out_sig_bt[total_c_args-1] = T_INT; |
|
3092 } |
|
3093 } |
|
3094 |
|
3095 assert(i==total_args_passed, "validly parsed signature"); |
|
3096 |
|
3097 // Now get the compiled-Java layout as input arguments |
|
3098 int comp_args_on_stack; |
|
3099 comp_args_on_stack = SharedRuntime::java_calling_convention( |
|
3100 in_sig_bt, in_regs, total_args_passed, false); |
|
3101 |
|
3102 // We have received a description of where all the java arg are located |
|
3103 // on entry to the wrapper. We need to convert these args to where |
|
3104 // the a native (non-jni) function would expect them. To figure out |
|
3105 // where they go we convert the java signature to a C signature and remove |
|
3106 // T_VOID for any long/double we might have received. |
|
3107 |
|
3108 |
|
3109 // Now figure out where the args must be stored and how much stack space |
|
3110 // they require (neglecting out_preserve_stack_slots but space for storing |
|
3111 // the 1st six register arguments). It's weird see int_stk_helper. |
|
3112 // |
|
3113 int out_arg_slots; |
|
3114 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args); |
|
3115 |
|
3116 // Calculate the total number of stack slots we will need. |
|
3117 |
|
3118 // First count the abi requirement plus all of the outgoing args |
|
3119 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
|
3120 |
|
3121 // Plus a temp for possible converion of float/double/long register args |
|
3122 |
|
3123 int conversion_temp = stack_slots; |
|
3124 stack_slots += 2; |
|
3125 |
|
3126 |
|
3127 // Now space for the string(s) we must convert |
|
3128 |
|
3129 int string_locs = stack_slots; |
|
3130 stack_slots += total_strings * |
|
3131 (max_dtrace_string_size / VMRegImpl::stack_slot_size); |
|
3132 |
|
3133 // Ok The space we have allocated will look like: |
|
3134 // |
|
3135 // |
|
3136 // FP-> | | |
|
3137 // |---------------------| |
|
3138 // | string[n] | |
|
3139 // |---------------------| <- string_locs[n] |
|
3140 // | string[n-1] | |
|
3141 // |---------------------| <- string_locs[n-1] |
|
3142 // | ... | |
|
3143 // | ... | |
|
3144 // |---------------------| <- string_locs[1] |
|
3145 // | string[0] | |
|
3146 // |---------------------| <- string_locs[0] |
|
3147 // | temp | |
|
3148 // |---------------------| <- conversion_temp |
|
3149 // | outbound memory | |
|
3150 // | based arguments | |
|
3151 // | | |
|
3152 // |---------------------| |
|
3153 // | | |
|
3154 // SP-> | out_preserved_slots | |
|
3155 // |
|
3156 // |
|
3157 |
|
3158 // Now compute actual number of stack words we need rounding to make |
|
3159 // stack properly aligned. |
|
3160 stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word); |
|
3161 |
|
3162 int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
|
3163 |
|
3164 intptr_t start = (intptr_t)__ pc(); |
|
3165 |
|
3166 // First thing make an ic check to see if we should even be here |
|
3167 |
|
3168 { |
|
3169 Label L; |
|
3170 const Register temp_reg = G3_scratch; |
|
3171 Address ic_miss(temp_reg, SharedRuntime::get_ic_miss_stub()); |
|
3172 __ verify_oop(O0); |
|
3173 __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), temp_reg); |
|
3174 __ cmp(temp_reg, G5_inline_cache_reg); |
|
3175 __ brx(Assembler::equal, true, Assembler::pt, L); |
|
3176 __ delayed()->nop(); |
|
3177 |
|
3178 __ jump_to(ic_miss, 0); |
|
3179 __ delayed()->nop(); |
|
3180 __ align(CodeEntryAlignment); |
|
3181 __ bind(L); |
|
3182 } |
|
3183 |
|
3184 int vep_offset = ((intptr_t)__ pc()) - start; |
|
3185 |
|
3186 |
|
3187 // The instruction at the verified entry point must be 5 bytes or longer |
|
3188 // because it can be patched on the fly by make_non_entrant. The stack bang |
|
3189 // instruction fits that requirement. |
|
3190 |
|
3191 // Generate stack overflow check before creating frame |
|
3192 __ generate_stack_overflow_check(stack_size); |
|
3193 |
|
3194 assert(((intptr_t)__ pc() - start - vep_offset) >= 5, |
|
3195 "valid size for make_non_entrant"); |
|
3196 |
|
3197 // Generate a new frame for the wrapper. |
|
3198 __ save(SP, -stack_size, SP); |
|
3199 |
|
3200 // Frame is now completed as far a size and linkage. |
|
3201 |
|
3202 int frame_complete = ((intptr_t)__ pc()) - start; |
|
3203 |
|
3204 #ifdef ASSERT |
|
3205 bool reg_destroyed[RegisterImpl::number_of_registers]; |
|
3206 bool freg_destroyed[FloatRegisterImpl::number_of_registers]; |
|
3207 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { |
|
3208 reg_destroyed[r] = false; |
|
3209 } |
|
3210 for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) { |
|
3211 freg_destroyed[f] = false; |
|
3212 } |
|
3213 |
|
3214 #endif /* ASSERT */ |
|
3215 |
|
3216 VMRegPair zero; |
|
3217 const Register g0 = G0; // without this we get a compiler warning (why??) |
|
3218 zero.set2(g0->as_VMReg()); |
|
3219 |
|
3220 int c_arg, j_arg; |
|
3221 |
|
3222 Register conversion_off = noreg; |
|
3223 |
|
3224 for (j_arg = first_arg_to_pass, c_arg = 0 ; |
|
3225 j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
|
3226 |
|
3227 VMRegPair src = in_regs[j_arg]; |
|
3228 VMRegPair dst = out_regs[c_arg]; |
|
3229 |
|
3230 #ifdef ASSERT |
|
3231 if (src.first()->is_Register()) { |
|
3232 assert(!reg_destroyed[src.first()->as_Register()->encoding()], "ack!"); |
|
3233 } else if (src.first()->is_FloatRegister()) { |
|
3234 assert(!freg_destroyed[src.first()->as_FloatRegister()->encoding( |
|
3235 FloatRegisterImpl::S)], "ack!"); |
|
3236 } |
|
3237 if (dst.first()->is_Register()) { |
|
3238 reg_destroyed[dst.first()->as_Register()->encoding()] = true; |
|
3239 } else if (dst.first()->is_FloatRegister()) { |
|
3240 freg_destroyed[dst.first()->as_FloatRegister()->encoding( |
|
3241 FloatRegisterImpl::S)] = true; |
|
3242 } |
|
3243 #endif /* ASSERT */ |
|
3244 |
|
3245 switch (in_sig_bt[j_arg]) { |
|
3246 case T_ARRAY: |
|
3247 case T_OBJECT: |
|
3248 { |
|
3249 if (out_sig_bt[c_arg] == T_BYTE || out_sig_bt[c_arg] == T_SHORT || |
|
3250 out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { |
|
3251 // need to unbox a one-slot value |
|
3252 Register in_reg = L0; |
|
3253 Register tmp = L2; |
|
3254 if ( src.first()->is_reg() ) { |
|
3255 in_reg = src.first()->as_Register(); |
|
3256 } else { |
|
3257 assert(Assembler::is_simm13(reg2offset(src.first()) + STACK_BIAS), |
|
3258 "must be"); |
|
3259 __ ld_ptr(FP, reg2offset(src.first()) + STACK_BIAS, in_reg); |
|
3260 } |
|
3261 // If the final destination is an acceptable register |
|
3262 if ( dst.first()->is_reg() ) { |
|
3263 if ( dst.is_single_phys_reg() || out_sig_bt[c_arg] != T_LONG ) { |
|
3264 tmp = dst.first()->as_Register(); |
|
3265 } |
|
3266 } |
|
3267 |
|
3268 Label skipUnbox; |
|
3269 if ( wordSize == 4 && out_sig_bt[c_arg] == T_LONG ) { |
|
3270 __ mov(G0, tmp->successor()); |
|
3271 } |
|
3272 __ br_null(in_reg, true, Assembler::pn, skipUnbox); |
|
3273 __ delayed()->mov(G0, tmp); |
|
3274 |
|
3275 BasicType bt = out_sig_bt[c_arg]; |
|
3276 int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); |
|
3277 switch (bt) { |
|
3278 case T_BYTE: |
|
