Mercurial > hg > graal-compiler
comparison src/cpu/sparc/vm/sharedRuntime_sparc.cpp @ 0:a61af66fc99e jdk7-b24
Initial load
author | duke |
---|---|
date | Sat, 01 Dec 2007 00:00:00 +0000 |
parents | |
children | ba764ed4b6f2 |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:a61af66fc99e |
---|---|
1 /* | |
2 * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. | |
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 #include "incls/_precompiled.incl" | |
26 #include "incls/_sharedRuntime_sparc.cpp.incl" | |
27 | |
28 #define __ masm-> | |
29 | |
30 #ifdef COMPILER2 | |
31 UncommonTrapBlob* SharedRuntime::_uncommon_trap_blob; | |
32 #endif // COMPILER2 | |
33 | |
34 DeoptimizationBlob* SharedRuntime::_deopt_blob; | |
35 SafepointBlob* SharedRuntime::_polling_page_safepoint_handler_blob; | |
36 SafepointBlob* SharedRuntime::_polling_page_return_handler_blob; | |
37 RuntimeStub* SharedRuntime::_wrong_method_blob; | |
38 RuntimeStub* SharedRuntime::_ic_miss_blob; | |
39 RuntimeStub* SharedRuntime::_resolve_opt_virtual_call_blob; | |
40 RuntimeStub* SharedRuntime::_resolve_virtual_call_blob; | |
41 RuntimeStub* SharedRuntime::_resolve_static_call_blob; | |
42 | |
43 class RegisterSaver { | |
44 | |
45 // Used for saving volatile registers. This is Gregs, Fregs, I/L/O. | |
46 // The Oregs are problematic. In the 32bit build the compiler can | |
47 // have O registers live with 64 bit quantities. A window save will | |
48 // cut the heads off of the registers. We have to do a very extensive | |
49 // stack dance to save and restore these properly. | |
50 | |
51 // Note that the Oregs problem only exists if we block at either a polling | |
52 // page exception a compiled code safepoint that was not originally a call | |
53 // or deoptimize following one of these kinds of safepoints. | |
54 | |
55 // Lots of registers to save. For all builds, a window save will preserve | |
56 // the %i and %l registers. For the 32-bit longs-in-two entries and 64-bit | |
57 // builds a window-save will preserve the %o registers. In the LION build | |
58 // we need to save the 64-bit %o registers which requires we save them | |
59 // before the window-save (as then they become %i registers and get their | |
60 // heads chopped off on interrupt). We have to save some %g registers here | |
61 // as well. | |
62 enum { | |
63 // This frame's save area. Includes extra space for the native call: | |
64 // vararg's layout space and the like. Briefly holds the caller's | |
65 // register save area. | |
66 call_args_area = frame::register_save_words_sp_offset + | |
67 frame::memory_parameter_word_sp_offset*wordSize, | |
68 // Make sure save locations are always 8 byte aligned. | |
69 // can't use round_to because it doesn't produce compile time constant | |
70 start_of_extra_save_area = ((call_args_area + 7) & ~7), | |
71 g1_offset = start_of_extra_save_area, // g-regs needing saving | |
72 g3_offset = g1_offset+8, | |
73 g4_offset = g3_offset+8, | |
74 g5_offset = g4_offset+8, | |
75 o0_offset = g5_offset+8, | |
76 o1_offset = o0_offset+8, | |
77 o2_offset = o1_offset+8, | |
78 o3_offset = o2_offset+8, | |
79 o4_offset = o3_offset+8, | |
80 o5_offset = o4_offset+8, | |
81 start_of_flags_save_area = o5_offset+8, | |
82 ccr_offset = start_of_flags_save_area, | |
83 fsr_offset = ccr_offset + 8, | |
84 d00_offset = fsr_offset+8, // Start of float save area | |
85 register_save_size = d00_offset+8*32 | |
86 }; | |
87 | |
88 | |
89 public: | |
90 | |
91 static int Oexception_offset() { return o0_offset; }; | |
92 static int G3_offset() { return g3_offset; }; | |
93 static int G5_offset() { return g5_offset; }; | |
94 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words); | |
95 static void restore_live_registers(MacroAssembler* masm); | |
96 | |
97 // During deoptimization only the result register need to be restored | |
98 // all the other values have already been extracted. | |
99 | |
100 static void restore_result_registers(MacroAssembler* masm); | |
101 }; | |
102 | |
103 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) { | |
104 // Record volatile registers as callee-save values in an OopMap so their save locations will be | |
105 // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for | |
106 // deoptimization; see compiledVFrame::create_stack_value). The caller's I, L and O registers | |
107 // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame | |
108 // (as the stub's I's) when the runtime routine called by the stub creates its frame. | |
109 int i; | |
110 // Always make the frame size 16 bytr aligned. | |
111 int frame_size = round_to(additional_frame_words + register_save_size, 16); | |
112 // OopMap frame size is in c2 stack slots (sizeof(jint)) not bytes or words | |
113 int frame_size_in_slots = frame_size / sizeof(jint); | |
114 // CodeBlob frame size is in words. | |
115 *total_frame_words = frame_size / wordSize; | |
116 // OopMap* map = new OopMap(*total_frame_words, 0); | |
117 OopMap* map = new OopMap(frame_size_in_slots, 0); | |