3279 __ ldub(in_reg, box_offset, tmp); break; |
|
3280 case T_SHORT: |
|
3281 __ lduh(in_reg, box_offset, tmp); break; |
|
3282 case T_INT: |
|
3283 __ ld(in_reg, box_offset, tmp); break; |
|
3284 case T_LONG: |
|
3285 __ ld_long(in_reg, box_offset, tmp); break; |
|
3286 default: ShouldNotReachHere(); |
|
3287 } |
|
3288 |
|
3289 __ bind(skipUnbox); |
|
3290 // If tmp wasn't final destination copy to final destination |
|
3291 if (tmp == L2) { |
|
3292 VMRegPair tmp_as_VM = reg64_to_VMRegPair(L2); |
|
3293 if (out_sig_bt[c_arg] == T_LONG) { |
|
3294 long_move(masm, tmp_as_VM, dst); |
|
3295 } else { |
|
3296 move32_64(masm, tmp_as_VM, out_regs[c_arg]); |
|
3297 } |
|
3298 } |
|
3299 if (out_sig_bt[c_arg] == T_LONG) { |
|
3300 assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
|
3301 ++c_arg; // move over the T_VOID to keep the loop indices in sync |
|
3302 } |
|
3303 } else if (out_sig_bt[c_arg] == T_ADDRESS) { |
|
3304 Register s = |
|
3305 src.first()->is_reg() ? src.first()->as_Register() : L2; |
|
3306 Register d = |
|
3307 dst.first()->is_reg() ? dst.first()->as_Register() : L2; |
|
3308 |
|
3309 // We store the oop now so that the conversion pass can reach |
|
3310 // while in the inner frame. This will be the only store if |
|
3311 // the oop is NULL. |
|
3312 if (s != L2) { |
|
3313 // src is register |
|
3314 if (d != L2) { |
|
3315 // dst is register |
|
3316 __ mov(s, d); |
|
3317 } else { |
|
3318 assert(Assembler::is_simm13(reg2offset(dst.first()) + |
|
3319 STACK_BIAS), "must be"); |
|
3320 __ st_ptr(s, SP, reg2offset(dst.first()) + STACK_BIAS); |
|
3321 } |
|
3322 } else { |
|
3323 // src not a register |
|
3324 assert(Assembler::is_simm13(reg2offset(src.first()) + |
|
3325 STACK_BIAS), "must be"); |
|
3326 __ ld_ptr(FP, reg2offset(src.first()) + STACK_BIAS, d); |
|
3327 if (d == L2) { |
|
3328 assert(Assembler::is_simm13(reg2offset(dst.first()) + |
|
3329 STACK_BIAS), "must be"); |
|
3330 __ st_ptr(d, SP, reg2offset(dst.first()) + STACK_BIAS); |
|
3331 } |
|
3332 } |
|
3333 } else if (out_sig_bt[c_arg] != T_VOID) { |
|
3334 // Convert the arg to NULL |
|
3335 if (dst.first()->is_reg()) { |
|
3336 __ mov(G0, dst.first()->as_Register()); |
|
3337 } else { |
|
3338 assert(Assembler::is_simm13(reg2offset(dst.first()) + |
|
3339 STACK_BIAS), "must be"); |
|
3340 __ st_ptr(G0, SP, reg2offset(dst.first()) + STACK_BIAS); |
|
3341 } |
|
3342 } |
|
3343 } |
|
3344 break; |
|
3345 case T_VOID: |
|
3346 break; |
|
3347 |
|
3348 case T_FLOAT: |
|
3349 if (src.first()->is_stack()) { |
|
3350 // Stack to stack/reg is simple |
|
3351 move32_64(masm, src, dst); |
|
3352 } else { |
|
3353 if (dst.first()->is_reg()) { |
|
3354 // freg -> reg |
|
3355 int off = |
|
3356 STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
|
3357 Register d = dst.first()->as_Register(); |
|
3358 if (Assembler::is_simm13(off)) { |
|
3359 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
|
3360 SP, off); |
|
3361 __ ld(SP, off, d); |
|
3362 } else { |
|
3363 if (conversion_off == noreg) { |
|
3364 __ set(off, L6); |
|
3365 conversion_off = L6; |
|
3366 } |
|
3367 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
|
3368 SP, conversion_off); |
|
3369 __ ld(SP, conversion_off , d); |
|
3370 } |
|
3371 } else { |
|
3372 // freg -> mem |
|
3373 int off = STACK_BIAS + reg2offset(dst.first()); |
|
3374 if (Assembler::is_simm13(off)) { |
|
3375 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
|
3376 SP, off); |
|
3377 } else { |
|
3378 if (conversion_off == noreg) { |
|
3379 __ set(off, L6); |
|
3380 conversion_off = L6; |
|
3381 } |
|
3382 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), |
|
3383 SP, conversion_off); |
|
3384 } |
|
3385 } |
|
3386 } |
|
3387 break; |
|
3388 |
|
3389 case T_DOUBLE: |
|
3390 assert( j_arg + 1 < total_args_passed && |
|
3391 in_sig_bt[j_arg + 1] == T_VOID && |
|
3392 out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
|
3393 if (src.first()->is_stack()) { |
|
3394 // Stack to stack/reg is simple |
|
3395 long_move(masm, src, dst); |
|
3396 } else { |
|
3397 Register d = dst.first()->is_reg() ? dst.first()->as_Register() : L2; |
|
3398 |
|
3399 // Destination could be an odd reg on 32bit in which case |
|
3400 // we can't load direct to the destination. |
|
3401 |
|
3402 if (!d->is_even() && wordSize == 4) { |
|
3403 d = L2; |
|
3404 } |
|
3405 int off = STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
|
3406 if (Assembler::is_simm13(off)) { |
|
3407 __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), |
|
3408 SP, off); |
|
3409 __ ld_long(SP, off, d); |
|
3410 } else { |
|
3411 if (conversion_off == noreg) { |
|
3412 __ set(off, L6); |
|
3413 conversion_off = L6; |
|
3414 } |
|
3415 __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), |
|
3416 SP, conversion_off); |
|
3417 __ ld_long(SP, conversion_off, d); |
|
3418 } |
|
3419 if (d == L2) { |
|
3420 long_move(masm, reg64_to_VMRegPair(L2), dst); |
|
3421 } |
|
3422 } |
|
3423 break; |
|
3424 |
|
3425 case T_LONG : |
|
3426 // 32bit can't do a split move of something like g1 -> O0, O1 |
|
3427 // so use a memory temp |
|
3428 if (src.is_single_phys_reg() && wordSize == 4) { |
|
3429 Register tmp = L2; |
|
3430 if (dst.first()->is_reg() && |
|
3431 (wordSize == 8 || dst.first()->as_Register()->is_even())) { |
|
3432 tmp = dst.first()->as_Register(); |
|
3433 } |
|
3434 |
|
3435 int off = STACK_BIAS + conversion_temp * VMRegImpl::stack_slot_size; |
|
3436 if (Assembler::is_simm13(off)) { |
|
3437 __ stx(src.first()->as_Register(), SP, off); |
|
3438 __ ld_long(SP, off, tmp); |
|
3439 } else { |
|
3440 if (conversion_off == noreg) { |
|
3441 __ set(off, L6); |
|
3442 conversion_off = L6; |
|
3443 } |
|
3444 __ stx(src.first()->as_Register(), SP, conversion_off); |
|
3445 __ ld_long(SP, conversion_off, tmp); |
|
3446 } |
|
3447 |
|
3448 if (tmp == L2) { |
|
3449 long_move(masm, reg64_to_VMRegPair(L2), dst); |
|
3450 } |
|
3451 } else { |
|
3452 long_move(masm, src, dst); |
|
3453 } |
|
3454 break; |
|
3455 |
|
3456 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
|
3457 |
|
3458 default: |
|
3459 move32_64(masm, src, dst); |
|
3460 } |
|
3461 } |
|
3462 |
|
3463 |
|
3464 // If we have any strings we must store any register based arg to the stack |
|
3465 // This includes any still live xmm registers too. |
|
3466 |
|
3467 if (total_strings > 0 ) { |
|
3468 |
|
3469 // protect all the arg registers |
|
3470 __ save_frame(0); |
|
3471 __ mov(G2_thread, L7_thread_cache); |
|
3472 const Register L2_string_off = L2; |
|
3473 |
|
3474 // Get first string offset |
|
3475 __ set(string_locs * VMRegImpl::stack_slot_size, L2_string_off); |
|
3476 |
|
3477 for (c_arg = 0 ; c_arg < total_c_args ; c_arg++ ) { |
|
3478 if (out_sig_bt[c_arg] == T_ADDRESS) { |
|
3479 |
|
3480 VMRegPair dst = out_regs[c_arg]; |
|
3481 const Register d = dst.first()->is_reg() ? |
|
3482 dst.first()->as_Register()->after_save() : noreg; |
|
3483 |
|
3484 // It's a string the oop and it was already copied to the out arg |
|
3485 // position |
|
3486 if (d != noreg) { |
|
3487 __ mov(d, O0); |
|
3488 } else { |
|
3489 assert(Assembler::is_simm13(reg2offset(dst.first()) + STACK_BIAS), |
|
3490 "must be"); |
|
3491 __ ld_ptr(FP, reg2offset(dst.first()) + STACK_BIAS, O0); |
|
3492 } |
|
3493 Label skip; |
|
3494 |
|
3495 __ br_null(O0, false, Assembler::pn, skip); |
|
3496 __ delayed()->add(FP, L2_string_off, O1); |
|
3497 |
|
3498 if (d != noreg) { |
|
3499 __ mov(O1, d); |
|
3500 } else { |
|
3501 assert(Assembler::is_simm13(reg2offset(dst.first()) + STACK_BIAS), |
|
3502 "must be"); |
|
3503 __ st_ptr(O1, FP, reg2offset(dst.first()) + STACK_BIAS); |
|
3504 } |
|
3505 |
|
3506 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::get_utf), |
|
3507 relocInfo::runtime_call_type); |
|
3508 __ delayed()->add(L2_string_off, max_dtrace_string_size, L2_string_off); |
|
3509 |
|
3510 __ bind(skip); |
|
3511 |
|
3512 } |
|
3513 |
|
3514 } |
|
3515 __ mov(L7_thread_cache, G2_thread); |
|
3516 __ restore(); |
|
3517 |
|
3518 } |
|
3519 |
|
3520 |
|
3521 // Ok now we are done. Need to place the nop that dtrace wants in order to |
|
3522 // patch in the trap |
|
3523 |
|
3524 int patch_offset = ((intptr_t)__ pc()) - start; |
|
3525 |
|
3526 __ nop(); |
|
3527 |
|
3528 |
|
3529 // Return |
|
3530 |
|
3531 __ ret(); |
|
3532 __ delayed()->restore(); |
|
3533 |
|
3534 __ flush(); |
|
3535 |
|
3536 nmethod *nm = nmethod::new_dtrace_nmethod( |
|
3537 method, masm->code(), vep_offset, patch_offset, frame_complete, |
|
3538 stack_slots / VMRegImpl::slots_per_word); |
|
3539 return nm; |
|
3540 |
|
3541 } |
|
3542 |
|
3543 #endif // HAVE_DTRACE_H |
|
3544 |
|
3545 // this function returns the adjust size (in number of words) to a c2i adapter |
|
3546 // activation for use during deoptimization |
|
3547 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) { |
|
3548 return (callee_locals - callee_parameters) * Interpreter::stackElementWords; |
|
3549 } |
|
3550 |
|
3551 // "Top of Stack" slots that may be unused by the calling convention but must |
|
3552 // otherwise be preserved. |
|
3553 // On Intel these are not necessary and the value can be zero. |
|
3554 // On Sparc this describes the words reserved for storing a register window |
|
3555 // when an interrupt occurs. |
|
3556 uint SharedRuntime::out_preserve_stack_slots() { |
|
3557 //return frame::register_save_words * VMRegImpl::slots_per_word; |
|
3558 return 0; |
|
3559 } |
|
3560 /* |
|
3561 static void gen_new_frame(MacroAssembler* masm, bool deopt) { |
|
3562 // |
|
3563 // Common out the new frame generation for deopt and uncommon trap |
|
3564 // |
|
3565 Register G3pcs = G3_scratch; // Array of new pcs (input) |
|
3566 Register Oreturn0 = O0; |
|
3567 Register Oreturn1 = O1; |
|
3568 Register O2UnrollBlock = O2; |
|
3569 Register O3array = O3; // Array of frame sizes (input) |
|
3570 Register O4array_size = O4; // number of frames (input) |
|
3571 Register O7frame_size = O7; // number of frames (input) |
|
3572 |
|
3573 __ ld_ptr(O3array, 0, O7frame_size); |
|
3574 __ sub(G0, O7frame_size, O7frame_size); |
|
3575 __ save(SP, O7frame_size, SP); |
|
3576 __ ld_ptr(G3pcs, 0, I7); // load frame's new pc |
|
3577 |
|
3578 #ifdef ASSERT |
|
3579 // make sure that the frames are aligned properly |
|
3580 #ifndef _LP64 |
|
3581 __ btst(wordSize*2-1, SP); |
|
3582 __ breakpoint_trap(Assembler::notZero); |
|
3583 #endif |
|
3584 #endif |
|
3585 |
|
3586 // Deopt needs to pass some extra live values from frame to frame |
|
3587 |
|
3588 if (deopt) { |
|
3589 __ mov(Oreturn0->after_save(), Oreturn0); |
|
3590 __ mov(Oreturn1->after_save(), Oreturn1); |
|
3591 } |
|
3592 |
|
3593 __ mov(O4array_size->after_save(), O4array_size); |
|
3594 __ sub(O4array_size, 1, O4array_size); |
|
3595 __ mov(O3array->after_save(), O3array); |
|
3596 __ mov(O2UnrollBlock->after_save(), O2UnrollBlock); |
|
3597 __ add(G3pcs, wordSize, G3pcs); // point to next pc value |
|
3598 |
|
3599 #ifdef ASSERT |
|
3600 // trash registers to show a clear pattern in backtraces |
|
3601 __ set(0xDEAD0000, I0); |
|
3602 __ add(I0, 2, I1); |
|
3603 __ add(I0, 4, I2); |
|
3604 __ add(I0, 6, I3); |
|
3605 __ add(I0, 8, I4); |
|
3606 // Don't touch I5 could have valuable savedSP |
|
3607 __ set(0xDEADBEEF, L0); |
|
3608 __ mov(L0, L1); |
|
3609 __ mov(L0, L2); |
|
3610 __ mov(L0, L3); |
|
3611 __ mov(L0, L4); |
|
3612 __ mov(L0, L5); |
|
3613 |
|
3614 // trash the return value as there is nothing to return yet |
|
3615 __ set(0xDEAD0001, O7); |
|
3616 #endif |
|
3617 |
|
3618 __ mov(SP, O5_savedSP); |
|
3619 } |
|
3620 |
|
3621 |
|
3622 static void make_new_frames(MacroAssembler* masm, bool deopt) { |
|
3623 // |
|
3624 // loop through the UnrollBlock info and create new frames |
|
3625 // |
|
3626 Register G3pcs = G3_scratch; |
|
3627 Register Oreturn0 = O0; |
|
3628 Register Oreturn1 = O1; |
|
3629 Register O2UnrollBlock = O2; |
|
3630 Register O3array = O3; |
|
3631 Register O4array_size = O4; |
|
3632 Label loop; |
|
3633 |
|
3634 // Before we make new frames, check to see if stack is available. |
|
3635 // Do this after the caller's return address is on top of stack |
|
3636 if (UseStackBanging) { |
|
3637 // Get total frame size for interpreted frames |
|
3638 __ ld(Address(O2UnrollBlock, 0, |
|
3639 Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()), O4); |
|
3640 __ bang_stack_size(O4, O3, G3_scratch); |
|
3641 } |
|
3642 |
|
3643 __ ld(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()), O4array_size); |
|
3644 __ ld_ptr(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()), G3pcs); |
|
3645 |
|
3646 __ ld_ptr(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()), O3array); |
|
3647 |
|
3648 // Adjust old interpreter frame to make space for new frame's extra java locals |
|
3649 // |
|
3650 // We capture the original sp for the transition frame only because it is needed in |
|
3651 // order to properly calculate interpreter_sp_adjustment. Even though in real life |
|
3652 // every interpreter frame captures a savedSP it is only needed at the transition |
|
3653 // (fortunately). If we had to have it correct everywhere then we would need to |
|
3654 // be told the sp_adjustment for each frame we create. If the frame size array |
|
3655 // were to have twice the frame count entries then we could have pairs [sp_adjustment, frame_size] |
|
3656 // for each frame we create and keep up the illusion every where. |
|
3657 // |
|
3658 |
|
3659 __ ld(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()), O7); |
|
3660 __ mov(SP, O5_savedSP); // remember initial sender's original sp before adjustment |
|
3661 __ sub(SP, O7, SP); |
|
3662 |
|
3663 #ifdef ASSERT |
|
3664 // make sure that there is at least one entry in the array |
|
3665 __ tst(O4array_size); |
|
3666 __ breakpoint_trap(Assembler::zero); |
|
3667 #endif |
|
3668 |
|
3669 // Now push the new interpreter frames |
|
3670 __ bind(loop); |
|
3671 |
|
3672 // allocate a new frame, filling the registers |
|
3673 |
|
3674 gen_new_frame(masm, deopt); // allocate an interpreter frame |
|
3675 |
|
3676 __ tst(O4array_size); |
|