118 | |
119 #if !defined(_LP64) | |
120 | |
121 // Save 64-bit O registers; they will get their heads chopped off on a 'save'. | |
122 __ stx(O0, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8); | |
123 __ stx(O1, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8); | |
124 __ stx(O2, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+2*8); | |
125 __ stx(O3, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+3*8); | |
126 __ stx(O4, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+4*8); | |
127 __ stx(O5, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+5*8); | |
128 #endif /* _LP64 */ | |
129 | |
130 __ save(SP, -frame_size, SP); | |
131 | |
132 #ifndef _LP64 | |
133 // Reload the 64 bit Oregs. Although they are now Iregs we load them | |
134 // to Oregs here to avoid interrupts cutting off their heads | |
135 | |
136 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8, O0); | |
137 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8, O1); | |
138 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+2*8, O2); | |
139 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+3*8, O3); | |
140 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+4*8, O4); | |
141 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+5*8, O5); | |
142 | |
143 __ stx(O0, SP, o0_offset+STACK_BIAS); | |
144 map->set_callee_saved(VMRegImpl::stack2reg((o0_offset + 4)>>2), O0->as_VMReg()); | |
145 | |
146 __ stx(O1, SP, o1_offset+STACK_BIAS); | |
147 | |
148 map->set_callee_saved(VMRegImpl::stack2reg((o1_offset + 4)>>2), O1->as_VMReg()); | |
149 | |
150 __ stx(O2, SP, o2_offset+STACK_BIAS); | |
151 map->set_callee_saved(VMRegImpl::stack2reg((o2_offset + 4)>>2), O2->as_VMReg()); | |
152 | |
153 __ stx(O3, SP, o3_offset+STACK_BIAS); | |
154 map->set_callee_saved(VMRegImpl::stack2reg((o3_offset + 4)>>2), O3->as_VMReg()); | |
155 | |
156 __ stx(O4, SP, o4_offset+STACK_BIAS); | |
157 map->set_callee_saved(VMRegImpl::stack2reg((o4_offset + 4)>>2), O4->as_VMReg()); | |
158 | |
159 __ stx(O5, SP, o5_offset+STACK_BIAS); | |
160 map->set_callee_saved(VMRegImpl::stack2reg((o5_offset + 4)>>2), O5->as_VMReg()); | |
161 #endif /* _LP64 */ | |
162 | |
163 // Save the G's | |
164 __ stx(G1, SP, g1_offset+STACK_BIAS); | |
165 map->set_callee_saved(VMRegImpl::stack2reg((g1_offset + 4)>>2), G1->as_VMReg()); | |
166 | |
167 __ stx(G3, SP, g3_offset+STACK_BIAS); | |
168 map->set_callee_saved(VMRegImpl::stack2reg((g3_offset + 4)>>2), G3->as_VMReg()); | |
169 | |
170 __ stx(G4, SP, g4_offset+STACK_BIAS); | |
171 map->set_callee_saved(VMRegImpl::stack2reg((g4_offset + 4)>>2), G4->as_VMReg()); | |
172 | |
173 __ stx(G5, SP, g5_offset+STACK_BIAS); | |
174 map->set_callee_saved(VMRegImpl::stack2reg((g5_offset + 4)>>2), G5->as_VMReg()); | |
175 | |
176 // This is really a waste but we'll keep things as they were for now | |
177 if (true) { | |
178 #ifndef _LP64 | |
179 map->set_callee_saved(VMRegImpl::stack2reg((o0_offset)>>2), O0->as_VMReg()->next()); | |
180 map->set_callee_saved(VMRegImpl::stack2reg((o1_offset)>>2), O1->as_VMReg()->next()); | |
181 map->set_callee_saved(VMRegImpl::stack2reg((o2_offset)>>2), O2->as_VMReg()->next()); | |
182 map->set_callee_saved(VMRegImpl::stack2reg((o3_offset)>>2), O3->as_VMReg()->next()); | |
183 map->set_callee_saved(VMRegImpl::stack2reg((o4_offset)>>2), O4->as_VMReg()->next()); | |
184 map->set_callee_saved(VMRegImpl::stack2reg((o5_offset)>>2), O5->as_VMReg()->next()); | |
185 #endif /* _LP64 */ | |
186 map->set_callee_saved(VMRegImpl::stack2reg((g1_offset)>>2), G1->as_VMReg()->next()); | |
187 map->set_callee_saved(VMRegImpl::stack2reg((g3_offset)>>2), G3->as_VMReg()->next()); | |
188 map->set_callee_saved(VMRegImpl::stack2reg((g4_offset)>>2), G4->as_VMReg()->next()); | |
189 map->set_callee_saved(VMRegImpl::stack2reg((g5_offset)>>2), G5->as_VMReg()->next()); | |
190 } | |
191 | |
192 | |
193 // Save the flags | |
194 __ rdccr( G5 ); | |
195 __ stx(G5, SP, ccr_offset+STACK_BIAS); | |
196 __ stxfsr(SP, fsr_offset+STACK_BIAS); | |
197 | |
198 // Save all the FP registers | |
199 int offset = d00_offset; | |
200 for( int i=0; i<64; i+=2 ) { | |
201 FloatRegister f = as_FloatRegister(i); | |
202 __ stf(FloatRegisterImpl::D, f, SP, offset+STACK_BIAS); | |
203 map->set_callee_saved(VMRegImpl::stack2reg(offset>>2), f->as_VMReg()); | |
204 if (true) { | |
205 map->set_callee_saved(VMRegImpl::stack2reg((offset + sizeof(float))>>2), f->as_VMReg()->next()); | |
206 } | |
207 offset += sizeof(double); | |
208 } | |
209 | |
210 // And we're done. | |
211 | |
212 return map; | |
213 } | |
214 | |
215 | |
216 // Pop the current frame and restore all the registers that we | |
217 // saved. | |
218 void RegisterSaver::restore_live_registers(MacroAssembler* masm) { | |
219 | |
220 // Restore all the FP registers | |
221 for( int i=0; i<64; i+=2 ) { | |
222 __ ldf(FloatRegisterImpl::D, SP, d00_offset+i*sizeof(float)+STACK_BIAS, as_FloatRegister(i)); | |
223 } | |
224 | |
225 __ ldx(SP, ccr_offset+STACK_BIAS, G1); | |
226 __ wrccr (G1) ; | |
227 | |