3677 __ br(Assembler::notZero, false, Assembler::pn, loop); |
|
3678 __ delayed()->add(O3array, wordSize, O3array); |
|
3679 __ ld_ptr(G3pcs, 0, O7); // load final frame new pc |
|
3680 |
|
3681 } |
|
3682 */ |
|
3683 |
|
3684 //------------------------------generate_deopt_blob---------------------------- |
|
3685 // Ought to generate an ideal graph & compile, but here's some SPARC ASM |
|
3686 // instead. |
|
3687 void SharedRuntime::generate_deopt_blob() { |
|
3688 // allocate space for the code |
|
3689 ResourceMark rm; |
|
3690 // setup code generation tools |
|
3691 //CodeBuffer buffer ("deopt_blob", 4000, 2048); |
|
3692 CodeBuffer buffer ("deopt_blob", 8000, 2048);//aoqi FIXME for debug |
|
3693 MacroAssembler* masm = new MacroAssembler( & buffer); |
|
3694 int frame_size_in_words; |
|
3695 OopMap* map = NULL; |
|
3696 // Account for the extra args we place on the stack |
|
3697 // by the time we call fetch_unroll_info |
|
3698 const int additional_words = 2; // deopt kind, thread |
|
3699 |
|
3700 OopMapSet *oop_maps = new OopMapSet(); |
|
3701 |
|
3702 address start = __ pc(); |
|
3703 Label cont; |
|
3704 // we use S3 for DeOpt reason register |
|
3705 Register reason = S3; |
|
3706 // use S6 for thread register |
|
3707 Register thread = TREG; |
|
3708 // use S7 for fetch_unroll_info returned UnrollBlock |
|
3709 Register unroll = S7; |
|
3710 // Prolog for non exception case! |
|
3711 // Correct the return address we were given. |
|
3712 //FIXME, return address is on the tos or Ra? |
|
3713 __ addi(RA, RA, - (NativeCall::return_address_offset)); |
|
3714 // Save everything in sight. |
|
3715 map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
|
3716 // Normal deoptimization |
|
3717 __ move(reason, Deoptimization::Unpack_deopt); |
|
3718 __ b(cont); |
|
3719 __ delayed()->nop(); |
|
3720 |
|
3721 int reexecute_offset = __ pc() - start; |
|
3722 |
|
3723 // Reexecute case |
|
3724 // return address is the pc describes what bci to do re-execute at |
|
3725 |
|
3726 // No need to update map as each call to save_live_registers will produce identical oopmap |
|
3727 //__ addi(RA, RA, - (NativeCall::return_address_offset)); |
|
3728 (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
|
3729 __ move(reason, Deoptimization::Unpack_reexecute); |
|
3730 __ b(cont); |
|
3731 __ delayed()->nop(); |
|
3732 |
|
3733 int exception_offset = __ pc() - start; |
|
3734 // Prolog for exception case |
|
3735 |
|
3736 // all registers are dead at this entry point, except for eax and |
|
3737 // edx which contain the exception oop and exception pc |
|
3738 // respectively. Set them in TLS and fall thru to the |
|
3739 // unpack_with_exception_in_tls entry point. |
|
3740 |
|
3741 __ get_thread(thread); |
|
3742 __ st_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
|
3743 __ st_ptr(V0, thread, in_bytes(JavaThread::exception_oop_offset())); |
|
3744 int exception_in_tls_offset = __ pc() - start; |
|
3745 // new implementation because exception oop is now passed in JavaThread |
|
3746 |
|
3747 // Prolog for exception case |
|
3748 // All registers must be preserved because they might be used by LinearScan |
|
3749 // Exceptiop oop and throwing PC are passed in JavaThread |
|
3750 // tos: stack at point of call to method that threw the exception (i.e. only |
|
3751 // args are on the stack, no return address) |
|
3752 |
|
3753 // Return address will be patched later with the throwing pc. The correct value is not |
|
3754 // available now because loading it from memory would destroy registers. |
|
3755 // Save everything in sight. |
|
3756 // No need to update map as each call to save_live_registers will produce identical oopmap |
|
3757 __ addi(RA, RA, - (NativeCall::return_address_offset)); |
|
3758 (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words); |
|
3759 |
|
3760 // Now it is safe to overwrite any register |
|
3761 // store the correct deoptimization type |
|
3762 __ move(reason, Deoptimization::Unpack_exception); |
|
3763 // load throwing pc from JavaThread and patch it as the return address |
|
3764 // of the current frame. Then clear the field in JavaThread |
|
3765 __ get_thread(thread); |
|
3766 __ ld_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
|
3767 __ st_ptr(V1, SP, RegisterSaver::raOffset() * wordSize); //save ra |
|
3768 __ st_ptr(R0, thread, in_bytes(JavaThread::exception_pc_offset())); |
|
3769 |
|
3770 |
|
3771 #ifdef ASSERT |
|
3772 // verify that there is really an exception oop in JavaThread |
|
3773 __ ld_ptr(AT, thread, in_bytes(JavaThread::exception_oop_offset())); |
|
3774 __ verify_oop(AT); |
|
3775 // verify that there is no pending exception |
|
3776 Label no_pending_exception; |
|
3777 __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
3778 __ beq(AT, R0, no_pending_exception); |
|
3779 __ delayed()->nop(); |
|
3780 __ stop("must not have pending exception here"); |
|
3781 __ bind(no_pending_exception); |
|
3782 #endif |
|
3783 __ bind(cont); |
|
3784 // Compiled code leaves the floating point stack dirty, empty it. |
|
3785 __ empty_FPU_stack(); |
|
3786 |
|
3787 |
|
3788 // Call C code. Need thread and this frame, but NOT official VM entry |
|
3789 // crud. We cannot block on this call, no GC can happen. |
|
3790 #ifndef OPT_THREAD |
|
3791 __ get_thread(thread); |
|
3792 #endif |
|
3793 |
|
3794 /* |
|
3795 * |
|
3796 0x000000555bd82aec: dadd a0, s6, zero ; __ move(A0, thread); |
|
3797 0x000000555bd82af0: daddi sp, sp, 0xfffffff0 ; __ addi(SP, SP, -additional_words * wordSize); |
|
3798 0x000000555bd82af4: sd sp, 0x1c8(s6) ; __ set_last_Java_frame(thread, NOREG, NOREG, NULL); |
|
3799 0x000000555bd82af8: lui at, 0x0 ; __ li64(AT, save_pc); |
|
3800 0x000000555bd82afc: ori at, at, 0x55 |
|
3801 0x000000555bd82b00: dsll at, at, 16 |
|
3802 0x000000555bd82b04: ori at, at, 0x5bd8 |
|
3803 0x000000555bd82b08: dsll at, at, 16 |
|
3804 0x000000555bd82b0c: ori at, at, 0x2b34 ; save_pc = pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 4 |
|
3805 0x000000555bd82b10: sd at, 0x1d0(s6) |
|
3806 0x000000555bd82b14: lui t9, 0x0 |
|
3807 0x000000555bd82b18: ori t9, t9, 0x55 |
|
3808 0x000000555bd82b1c: dsll t9, t9, 16 |
|
3809 0x000000555bd82b20: ori t9, t9, 0x5aa6 |
|
3810 0x000000555bd82b24: dsll t9, t9, 16 |
|
3811 0x000000555bd82b28: ori t9, t9, 0x4074 |
|
3812 0x000000555bd82b2c: jalr t9 |
|
3813 0x000000555bd82b30: sll zero, zero, 0 |
|
3814 |
|
3815 0x000000555bd82b34: daddiu sp, sp, 0x10 ; save_pc |
|
3816 */ |
|
3817 __ move(A0, thread); |
|
3818 __ addi(SP, SP, -additional_words * wordSize); |
|
3819 |
|
3820 __ set_last_Java_frame(NOREG, NOREG, NULL); |
|
3821 |
|
3822 // Call fetch_unroll_info(). Need thread and this frame, but NOT official VM entry - cannot block on |
|
3823 // this call, no GC can happen. Call should capture return values. |
|
3824 |
|
3825 __ relocate(relocInfo::internal_pc_type); |
|
3826 { |
|
3827 intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 4; |
|
3828 __ li48(AT, save_pc); |
|
3829 } |
|
3830 __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