228 // Restore the G's | |
229 // Note that G2 (AKA GThread) must be saved and restored separately. | |
230 // TODO-FIXME: save and restore some of the other ASRs, viz., %asi and %gsr. | |
231 | |
232 __ ldx(SP, g1_offset+STACK_BIAS, G1); | |
233 __ ldx(SP, g3_offset+STACK_BIAS, G3); | |
234 __ ldx(SP, g4_offset+STACK_BIAS, G4); | |
235 __ ldx(SP, g5_offset+STACK_BIAS, G5); | |
236 | |
237 | |
238 #if !defined(_LP64) | |
239 // Restore the 64-bit O's. | |
240 __ ldx(SP, o0_offset+STACK_BIAS, O0); | |
241 __ ldx(SP, o1_offset+STACK_BIAS, O1); | |
242 __ ldx(SP, o2_offset+STACK_BIAS, O2); | |
243 __ ldx(SP, o3_offset+STACK_BIAS, O3); | |
244 __ ldx(SP, o4_offset+STACK_BIAS, O4); | |
245 __ ldx(SP, o5_offset+STACK_BIAS, O5); | |
246 | |
247 // And temporarily place them in TLS | |
248 | |
249 __ stx(O0, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8); | |
250 __ stx(O1, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8); | |
251 __ stx(O2, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+2*8); | |
252 __ stx(O3, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+3*8); | |
253 __ stx(O4, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+4*8); | |
254 __ stx(O5, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+5*8); | |
255 #endif /* _LP64 */ | |
256 | |
257 // Restore flags | |
258 | |
259 __ ldxfsr(SP, fsr_offset+STACK_BIAS); | |
260 | |
261 __ restore(); | |
262 | |
263 #if !defined(_LP64) | |
264 // Now reload the 64bit Oregs after we've restore the window. | |
265 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8, O0); | |
266 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8, O1); | |
267 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+2*8, O2); | |
268 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+3*8, O3); | |
269 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+4*8, O4); | |
270 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+5*8, O5); | |
271 #endif /* _LP64 */ | |
272 | |
273 } | |
274 | |
275 // Pop the current frame and restore the registers that might be holding | |
276 // a result. | |
277 void RegisterSaver::restore_result_registers(MacroAssembler* masm) { | |
278 | |
279 #if !defined(_LP64) | |
280 // 32bit build returns longs in G1 | |
281 __ ldx(SP, g1_offset+STACK_BIAS, G1); | |
282 | |
283 // Retrieve the 64-bit O's. | |
284 __ ldx(SP, o0_offset+STACK_BIAS, O0); | |
285 __ ldx(SP, o1_offset+STACK_BIAS, O1); | |
286 // and save to TLS | |
287 __ stx(O0, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8); | |
288 __ stx(O1, G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8); | |
289 #endif /* _LP64 */ | |
290 | |
291 __ ldf(FloatRegisterImpl::D, SP, d00_offset+STACK_BIAS, as_FloatRegister(0)); | |
292 | |
293 __ restore(); | |
294 | |
295 #if !defined(_LP64) | |
296 // Now reload the 64bit Oregs after we've restore the window. | |
297 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+0*8, O0); | |
298 __ ldx(G2_thread, JavaThread::o_reg_temps_offset_in_bytes()+1*8, O1); | |
299 #endif /* _LP64 */ | |
300 | |
301 } | |
302 | |
303 // The java_calling_convention describes stack locations as ideal slots on | |
304 // a frame with no abi restrictions. Since we must observe abi restrictions | |
305 // (like the placement of the register window) the slots must be biased by | |
306 // the following value. | |
307 static int reg2offset(VMReg r) { | |
308 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; | |
309 } | |
310 | |
311 // --------------------------------------------------------------------------- | |
312 // Read the array of BasicTypes from a signature, and compute where the | |
313 // arguments should go. Values in the VMRegPair regs array refer to 4-byte (VMRegImpl::stack_slot_size) | |
314 // quantities. Values less than VMRegImpl::stack0 are registers, those above | |
315 // refer to 4-byte stack slots. All stack slots are based off of the window | |
316 // top. VMRegImpl::stack0 refers to the first slot past the 16-word window, | |
317 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register | |
318 // values 0-63 (up to RegisterImpl::number_of_registers) are the 64-bit | |
319 // integer registers. Values 64-95 are the (32-bit only) float registers. | |
320 // Each 32-bit quantity is given its own number, so the integer registers | |
321 // (in either 32- or 64-bit builds) use 2 numbers. For example, there is | |
322 // an O0-low and an O0-high. Essentially, all int register numbers are doubled. | |
323 | |
324 // Register results are passed in O0-O5, for outgoing call arguments. To | |
325 // convert to incoming arguments, convert all O's to I's. The regs array | |
326 // refer to the low and hi 32-bit words of 64-bit registers or stack slots. | |
327 // If the regs[].second() field is set to VMRegImpl::Bad(), it means it's unused (a | |
328 // 32-bit value was passed). If both are VMRegImpl::Bad(), it means no value was | |
329 // passed (used as a placeholder for the other half of longs and doubles in | |
330 // the 64-bit build). regs[].second() is either VMRegImpl::Bad() or regs[].second() is | |
331 // regs[].first()+1 (regs[].first() may be misaligned in the C calling convention). | |
332 // Sparc never passes a value in regs[].second() but not regs[].first() (regs[].first() | |
333 // == VMRegImpl::Bad() && regs[].second() != VMRegImpl::Bad()) nor unrelated values in the | |
334 // same VMRegPair. | |
335 | |
336 // Note: the INPUTS in sig_bt are in units of Java argument words, which are | |
337 // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit | |
338 // units regardless of build. | |
339 | |
340 | |
341 // --------------------------------------------------------------------------- | |
342 // The compiled Java calling convention. The Java convention always passes | |
343 // 64-bit values in adjacent aligned locations (either registers or stack), | |
344 // floats in float registers and doubles in aligned float pairs. Values are | |
345 // packed in the registers. There is no backing varargs store for values in | |
346 // registers. In the 32-bit build, longs are passed in G1 and G4 (cannot be | |
347 // passed in I's, because longs in I's get their heads chopped off at | |
348 // interrupt). | |
349 int SharedRuntime::java_calling_convention(const BasicType *sig_bt, | |
350 VMRegPair *regs, | |
351 int total_args_passed, | |
352 int is_outgoing) { | |
353 assert(F31->as_VMReg()->is_reg(), "overlapping stack/register numbers"); | |
354 | |
355 // Convention is to pack the first 6 int/oop args into the first 6 registers | |
356 // (I0-I5), extras spill to the stack. Then pack the first 8 float args | |
357 // into F0-F7, extras spill to the stack. Then pad all register sets to | |
358 // align. Then put longs and doubles into the same registers as they fit, | |
359 // else spill to the stack. | |
360 const int int_reg_max = SPARC_ARGS_IN_REGS_NUM; | |
361 const int flt_reg_max = 8; | |
362 // | |
363 // Where 32-bit 1-reg longs start being passed | |
364 // In tiered we must pass on stack because c1 can't use a "pair" in a single reg. | |
365 // So make it look like we've filled all the G regs that c2 wants to use. | |
366 Register g_reg = TieredCompilation ? noreg : G1; | |
367 | |
368 // Count int/oop and float args. See how many stack slots we'll need and | |
369 // where the longs & doubles will go. | |
370 int int_reg_cnt = 0; | |
371 int flt_reg_cnt = 0; | |
372 // int stk_reg_pairs = frame::register_save_words*(wordSize>>2); | |
373 // int stk_reg_pairs = SharedRuntime::out_preserve_stack_slots(); | |
374 int stk_reg_pairs = 0; | |
375 for (int i = 0; i < total_args_passed; i++) { | |
376 switch (sig_bt[i]) { | |
377 case T_LONG: // LP64, longs compete with int args | |
378 assert(sig_bt[i+1] == T_VOID, ""); | |
379 #ifdef _LP64 | |
380 if (int_reg_cnt < int_reg_max) int_reg_cnt++; | |
381 #endif | |
382 break; | |
383 case T_OBJECT: | |
384 case T_ARRAY: | |
385 case T_ADDRESS: // Used, e.g., in slow-path locking for the lock's stack address | |
386 if (int_reg_cnt < int_reg_max) int_reg_cnt++; | |
387 #ifndef _LP64 | |
388 else stk_reg_pairs++; | |
389 #endif | |
390 break; | |
391 case T_INT: | |
392 case T_SHORT: | |
393 case T_CHAR: | |
394 case T_BYTE: | |
395 case T_BOOLEAN: | |
396 if (int_reg_cnt < int_reg_max) int_reg_cnt++; | |
397 else stk_reg_pairs++; | |
398 break; | |
399 case T_FLOAT: | |
400 if (flt_reg_cnt < flt_reg_max) flt_reg_cnt++; | |
401 else stk_reg_pairs++; | |
402 break; | |
403 case T_DOUBLE: | |
404 assert(sig_bt[i+1] == T_VOID, ""); | |
405 break; | |
406 case T_VOID: | |
407 break; | |
408 default: | |
409 ShouldNotReachHere(); | |
410 } | |
411 } | |
412 | |
413 // This is where the longs/doubles start on the stack. | |
414 stk_reg_pairs = (stk_reg_pairs+1) & ~1; // Round | |
415 | |
416 int int_reg_pairs = (int_reg_cnt+1) & ~1; // 32-bit 2-reg longs only | |
417 int flt_reg_pairs = (flt_reg_cnt+1) & ~1; | |
418 | |
419 // int stk_reg = frame::register_save_words*(wordSize>>2); | |
420 // int stk_reg = SharedRuntime::out_preserve_stack_slots(); | |
421 int stk_reg = 0; | |
422 int int_reg = 0; | |
423 int flt_reg = 0; | |
424 | |
425 // Now do the signature layout | |
426 for (int i = 0; i < total_args_passed; i++) { | |
427 switch (sig_bt[i]) { | |
428 case T_INT: | |
429 case T_SHORT: | |
430 case T_CHAR: | |
431 case T_BYTE: | |
432 case T_BOOLEAN: | |
433 #ifndef _LP64 | |
434 case T_OBJECT: | |
435 case T_ARRAY: | |
436 case T_ADDRESS: // Used, e.g., in slow-path locking for the lock's stack address | |
437 #endif // _LP64 | |
438 if (int_reg < int_reg_max) { | |
439 Register r = is_outgoing ? as_oRegister(int_reg++) : as_iRegister(int_reg++); | |
440 regs[i].set1(r->as_VMReg()); | |
441 } else { | |
442 regs[i].set1(VMRegImpl::stack2reg(stk_reg++)); | |
443 } | |
444 break; | |
445 | |
446 #ifdef _LP64 | |
447 case T_OBJECT: | |
448 case T_ARRAY: | |
449 case T_ADDRESS: // Used, e.g., in slow-path locking for the lock's stack address | |
450 if (int_reg < int_reg_max) { | |
451 Register r = is_outgoing ? as_oRegister(int_reg++) : as_iRegister(int_reg++); | |