|
3831 |
|
3832 __ call((address)Deoptimization::fetch_unroll_info); |
|
3833 //__ call(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info), relocInfo::runtime_call_type); |
|
3834 __ delayed()->nop(); |
|
3835 oop_maps->add_gc_map(__ pc() - start, map); |
|
3836 __ addiu(SP, SP, additional_words * wordSize); |
|
3837 __ get_thread(thread); |
|
3838 __ reset_last_Java_frame(false, true); |
|
3839 |
|
3840 // Load UnrollBlock into S7 |
|
3841 __ move(unroll, V0); |
|
3842 |
|
3843 |
|
3844 // Move the unpack kind to a safe place in the UnrollBlock because |
|
3845 // we are very short of registers |
|
3846 |
|
3847 Address unpack_kind(unroll, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes()); |
|
3848 //__ pop(reason); |
|
3849 __ sw(reason, unpack_kind); |
|
3850 // save the unpack_kind value |
|
3851 // Retrieve the possible live values (return values) |
|
3852 // All callee save registers representing jvm state |
|
3853 // are now in the vframeArray. |
|
3854 |
|
3855 Label noException; |
|
3856 __ move(AT, Deoptimization::Unpack_exception); |
|
3857 __ bne(AT, reason, noException);// Was exception pending? |
|
3858 __ delayed()->nop(); |
|
3859 __ ld_ptr(V0, thread, in_bytes(JavaThread::exception_oop_offset())); |
|
3860 __ ld_ptr(V1, thread, in_bytes(JavaThread::exception_pc_offset())); |
|
3861 __ st_ptr(R0, thread, in_bytes(JavaThread::exception_pc_offset())); |
|
3862 __ st_ptr(R0, thread, in_bytes(JavaThread::exception_oop_offset())); |
|
3863 |
|
3864 __ verify_oop(V0); |
|
3865 |
|
3866 // Overwrite the result registers with the exception results. |
|
3867 __ st_ptr(V0, SP, RegisterSaver::v0Offset()*wordSize); |
|
3868 __ st_ptr(V1, SP, RegisterSaver::v1Offset()*wordSize); |
|
3869 |
|
3870 __ bind(noException); |
|
3871 |
|
3872 |
|
3873 // Stack is back to only having register save data on the stack. |
|
3874 // Now restore the result registers. Everything else is either dead or captured |
|
3875 // in the vframeArray. |
|
3876 |
|
3877 RegisterSaver::restore_result_registers(masm); |
|
3878 // All of the register save area has been popped of the stack. Only the |
|
3879 // return address remains. |
|
3880 // Pop all the frames we must move/replace. |
|
3881 // Frame picture (youngest to oldest) |
|
3882 // 1: self-frame (no frame link) |
|
3883 // 2: deopting frame (no frame link) |
|
3884 // 3: caller of deopting frame (could be compiled/interpreted). |
|
3885 // |
|
3886 // Note: by leaving the return address of self-frame on the stack |
|
3887 // and using the size of frame 2 to adjust the stack |
|
3888 // when we are done the return to frame 3 will still be on the stack. |
|
3889 |
|
3890 // register for the sender's sp |
|
3891 Register sender_sp = Rsender; |
|
3892 // register for frame pcs |
|
3893 Register pcs = T0; |
|
3894 // register for frame sizes |
|
3895 Register sizes = T1; |
|
3896 // register for frame count |
|
3897 Register count = T3; |
|
3898 |
|
3899 // Pop deoptimized frame |
|
3900 __ lw(AT, unroll, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()); |
|
3901 __ add(SP, SP, AT); |
|
3902 // sp should be pointing at the return address to the caller (3) |
|
3903 |
|
3904 // Load array of frame pcs into pcs |
|
3905 __ ld_ptr(pcs, unroll, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()); |
|
3906 __ addi(SP, SP, wordSize); // trash the old pc |
|
3907 // Load array of frame sizes into T6 |
|
3908 __ ld_ptr(sizes, unroll, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()); |
|
3909 |
|
3910 |
|
3911 |
|
3912 // Load count of frams into T3 |
|
3913 __ lw(count, unroll, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()); |
|
3914 // Pick up the initial fp we should save |
|
3915 __ ld(FP, unroll, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()); |
|
3916 // Now adjust the caller's stack to make up for the extra locals |
|
3917 // but record the original sp so that we can save it in the skeletal interpreter |
|
3918 // frame and the stack walking of interpreter_sender will get the unextended sp |
|
3919 // value and not the "real" sp value. |
|
3920 __ move(sender_sp, SP); |
|
3921 __ lw(AT, unroll, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()); |
|
3922 __ sub(SP, SP, AT); |
|
3923 |
|
3924 // Push interpreter frames in a loop |
|
3925 /* |
|
3926 * |
|
3927 Loop: |
|
3928 0x000000555bd82d18: lw t2, 0x0(t1) ; lw sizes[i] <--- error lw->ld |
|
3929 0x000000555bd82d1c: ld at, 0x0(t0) ; ld pcs[i] |
|
3930 0x000000555bd82d20: daddi t2, t2, 0xfffffff0 ; t2 -= 16 |
|
3931 0x000000555bd82d24: daddi sp, sp, 0xfffffff0 |
|
3932 0x000000555bd82d28: sd fp, 0x0(sp) ; push fp |
|
3933 0x000000555bd82d2c: sd at, 0x8(sp) ; push at |
|
3934 0x000000555bd82d30: dadd fp, sp, zero ; fp <- sp |
|
3935 0x000000555bd82d34: dsub sp, sp, t2 ; sp -= t2 |
|
3936 0x000000555bd82d38: sd zero, 0xfffffff0(fp) ; __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
|
3937 0x000000555bd82d3c: sd s4, 0xfffffff8(fp) ; __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize); |
|
3938 0x000000555bd82d40: dadd s4, sp, zero ; move(sender_sp, SP); |
|
3939 0x000000555bd82d44: daddi t3, t3, 0xffffffff ; count -- |
|
3940 0x000000555bd82d48: daddi t1, t1, 0x4 ; sizes += 4 |
|
3941 0x000000555bd82d4c: bne t3, zero, 0x000000555bd82d18 |
|
3942 0x000000555bd82d50: daddi t0, t0, 0x4 ; <--- error t0 += 8 |
|
3943 */ |
|
3944 |
|
3945 // pcs[0] = frame_pcs[0] = deopt_sender.raw_pc(); regex.split |
|
3946 Label loop; |
|
3947 __ bind(loop); |
|
3948 __ ld(T2, sizes, 0); // Load frame size |
|
3949 __ ld_ptr(AT, pcs, 0); // save return address |
|
3950 __ addi(T2, T2, -2*wordSize); // we'll push pc and rbp, by hand |
|
3951 __ push2(AT, FP); |
|
3952 __ move(FP, SP); |
|
3953 __ sub(SP, SP, T2); // Prolog! |
|
3954 // This value is corrected by layout_activation_impl |
|
3955 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
|
3956 __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize);// Make it walkable |
|
3957 __ move(sender_sp, SP); // pass to next frame |
|
3958 __ addi(count, count, -1); // decrement counter |
|
3959 __ addi(sizes, sizes, wordSize); // Bump array pointer (sizes) |
|
3960 __ bne(count, R0, loop); |
|
3961 __ delayed()->addi(pcs, pcs, wordSize); // Bump array pointer (pcs) |
|
3962 __ ld(AT, pcs, 0); // frame_pcs[number_of_frames] = Interpreter::deopt_entry(vtos, 0); |
|
3963 // Re-push self-frame |
|
3964 __ push2(AT, FP); |
|
3965 __ move(FP, SP); |
|
3966 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
|
3967 __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize); |
|
3968 __ addi(SP, SP, -(frame_size_in_words - 2 - additional_words) * wordSize); |
|
3969 |
|
3970 // Restore frame locals after moving the frame |
|
3971 __ sd(V0, SP, RegisterSaver::v0Offset() * wordSize); |
|
3972 __ sd(V1, SP, RegisterSaver::v1Offset() * wordSize); |
|
3973 __ sdc1(F0, SP, RegisterSaver::fpResultOffset()* wordSize);// Pop float stack and store in local |
|
3974 __ sdc1(F1, SP, (RegisterSaver::fpResultOffset() + 1) * wordSize); |
|
3975 |
|
3976 |
|
3977 // Call unpack_frames(). Need thread and this frame, but NOT official VM entry - cannot block on |
|
3978 // this call, no GC can happen. |
|
3979 __ move(A1, reason); // exec_mode |
|
3980 __ get_thread(thread); |
|
3981 __ move(A0, thread); // thread |
|
3982 __ addi(SP, SP, (-additional_words) *wordSize); |
|
3983 |
|
3984 // set last_Java_sp, last_Java_fp |
|
3985 __ set_last_Java_frame(NOREG, FP, NULL); |
|
3986 |
|
3987 __ move(AT, -(StackAlignmentInBytes)); |
|
3988 __ andr(SP, SP, AT); // Fix stack alignment as required by ABI |
|
3989 |
|
3990 __ relocate(relocInfo::internal_pc_type); |
|
3991 { |
|
3992 intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 4; |
|
3993 __ li48(AT, save_pc); |
|
3994 } |
|
3995 __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
|
3996 |
|
3997 //__ call(Deoptimization::unpack_frames); |
|
3998 __ call(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames), relocInfo::runtime_call_type); |
|
3999 __ delayed()->nop(); |
|
4000 // Revert SP alignment after call since we're going to do some SP relative addressing below |
|
4001 __ ld(SP, thread, in_bytes(JavaThread::last_Java_sp_offset())); |
|
4002 // Set an oopmap for the call site |
|
4003 oop_maps->add_gc_map(__ offset(), new OopMap( frame_size_in_words , 0)); |
|
4004 |
|
4005 __ push(V0); |
|
4006 |
|
4007 __ get_thread(thread); |
|
4008 __ reset_last_Java_frame(false, false); |
|
4009 |
|
4010 // Collect return values |
|
4011 __ ld(V0, SP, (RegisterSaver::v0Offset() + additional_words +1) * wordSize); |
|
4012 __ ld(V1, SP, (RegisterSaver::v1Offset() + additional_words +1) * wordSize); |
|
4013 __ ldc1(F0, SP, RegisterSaver::fpResultOffset()* wordSize);// Pop float stack and store in local |
|
4014 __ ldc1(F1, SP, (RegisterSaver::fpResultOffset() + 1) * wordSize); |
|
4015 //FIXME, |
|
4016 // Clear floating point stack before returning to interpreter |
|
4017 __ empty_FPU_stack(); |
|
4018 //FIXME, we should consider about float and double |
|
4019 // Push a float or double return value if necessary. |
|
4020 __ leave(); |
|
4021 |
|
4022 // Jump to interpreter |
|
4023 __ jr(RA); |
|
4024 __ delayed()->nop(); |
|
4025 |
|
4026 masm->flush(); |
|
4027 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); |
|
4028 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); |
|
4029 } |
|
4030 |
|
4031 #ifdef COMPILER2 |
|
4032 |
|
4033 //------------------------------generate_uncommon_trap_blob-------------------- |
|
4034 // Ought to generate an ideal graph & compile, but here's some SPARC ASM |
|
4035 // instead. |
|
4036 void SharedRuntime::generate_uncommon_trap_blob() { |
|
4037 // allocate space for the code |
|
4038 ResourceMark rm; |
|
4039 // setup code generation tools |
|
4040 CodeBuffer buffer ("uncommon_trap_blob", 512*80 , 512*40 ); |
|
4041 MacroAssembler* masm = new MacroAssembler(&buffer); |
|
4042 |
|
4043 enum frame_layout { |
|
4044 s0_off, s0_off2, |
|
4045 s1_off, s1_off2, |
|
4046 s2_off, s2_off2, |
|
4047 s3_off, s3_off2, |
|
4048 s4_off, s4_off2, |
|
4049 s5_off, s5_off2, |
|
4050 s6_off, s6_off2, |
|
4051 s7_off, s7_off2, |
|
4052 fp_off, fp_off2, |
|
4053 return_off, return_off2, // slot for return address sp + 9 |
|
4054 framesize |
|
4055 }; |
|
4056 assert(framesize % 4 == 0, "sp not 16-byte aligned"); |
|
4057 |
|
4058 address start = __ pc(); |
|
4059 |
|
4060 // Push self-frame. |
|
4061 __ daddiu(SP, SP, -framesize * BytesPerInt); |
|
4062 |
|
4063 __ sd(RA, SP, return_off * BytesPerInt); |
|
4064 __ sd(FP, SP, fp_off * BytesPerInt); |
|
4065 |
|
4066 // Save callee saved registers. None for UseSSE=0, |
|
4067 // floats-only for UseSSE=1, and doubles for UseSSE=2. |
|
4068 __ sd(S0, SP, s0_off * BytesPerInt); |
|
4069 __ sd(S1, SP, s1_off * BytesPerInt); |
|
4070 __ sd(S2, SP, s2_off * BytesPerInt); |
|
4071 __ sd(S3, SP, s3_off * BytesPerInt); |
|
4072 __ sd(S4, SP, s4_off * BytesPerInt); |
|
4073 __ sd(S5, SP, s5_off * BytesPerInt); |
|
4074 __ sd(S6, SP, s6_off * BytesPerInt); |
|
4075 __ sd(S7, SP, s7_off * BytesPerInt); |
|
4076 |
|
4077 __ daddi(FP, SP, fp_off * BytesPerInt); |
|
4078 |
|
4079 // Clear the floating point exception stack |
|
4080 __ empty_FPU_stack(); |
|
4081 |
|
4082 Register thread = TREG; |
|
4083 |
|
4084 #ifndef OPT_THREAD |
|
4085 __ get_thread(thread); |
|
4086 #endif |
|
4087 // set last_Java_sp |
|
4088 __ set_last_Java_frame(NOREG, FP, NULL); |
|
4089 __ relocate(relocInfo::internal_pc_type); |
|
4090 assert(NativeCall::return_address_offset == 24, "in sharedRuntime return_address_offset"); |
|
4091 { |
|
4092 long save_pc = (long)__ pc() + 28 + NativeCall::return_address_offset; |
|
4093 __ li48(AT, (long)save_pc); |
|
4094 __ sd(AT, thread, in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
|
4095 } |
|
4096 // Call C code. Need thread but NOT official VM entry |
|
4097 // crud. We cannot block on this call, no GC can happen. Call should |
|
4098 // capture callee-saved registers as well as return values. |
|
4099 __ move(A0, thread); |
|
4100 // argument already in T0 |
|
4101 __ move(A1, T0); |
|
4102 __ li48(T9, (long)Deoptimization::uncommon_trap); |
|
4103 __ jalr(T9); |
|
4104 __ delayed()->nop(); |
|
4105 |
|
4106 // Set an oopmap for the call site |
|
4107 OopMapSet *oop_maps = new OopMapSet(); |
|
4108 OopMap* map = new OopMap( framesize, 0 ); |
|
4109 |
|
4110 map->set_callee_saved( VMRegImpl::stack2reg(s0_off ), S0->as_VMReg() ); |
|
4111 map->set_callee_saved( VMRegImpl::stack2reg(s1_off ), S1->as_VMReg() ); |
|
4112 map->set_callee_saved( VMRegImpl::stack2reg(s2_off ), S2->as_VMReg() ); |
|
4113 map->set_callee_saved( VMRegImpl::stack2reg(s3_off ), S3->as_VMReg() ); |
|
4114 map->set_callee_saved( VMRegImpl::stack2reg(s4_off ), S4->as_VMReg() ); |
|
4115 map->set_callee_saved( VMRegImpl::stack2reg(s5_off ), S5->as_VMReg() ); |
|
4116 map->set_callee_saved( VMRegImpl::stack2reg(s6_off ), S6->as_VMReg() ); |
|
4117 map->set_callee_saved( VMRegImpl::stack2reg(s7_off ), S7->as_VMReg() ); |
|
4118 |
|
4119 //oop_maps->add_gc_map( __ offset(), true, map); |
|
4120 oop_maps->add_gc_map( __ offset(), map); |
|
4121 |
|
4122 #ifndef OPT_THREAD |
|
4123 __ get_thread(thread); |
|
4124 #endif |
|
4125 __ reset_last_Java_frame(false,false); |
|
4126 |
|
4127 // Load UnrollBlock into S7 |
|
4128 Register unroll = S7; |
|
4129 __ move(unroll, V0); |
|
4130 |
|
4131 // Pop all the frames we must move/replace. |
|
4132 // |
|
4133 // Frame picture (youngest to oldest) |
|
4134 // 1: self-frame (no frame link) |
|
4135 // 2: deopting frame (no frame link) |
|
4136 // 3: possible-i2c-adapter-frame |
|
4137 // 4: caller of deopting frame (could be compiled/interpreted. If interpreted we will create an |
|
4138 // and c2i here) |
|
4139 |
|
4140 // Pop self-frame. We have no frame, and must rely only on EAX and ESP. |
|
4141 __ daddiu(SP, SP, framesize * BytesPerInt); |
|
4142 |
|
4143 // Pop deoptimized frame |
|
4144 __ lw(AT, unroll, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()); |
|
4145 __ dadd(SP, SP, AT); |
|
4146 |
|
4147 // register for frame pcs |