452 regs[i].set2(r->as_VMReg()); | |
453 } else { | |
454 regs[i].set2(VMRegImpl::stack2reg(stk_reg_pairs)); | |
455 stk_reg_pairs += 2; | |
456 } | |
457 break; | |
458 #endif // _LP64 | |
459 | |
460 case T_LONG: | |
461 assert(sig_bt[i+1] == T_VOID, "expecting VOID in other half"); | |
462 #ifdef COMPILER2 | |
463 #ifdef _LP64 | |
464 // Can't be tiered (yet) | |
465 if (int_reg < int_reg_max) { | |
466 Register r = is_outgoing ? as_oRegister(int_reg++) : as_iRegister(int_reg++); | |
467 regs[i].set2(r->as_VMReg()); | |
468 } else { | |
469 regs[i].set2(VMRegImpl::stack2reg(stk_reg_pairs)); | |
470 stk_reg_pairs += 2; | |
471 } | |
472 #else | |
473 // For 32-bit build, can't pass longs in O-regs because they become | |
474 // I-regs and get trashed. Use G-regs instead. G1 and G4 are almost | |
475 // spare and available. This convention isn't used by the Sparc ABI or | |
476 // anywhere else. If we're tiered then we don't use G-regs because c1 | |
477 // can't deal with them as a "pair". | |
478 // G0: zero | |
479 // G1: 1st Long arg | |
480 // G2: global allocated to TLS | |
481 // G3: used in inline cache check | |
482 // G4: 2nd Long arg | |
483 // G5: used in inline cache check | |
484 // G6: used by OS | |
485 // G7: used by OS | |
486 | |
487 if (g_reg == G1) { | |
488 regs[i].set2(G1->as_VMReg()); // This long arg in G1 | |
489 g_reg = G4; // Where the next arg goes | |
490 } else if (g_reg == G4) { | |
491 regs[i].set2(G4->as_VMReg()); // The 2nd long arg in G4 | |
492 g_reg = noreg; // No more longs in registers | |
493 } else { | |
494 regs[i].set2(VMRegImpl::stack2reg(stk_reg_pairs)); | |
495 stk_reg_pairs += 2; | |
496 } | |
497 #endif // _LP64 | |
498 #else // COMPILER2 | |
499 if (int_reg_pairs + 1 < int_reg_max) { | |
500 if (is_outgoing) { | |
501 regs[i].set_pair(as_oRegister(int_reg_pairs + 1)->as_VMReg(), as_oRegister(int_reg_pairs)->as_VMReg()); | |
502 } else { | |
503 regs[i].set_pair(as_iRegister(int_reg_pairs + 1)->as_VMReg(), as_iRegister(int_reg_pairs)->as_VMReg()); | |
504 } | |
505 int_reg_pairs += 2; | |
506 } else { | |
507 regs[i].set2(VMRegImpl::stack2reg(stk_reg_pairs)); | |
508 stk_reg_pairs += 2; | |
509 } | |
510 #endif // COMPILER2 | |
511 break; | |
512 | |
513 case T_FLOAT: | |
514 if (flt_reg < flt_reg_max) regs[i].set1(as_FloatRegister(flt_reg++)->as_VMReg()); | |
515 else regs[i].set1( VMRegImpl::stack2reg(stk_reg++)); | |
516 break; | |
517 case T_DOUBLE: | |
518 assert(sig_bt[i+1] == T_VOID, "expecting half"); | |
519 if (flt_reg_pairs + 1 < flt_reg_max) { | |
520 regs[i].set2(as_FloatRegister(flt_reg_pairs)->as_VMReg()); | |
521 flt_reg_pairs += 2; | |
522 } else { | |
523 regs[i].set2(VMRegImpl::stack2reg(stk_reg_pairs)); | |
524 stk_reg_pairs += 2; | |
525 } | |
526 break; | |
527 case T_VOID: regs[i].set_bad(); break; // Halves of longs & doubles | |
528 default: | |
529 ShouldNotReachHere(); | |
530 } | |
531 } | |
532 | |
533 // retun the amount of stack space these arguments will need. | |
534 return stk_reg_pairs; | |
535 | |
536 } | |
537 | |
538 // Helper class mostly to avoid passing masm everywhere, and handle store | |
539 // displacement overflow logic for LP64 | |
540 class AdapterGenerator { | |
541 MacroAssembler *masm; | |
542 #ifdef _LP64 | |
543 Register Rdisp; | |
544 void set_Rdisp(Register r) { Rdisp = r; } | |
545 #endif // _LP64 | |
546 | |
547 void patch_callers_callsite(); | |
548 void tag_c2i_arg(frame::Tag t, Register base, int st_off, Register scratch); | |
549 | |
550 // base+st_off points to top of argument | |
551 int arg_offset(const int st_off) { return st_off + Interpreter::value_offset_in_bytes(); } | |
552 int next_arg_offset(const int st_off) { | |
553 return st_off - Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); | |
554 } | |
555 | |
556 #ifdef _LP64 | |
557 // On _LP64 argument slot values are loaded first into a register | |
558 // because they might not fit into displacement. | |
559 Register arg_slot(const int st_off); | |
560 Register next_arg_slot(const int st_off); | |
561 #else | |
562 int arg_slot(const int st_off) { return arg_offset(st_off); } | |
563 int next_arg_slot(const int st_off) { return next_arg_offset(st_off); } | |
564 #endif // _LP64 | |
565 | |
566 // Stores long into offset pointed to by base | |
567 void store_c2i_long(Register r, Register base, | |
568 const int st_off, bool is_stack); | |
569 void store_c2i_object(Register r, Register base, | |
570 const int st_off); | |
571 void store_c2i_int(Register r, Register base, | |
572 const int st_off); | |
573 void store_c2i_double(VMReg r_2, | |
574 VMReg r_1, Register base, const int st_off); | |
575 void store_c2i_float(FloatRegister f, Register base, | |
576 const int st_off); | |
577 | |
578 public: | |
579 void gen_c2i_adapter(int total_args_passed, | |
580 // VMReg max_arg, | |
581 int comp_args_on_stack, // VMRegStackSlots | |
582 const BasicType *sig_bt, | |
583 const VMRegPair *regs, | |
584 Label& skip_fixup); | |
585 void gen_i2c_adapter(int total_args_passed, | |
586 // VMReg max_arg, | |
587 int comp_args_on_stack, // VMRegStackSlots | |
588 const BasicType *sig_bt, | |
589 const VMRegPair *regs); | |
590 | |
591 AdapterGenerator(MacroAssembler *_masm) : masm(_masm) {} | |
592 }; | |
593 | |
594 | |
595 // Patch the callers callsite with entry to compiled code if it exists. | |
596 void AdapterGenerator::patch_callers_callsite() { | |
597 Label L; | |
598 __ ld_ptr(G5_method, in_bytes(methodOopDesc::code_offset()), G3_scratch); | |
599 __ br_null(G3_scratch, false, __ pt, L); | |
600 // Schedule the branch target address early. | |
601 __ delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), G3_scratch); | |
602 // Call into the VM to patch the caller, then jump to compiled callee | |
603 __ save_frame(4); // Args in compiled layout; do not blow them | |
604 | |
605 // Must save all the live Gregs the list is: | |
606 // G1: 1st Long arg (32bit build) | |
607 // G2: global allocated to TLS | |
608 // G3: used in inline cache check (scratch) | |
609 // G4: 2nd Long arg (32bit build); | |
610 // G5: used in inline cache check (methodOop) | |
611 | |
612 // The longs must go to the stack by hand since in the 32 bit build they can be trashed by window ops. | |
613 | |
614 #ifdef _LP64 | |
615 // mov(s,d) | |
616 __ mov(G1, L1); | |
617 __ mov(G4, L4); | |
618 __ mov(G5_method, L5); | |
619 __ mov(G5_method, O0); // VM needs target method | |
620 __ mov(I7, O1); // VM needs caller's callsite | |
621 // Must be a leaf call... | |
622 // can be very far once the blob has been relocated | |
623 Address dest(O7, CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)); | |
624 __ relocate(relocInfo::runtime_call_type); | |
625 __ jumpl_to(dest, O7); | |
626 __ delayed()->mov(G2_thread, L7_thread_cache); | |
627 __ mov(L7_thread_cache, G2_thread); | |
628 __ mov(L1, G1); | |
629 __ mov(L4, G4); | |
630 __ mov(L5, G5_method); | |
631 #else | |
632 __ stx(G1, FP, -8 + STACK_BIAS); | |
633 __ stx(G4, FP, -16 + STACK_BIAS); | |
634 __ mov(G5_method, L5); | |
635 __ mov(G5_method, O0); // VM needs target method | |
636 __ mov(I7, O1); // VM needs caller's callsite | |
637 // Must be a leaf call... | |
638 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite), relocInfo::runtime_call_type); | |
639 __ delayed()->mov(G2_thread, L7_thread_cache); | |
640 __ mov(L7_thread_cache, G2_thread); | |
641 __ ldx(FP, -8 + STACK_BIAS, G1); | |
642 __ ldx(FP, -16 + STACK_BIAS, G4); | |
643 __ mov(L5, G5_method); | |
644 __ ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), G3_scratch); | |
645 #endif /* _LP64 */ | |
646 | |
647 __ restore(); // Restore args | |
648 __ bind(L); | |
649 } | |
650 | |
651 void AdapterGenerator::tag_c2i_arg(frame::Tag t, Register base, int st_off, | |
652 Register scratch) { | |
653 if (TaggedStackInterpreter) { | |
654 int tag_off = st_off + Interpreter::tag_offset_in_bytes(); | |
655 #ifdef _LP64 | |
656 Register tag_slot = Rdisp; | |
657 __ set(tag_off, tag_slot); | |
658 #else | |
659 int tag_slot = tag_off; | |
660 #endif // _LP64 | |
661 // have to store zero because local slots can be reused (rats!) | |
662 if (t == frame::TagValue) { | |
663 __ st_ptr(G0, base, tag_slot); | |
664 } else if (t == frame::TagCategory2) { | |
665 __ st_ptr(G0, base, tag_slot); | |
666 int next_tag_off = st_off - Interpreter::stackElementSize() + | |
667 Interpreter::tag_offset_in_bytes(); | |
668 #ifdef _LP64 | |
669 __ set(next_tag_off, tag_slot); | |
670 #else | |
671 tag_slot = next_tag_off; | |
672 #endif // _LP64 | |
673 __ st_ptr(G0, base, tag_slot); | |
674 } else { | |
675 __ mov(t, scratch); | |
676 __ st_ptr(scratch, base, tag_slot); | |
677 } | |
678 } | |
679 } | |
680 | |
681 #ifdef _LP64 | |
682 Register AdapterGenerator::arg_slot(const int st_off) { | |
683 __ set( arg_offset(st_off), Rdisp); | |
684 return Rdisp; | |
685 } | |
686 | |
687 Register AdapterGenerator::next_arg_slot(const int st_off){ | |
688 __ set( next_arg_offset(st_off), Rdisp); | |
689 return Rdisp; | |
690 } | |
691 #endif // _LP64 | |
692 | |
693 // Stores long into offset pointed to by base | |
694 void AdapterGenerator::store_c2i_long(Register r, Register base, | |
695 const int st_off, bool is_stack) { | |
696 #ifdef COMPILER2 | |
697 #ifdef _LP64 | |
698 // In V9, longs are given 2 64-bit slots in the interpreter, but the | |
699 // data is passed in only 1 slot. | |
700 __ stx(r, base, next_arg_slot(st_off)); | |
701 #else | |
702 // Misaligned store of 64-bit data | |
703 __ stw(r, base, arg_slot(st_off)); // lo bits | |
704 __ srlx(r, 32, r); | |
705 __ stw(r, base, next_arg_slot(st_off)); // hi bits | |
706 #endif // _LP64 | |
707 #else | |
708 if (is_stack) { | |
709 // Misaligned store of 64-bit data | |
710 __ stw(r, base, arg_slot(st_off)); // lo bits | |
711 __ srlx(r, 32, r); | |
712 __ stw(r, base, next_arg_slot(st_off)); // hi bits | |
713 } else { | |
714 __ stw(r->successor(), base, arg_slot(st_off) ); // lo bits | |
715 __ stw(r , base, next_arg_slot(st_off)); // hi bits | |