|
4148 Register pcs = T8; |
|
4149 // register for frame sizes |
|
4150 Register sizes = T9; |
|
4151 // register for frame count |
|
4152 Register count = T3; |
|
4153 // register for the sender's sp |
|
4154 Register sender_sp = T1; |
|
4155 |
|
4156 // sp should be pointing at the return address to the caller (4) |
|
4157 // Load array of frame pcs into ECX |
|
4158 __ ld(pcs, unroll, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()); |
|
4159 |
|
4160 /* 2012/9/7 Not needed in MIPS |
|
4161 __ addiu(SP, SP, wordSize); |
|
4162 */ |
|
4163 |
|
4164 // Load array of frame sizes into ESI |
|
4165 __ ld(sizes, unroll, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()); |
|
4166 __ lwu(count, unroll, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()); |
|
4167 |
|
4168 // Pick up the initial fp we should save |
|
4169 __ ld(FP, unroll, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()); |
|
4170 // Now adjust the caller's stack to make up for the extra locals |
|
4171 // but record the original sp so that we can save it in the skeletal interpreter |
|
4172 // frame and the stack walking of interpreter_sender will get the unextended sp |
|
4173 // value and not the "real" sp value. |
|
4174 |
|
4175 __ move(sender_sp, SP); |
|
4176 __ lw(AT, unroll, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()); |
|
4177 __ dsub(SP, SP, AT); |
|
4178 // Push interpreter frames in a loop |
|
4179 Label loop; |
|
4180 __ bind(loop); |
|
4181 __ ld(T2, sizes, 0); // Load frame size |
|
4182 __ ld(AT, pcs, 0); // save return address |
|
4183 __ daddi(T2, T2, -2*wordSize); // we'll push pc and rbp, by hand |
|
4184 __ push2(AT, FP); |
|
4185 __ move(FP, SP); |
|
4186 __ dsub(SP, SP, T2); // Prolog! |
|
4187 // This value is corrected by layout_activation_impl |
|
4188 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize); |
|
4189 __ sd(sender_sp, FP, frame::interpreter_frame_sender_sp_offset * wordSize);// Make it walkable |
|
4190 __ move(sender_sp, SP); // pass to next frame |
|
4191 __ daddi(count, count, -1); // decrement counter |
|
4192 __ daddi(sizes, sizes, wordSize); // Bump array pointer (sizes) |
|
4193 __ addi(pcs, pcs, wordSize); // Bump array pointer (pcs) |
|
4194 __ bne(count, R0, loop); |
|
4195 __ delayed()->nop(); // Bump array pointer (pcs) |
|
4196 |
|
4197 __ ld(RA, pcs, 0); |
|
4198 |
|
4199 // Re-push self-frame |
|
4200 __ daddi(SP, SP, - 2 * wordSize); // save old & set new FP |
|
4201 __ sd(FP, SP, 0 * wordSize); // save final return address |
|
4202 __ sd(RA, SP, 1 * wordSize); |
|
4203 __ move(FP, SP); |
|
4204 __ daddi(SP, SP, -(framesize / 2 - 2) * wordSize); |
|
4205 |
|
4206 // set last_Java_sp, last_Java_fp |
|
4207 __ set_last_Java_frame(NOREG, FP, NULL); |
|
4208 |
|
4209 __ move(AT, -(StackAlignmentInBytes)); |
|
4210 __ andr(SP, SP, AT); // Fix stack alignment as required by ABI |
|
4211 |
|
4212 __ relocate(relocInfo::internal_pc_type); |
|
4213 { |
|
4214 long save_pc = (long)__ pc() + 28 + NativeCall::return_address_offset; |
|
4215 __ li48(AT, (long)save_pc); |
|
4216 } |
|
4217 __ sd(AT, thread,in_bytes(JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset())); |
|
4218 |
|
4219 // Call C code. Need thread but NOT official VM entry |
|
4220 // crud. We cannot block on this call, no GC can happen. Call should |
|
4221 // restore return values to their stack-slots with the new SP. |
|
4222 __ move(A0, thread); |
|
4223 __ move(A1, Deoptimization::Unpack_uncommon_trap); |
|
4224 __ li48(T9, (long)Deoptimization::unpack_frames); |
|
4225 __ jalr(T9); |
|
4226 __ delayed()->nop(); |
|
4227 // Set an oopmap for the call site |
|
4228 //oop_maps->add_gc_map( __ offset(), true, new OopMap( framesize, 0 ) ); |
|
4229 oop_maps->add_gc_map( __ offset(), new OopMap( framesize, 0 ) );//Fu |
|
4230 |
|
4231 __ reset_last_Java_frame(true,true); |
|
4232 |
|
4233 // Pop self-frame. |
|
4234 __ leave(); // Epilog! |
|
4235 |
|
4236 // Jump to interpreter |
|
4237 __ jr(RA); |
|
4238 __ delayed()->nop(); |
|
4239 // ------------- |
|
4240 // make sure all code is generated |
|
4241 masm->flush(); |
|
4242 |
|
4243 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize / 2); |
|
4244 } |
|
4245 |
|
4246 #endif // COMPILER2 |
|
4247 |
|
4248 //------------------------------generate_handler_blob------------------- |
|
4249 // |
|
4250 // Generate a special Compile2Runtime blob that saves all registers, and sets |
|
4251 // up an OopMap and calls safepoint code to stop the compiled code for |
|
4252 // a safepoint. |
|
4253 // |
|
4254 // This blob is jumped to (via a breakpoint and the signal handler) from a |
|
4255 // safepoint in compiled code. |
|
4256 |
|
4257 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int pool_type) { |
|
4258 |
|
4259 // Account for thread arg in our frame |
|
4260 const int additional_words = 0; |
|
4261 int frame_size_in_words; |
|
4262 |
|
4263 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
|
4264 |
|
4265 ResourceMark rm; |
|
4266 OopMapSet *oop_maps = new OopMapSet(); |
|
4267 OopMap* map; |
|
4268 |
|
4269 // allocate space for the code |
|
4270 // setup code generation tools |
|
4271 CodeBuffer buffer ("handler_blob", 2048, 512); |
|
4272 MacroAssembler* masm = new MacroAssembler( &buffer); |
|
4273 |
|
4274 const Register thread = TREG; |
|
4275 address start = __ pc(); |
|
4276 address call_pc = NULL; |
|
4277 bool cause_return = (pool_type == POLL_AT_RETURN); |
|
4278 bool save_vectors = (pool_type == POLL_AT_VECTOR_LOOP); |
|
4279 |
|
4280 // If cause_return is true we are at a poll_return and there is |
|
4281 // the return address in RA to the caller on the nmethod |
|
4282 // that is safepoint. We can leave this return in RA and |
|
4283 // effectively complete the return and safepoint in the caller. |
|
4284 // Otherwise we load exception pc to RA. |
|
4285 __ push(thread); |
|
4286 #ifndef OPT_THREAD |
|
4287 __ get_thread(thread); |
|
4288 #endif |
|
4289 |
|
4290 if(!cause_return) { |
|
4291 __ ld_ptr(RA, Address(thread, JavaThread::saved_exception_pc_offset())); |
|
4292 } |
|
4293 |
|
4294 __ pop(thread); |
|
4295 map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, save_vectors); |
|
4296 |
|
4297 #ifndef OPT_THREAD |
|
4298 __ get_thread(thread); |
|
4299 #endif |
|
4300 // The following is basically a call_VM. However, we need the precise |
|
4301 // address of the call in order to generate an oopmap. Hence, we do all the |
|
4302 // work outselvs. |
|
4303 |
|
4304 __ move(A0, thread); |
|
4305 __ set_last_Java_frame(NOREG, NOREG, NULL); |
|
4306 |
|
4307 //__ relocate(relocInfo::internal_pc_type); |
|
4308 if (!cause_return) |
|
4309 { |
|
4310 /* |
|
4311 intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 4; |
|
4312 __ li48(AT, save_pc); |
|
4313 __ sd(AT, thread, in_bytes(JavaThread::last_Java_pc_offset())); |
|
4314 */ |
|
4315 } |
|
4316 |
|
4317 |
|
4318 // do the call |
|
4319 //__ lui(T9, Assembler::split_high((int)call_ptr)); |
|
4320 //__ addiu(T9, T9, Assembler::split_low((int)call_ptr)); |
|
4321 __ call(call_ptr); |