716 } | |
717 #endif // COMPILER2 | |
718 tag_c2i_arg(frame::TagCategory2, base, st_off, r); | |
719 } | |
720 | |
721 void AdapterGenerator::store_c2i_object(Register r, Register base, | |
722 const int st_off) { | |
723 __ st_ptr (r, base, arg_slot(st_off)); | |
724 tag_c2i_arg(frame::TagReference, base, st_off, r); | |
725 } | |
726 | |
727 void AdapterGenerator::store_c2i_int(Register r, Register base, | |
728 const int st_off) { | |
729 __ st (r, base, arg_slot(st_off)); | |
730 tag_c2i_arg(frame::TagValue, base, st_off, r); | |
731 } | |
732 | |
733 // Stores into offset pointed to by base | |
734 void AdapterGenerator::store_c2i_double(VMReg r_2, | |
735 VMReg r_1, Register base, const int st_off) { | |
736 #ifdef _LP64 | |
737 // In V9, doubles are given 2 64-bit slots in the interpreter, but the | |
738 // data is passed in only 1 slot. | |
739 __ stf(FloatRegisterImpl::D, r_1->as_FloatRegister(), base, next_arg_slot(st_off)); | |
740 #else | |
741 // Need to marshal 64-bit value from misaligned Lesp loads | |
742 __ stf(FloatRegisterImpl::S, r_1->as_FloatRegister(), base, next_arg_slot(st_off)); | |
743 __ stf(FloatRegisterImpl::S, r_2->as_FloatRegister(), base, arg_slot(st_off) ); | |
744 #endif | |
745 tag_c2i_arg(frame::TagCategory2, base, st_off, G1_scratch); | |
746 } | |
747 | |
748 void AdapterGenerator::store_c2i_float(FloatRegister f, Register base, | |
749 const int st_off) { | |
750 __ stf(FloatRegisterImpl::S, f, base, arg_slot(st_off)); | |
751 tag_c2i_arg(frame::TagValue, base, st_off, G1_scratch); | |
752 } | |
753 | |
754 void AdapterGenerator::gen_c2i_adapter( | |
755 int total_args_passed, | |
756 // VMReg max_arg, | |
757 int comp_args_on_stack, // VMRegStackSlots | |
758 const BasicType *sig_bt, | |
759 const VMRegPair *regs, | |
760 Label& skip_fixup) { | |
761 | |
762 // Before we get into the guts of the C2I adapter, see if we should be here | |
763 // at all. We've come from compiled code and are attempting to jump to the | |
764 // interpreter, which means the caller made a static call to get here | |
765 // (vcalls always get a compiled target if there is one). Check for a | |
766 // compiled target. If there is one, we need to patch the caller's call. | |
767 // However we will run interpreted if we come thru here. The next pass | |
768 // thru the call site will run compiled. If we ran compiled here then | |
769 // we can (theorectically) do endless i2c->c2i->i2c transitions during | |
770 // deopt/uncommon trap cycles. If we always go interpreted here then | |
771 // we can have at most one and don't need to play any tricks to keep | |
772 // from endlessly growing the stack. | |
773 // | |
774 // Actually if we detected that we had an i2c->c2i transition here we | |
775 // ought to be able to reset the world back to the state of the interpreted | |
776 // call and not bother building another interpreter arg area. We don't | |
777 // do that at this point. | |
778 | |
779 patch_callers_callsite(); | |
780 | |
781 __ bind(skip_fixup); | |
782 | |
783 // Since all args are passed on the stack, total_args_passed*wordSize is the | |
784 // space we need. Add in varargs area needed by the interpreter. Round up | |
785 // to stack alignment. | |
786 const int arg_size = total_args_passed * Interpreter::stackElementSize(); | |
787 const int varargs_area = | |
788 (frame::varargs_offset - frame::register_save_words)*wordSize; | |
789 const int extraspace = round_to(arg_size + varargs_area, 2*wordSize); | |
790 | |
791 int bias = STACK_BIAS; | |
792 const int interp_arg_offset = frame::varargs_offset*wordSize + | |
793 (total_args_passed-1)*Interpreter::stackElementSize(); | |
794 | |
795 Register base = SP; | |
796 | |
797 #ifdef _LP64 | |
798 // In the 64bit build because of wider slots and STACKBIAS we can run | |
799 // out of bits in the displacement to do loads and stores. Use g3 as | |
800 // temporary displacement. | |
801 if (! __ is_simm13(extraspace)) { | |
802 __ set(extraspace, G3_scratch); | |
803 __ sub(SP, G3_scratch, SP); | |
804 } else { | |
805 __ sub(SP, extraspace, SP); | |
806 } | |
807 set_Rdisp(G3_scratch); | |
808 #else | |
809 __ sub(SP, extraspace, SP); | |
810 #endif // _LP64 | |
811 | |
812 // First write G1 (if used) to where ever it must go | |
813 for (int i=0; i<total_args_passed; i++) { | |
814 const int st_off = interp_arg_offset - (i*Interpreter::stackElementSize()) + bias; | |
815 VMReg r_1 = regs[i].first(); | |
816 VMReg r_2 = regs[i].second(); | |
817 if (r_1 == G1_scratch->as_VMReg()) { | |
818 if (sig_bt[i] == T_OBJECT || sig_bt[i] == T_ARRAY) { | |
819 store_c2i_object(G1_scratch, base, st_off); | |
820 } else if (sig_bt[i] == T_LONG) { | |
821 assert(!TieredCompilation, "should not use register args for longs"); | |
822 store_c2i_long(G1_scratch, base, st_off, false); | |
823 } else { | |
824 store_c2i_int(G1_scratch, base, st_off); | |
825 } | |
826 } | |
827 } | |
828 | |
829 // Now write the args into the outgoing interpreter space | |