|
4322 __ delayed()->nop(); |
|
4323 |
|
4324 // Set an oopmap for the call site. This oopmap will map all |
|
4325 // oop-registers and debug-info registers as callee-saved. This |
|
4326 // will allow deoptimization at this safepoint to find all possible |
|
4327 // debug-info recordings, as well as let GC find all oops. |
|
4328 oop_maps->add_gc_map(__ offset(), map); |
|
4329 |
|
4330 Label noException; |
|
4331 |
|
4332 // Clear last_Java_sp again |
|
4333 __ reset_last_Java_frame(false, false); |
|
4334 |
|
4335 __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
4336 __ beq(AT, R0, noException); |
|
4337 __ delayed()->nop(); |
|
4338 |
|
4339 // Exception pending |
|
4340 |
|
4341 RegisterSaver::restore_live_registers(masm, save_vectors); |
|
4342 //forward_exception_entry need return address on the stack |
|
4343 __ push(RA); |
|
4344 //__ lui(T9, Assembler::split_high((int)StubRoutines::forward_exception_entry())); |
|
4345 //__ addiu(T9, T9, Assembler::split_low((int)StubRoutines::forward_exception_entry())); |
|
4346 __ li(T9, StubRoutines::forward_exception_entry()); |
|
4347 __ jr(T9); |
|
4348 __ delayed()->nop(); |
|
4349 |
|
4350 // No exception case |
|
4351 __ bind(noException); |
|
4352 // Normal exit, register restoring and exit |
|
4353 RegisterSaver::restore_live_registers(masm, save_vectors); |
|
4354 __ jr(RA); |
|
4355 __ delayed()->nop(); |
|
4356 |
|
4357 masm->flush(); |
|
4358 |
|
4359 // Fill-out other meta info |
|
4360 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); |
|
4361 } |
|
4362 |
|
4363 // |
|
4364 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss |
|
4365 // |
|
4366 // Generate a stub that calls into vm to find out the proper destination |
|
4367 // of a java call. All the argument registers are live at this point |
|
4368 // but since this is generic code we don't know what they are and the caller |
|
4369 // must do any gc of the args. |
|
4370 // |
|
4371 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) { |
|
4372 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
|
4373 |
|
4374 // allocate space for the code |
|
4375 ResourceMark rm; |
|
4376 |
|
4377 //CodeBuffer buffer(name, 1000, 512); |
|
4378 //FIXME. aoqi. code_size |
|
4379 CodeBuffer buffer(name, 20000, 2048); |
|
4380 MacroAssembler* masm = new MacroAssembler(&buffer); |
|
4381 |
|
4382 int frame_size_words; |
|
4383 //we put the thread in A0 |
|
4384 |
|
4385 OopMapSet *oop_maps = new OopMapSet(); |
|
4386 OopMap* map = NULL; |
|
4387 |
|
4388 int start = __ offset(); |
|
4389 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_words); |
|
4390 |
|
4391 |
|
4392 int frame_complete = __ offset(); |
|
4393 |
|
4394 const Register thread = T8; |
|
4395 __ get_thread(thread); |
|
4396 |
|
4397 __ move(A0, thread); |
|
4398 __ set_last_Java_frame(noreg, FP, NULL); |
|
4399 //__ addi(SP, SP, -wordSize); |
|
4400 //align the stack before invoke native |
|
4401 __ move(AT, -(StackAlignmentInBytes)); |
|
4402 __ andr(SP, SP, AT); |
|
4403 __ relocate(relocInfo::internal_pc_type); |
|
4404 { |
|
4405 intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 1 * BytesPerInstWord; |
|
4406 //tty->print_cr(" %s :%d, name:%s, pc: %lx, save_pc: %lx, frame_size_words: %lx", __func__, __LINE__, name, __ pc(), save_pc, frame_size_words); //aoqi_test |
|
4407 __ li48(AT, save_pc); |
|
4408 } |
|
4409 __ sd(AT, thread, in_bytes(JavaThread::last_Java_pc_offset())); |
|
4410 |
|
4411 __ call(destination); |
|
4412 __ delayed()->nop(); |
|
4413 |
|
4414 // Set an oopmap for the call site. |
|
4415 // We need this not only for callee-saved registers, but also for volatile |
|
4416 // registers that the compiler might be keeping live across a safepoint. |
|
4417 oop_maps->add_gc_map( __ offset() - start, map); |
|
4418 // V0 contains the address we are going to jump to assuming no exception got installed |
|
4419 __ get_thread(thread); |
|
4420 __ ld_ptr(SP, thread, in_bytes(JavaThread::last_Java_sp_offset())); |
|
4421 // clear last_Java_sp |
|
4422 __ reset_last_Java_frame(true, true); |
|
4423 // check for pending exceptions |
|
4424 Label pending; |
|
4425 __ ld_ptr(AT, thread, in_bytes(Thread::pending_exception_offset())); |
|
4426 __ bne(AT, R0, pending); |
|
4427 __ delayed()->nop(); |
|
4428 // get the returned Method* |
|
4429 //FIXME, do mips need this ? |
|
4430 __ get_vm_result_2(Rmethod, thread); // Refer to OpenJDK8 |
|
4431 __ st_ptr(Rmethod, SP, RegisterSaver::methodOffset() * wordSize); |
|
4432 __ st_ptr(V0, SP, RegisterSaver::v0Offset() * wordSize); |
|
4433 RegisterSaver::restore_live_registers(masm); |
|
4434 |
|
4435 // We are back the the original state on entry and ready to go the callee method. |
|
4436 __ jr(V0); |
|
4437 __ delayed()->nop(); |
|
4438 // Pending exception after the safepoint |
|
4439 |
|
4440 __ bind(pending); |
|
4441 |
|
4442 RegisterSaver::restore_live_registers(masm); |
|
4443 |
|
4444 // exception pending => remove activation and forward to exception handler |
|
4445 //forward_exception_entry need return address on the stack |
|
4446 __ push(RA); |
|
4447 __ get_thread(thread); |
|
4448 __ st_ptr(R0, thread, in_bytes(JavaThread::vm_result_offset())); |
|
4449 __ ld_ptr(V0, thread, in_bytes(Thread::pending_exception_offset())); |
|
4450 __ jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); |
|
4451 __ delayed() -> nop(); |
|
4452 // ------------- |
|
4453 // make sure all code is generated |
|
4454 masm->flush(); |
|
4455 |
|
4456 RuntimeStub* tmp= RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true); |
|
4457 return tmp; |
|
4458 } |
|
4459 |
|
4460 /*void SharedRuntime::generate_stubs() { |
|
4461 _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, |
|
4462 SharedRuntime::handle_wrong_method),"wrong_method_stub"); |
|
4463 _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, |
|
4464 SharedRuntime::handle_wrong_method_ic_miss),"ic_miss_stub"); |
|
4465 _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, |
|
4466 SharedRuntime::resolve_opt_virtual_call_C),"resolve_opt_virtual_call"); |
|
4467 _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, |
|
4468 SharedRuntime::resolve_virtual_call_C),"resolve_virtual_call"); |
|
4469 _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, |
|
4470 SharedRuntime::resolve_static_call_C),"resolve_static_call"); |
|
4471 _polling_page_safepoint_handler_blob =generate_handler_blob(CAST_FROM_FN_PTR(address, |
|
4472 SafepointSynchronize::handle_polling_page_exception), false); |
|
4473 _polling_page_return_handler_blob =generate_handler_blob(CAST_FROM_FN_PTR(address, |
|
4474 SafepointSynchronize::handle_polling_page_exception), true); |
|
4475 generate_deopt_blob(); |
|
4476 #ifdef COMPILER2 |
|
4477 generate_uncommon_trap_blob(); |
|
4478 #endif // COMPILER2 |
|
4479 }*/ |
|
4480 |
|
4481 extern "C" int SpinPause() {return 0;} |
|
4482 // extern "C" int SafeFetch32 (int * adr, int errValue) {return 0;} ; |
|
4483 // extern "C" intptr_t SafeFetchN (intptr_t * adr, intptr_t errValue) {return *adr; } ; |