830 for (int i=0; i<total_args_passed; i++) { | |
831 const int st_off = interp_arg_offset - (i*Interpreter::stackElementSize()) + bias; | |
832 VMReg r_1 = regs[i].first(); | |
833 VMReg r_2 = regs[i].second(); | |
834 if (!r_1->is_valid()) { | |
835 assert(!r_2->is_valid(), ""); | |
836 continue; | |
837 } | |
838 // Skip G1 if found as we did it first in order to free it up | |
839 if (r_1 == G1_scratch->as_VMReg()) { | |
840 continue; | |
841 } | |
842 #ifdef ASSERT | |
843 bool G1_forced = false; | |
844 #endif // ASSERT | |
845 if (r_1->is_stack()) { // Pretend stack targets are loaded into G1 | |
846 #ifdef _LP64 | |
847 Register ld_off = Rdisp; | |
848 __ set(reg2offset(r_1) + extraspace + bias, ld_off); | |
849 #else | |
850 int ld_off = reg2offset(r_1) + extraspace + bias; | |
851 #ifdef ASSERT | |
852 G1_forced = true; | |
853 #endif // ASSERT | |
854 #endif // _LP64 | |
855 r_1 = G1_scratch->as_VMReg();// as part of the load/store shuffle | |
856 if (!r_2->is_valid()) __ ld (base, ld_off, G1_scratch); | |
857 else __ ldx(base, ld_off, G1_scratch); | |
858 } | |
859 | |
860 if (r_1->is_Register()) { | |
861 Register r = r_1->as_Register()->after_restore(); | |
862 if (sig_bt[i] == T_OBJECT || sig_bt[i] == T_ARRAY) { | |
863 store_c2i_object(r, base, st_off); | |
864 } else if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { | |
865 if (TieredCompilation) { | |
866 assert(G1_forced || sig_bt[i] != T_LONG, "should not use register args for longs"); | |
867 } | |
868 store_c2i_long(r, base, st_off, r_2->is_stack()); | |
869 } else { | |
870 store_c2i_int(r, base, st_off); | |
871 } | |
872 } else { | |
873 assert(r_1->is_FloatRegister(), ""); | |
874 if (sig_bt[i] == T_FLOAT) { | |
875 store_c2i_float(r_1->as_FloatRegister(), base, st_off); | |
876 } else { | |
877 assert(sig_bt[i] == T_DOUBLE, "wrong type"); | |
878 store_c2i_double(r_2, r_1, base, st_off); | |
879 } | |
880 } | |
881 } | |
882 | |
883 #ifdef _LP64 | |
884 // Need to reload G3_scratch, used for temporary displacements. | |
885 __ ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), G3_scratch); | |
886 | |
887 // Pass O5_savedSP as an argument to the interpreter. | |
888 // The interpreter will restore SP to this value before returning. | |
889 __ set(extraspace, G1); | |
890 __ add(SP, G1, O5_savedSP); | |
891 #else | |
892 // Pass O5_savedSP as an argument to the interpreter. | |
893 // The interpreter will restore SP to this value before returning. | |
894 __ add(SP, extraspace, O5_savedSP); | |
895 #endif // _LP64 | |
896 | |
897 __ mov((frame::varargs_offset)*wordSize - | |
898 1*Interpreter::stackElementSize()+bias+BytesPerWord, G1); | |
899 // Jump to the interpreter just as if interpreter was doing it. | |
900 __ jmpl(G3_scratch, 0, G0); | |
901 // Setup Lesp for the call. Cannot actually set Lesp as the current Lesp | |
902 // (really L0) is in use by the compiled frame as a generic temp. However, | |
903 // the interpreter does not know where its args are without some kind of | |
904 // arg pointer being passed in. Pass it in Gargs. | |
905 __ delayed()->add(SP, G1, Gargs); | |
906 } | |
907 | |
908 void AdapterGenerator::gen_i2c_adapter( | |
909 int total_args_passed, | |
910 // VMReg max_arg, | |
911 int comp_args_on_stack, // VMRegStackSlots | |
912 const BasicType *sig_bt, | |
913 const VMRegPair *regs) { | |
914 | |
915 // Generate an I2C adapter: adjust the I-frame to make space for the C-frame | |
916 // layout. Lesp was saved by the calling I-frame and will be restored on | |
917 // return. Meanwhile, outgoing arg space is all owned by the callee | |
918 // C-frame, so we can mangle it at will. After adjusting the frame size, | |
919 // hoist register arguments and repack other args according to the compiled | |
920 // code convention. Finally, end in a jump to the compiled code. The entry | |
921 // point address is the start of the buffer. | |
922 | |
923 // We will only enter here from an interpreted frame and never from after | |
924 // passing thru a c2i. Azul allowed this but we do not. If we lose the | |
925 // race and use a c2i we will remain interpreted for the race loser(s). | |
926 // This removes all sorts of headaches on the x86 side and also eliminates | |
927 // the possibility of having c2i -> i2c -> c2i -> ... endless transitions. | |
928 | |
929 // As you can see from the list of inputs & outputs there are not a lot | |
930 // of temp registers to work with: mostly G1, G3 & G4. | |
931 | |
932 // Inputs: | |
933 // G2_thread - TLS | |
934 // G5_method - Method oop | |
935 // O0 - Flag telling us to restore SP from O5 | |
936 // O4_args - Pointer to interpreter's args | |
937 // O5 - Caller's saved SP, to be restored if needed | |
938 // O6 - Current SP! | |
939 // O7 - Valid return address | |
940 // L0-L7, I0-I7 - Caller's temps (no frame pushed yet) | |
941 | |
942 // Outputs: | |
943 // G2_thread - TLS | |
944 // G1, G4 - Outgoing long args in 32-bit build | |
945 // O0-O5 - Outgoing args in compiled layout | |
946 // O6 - Adjusted or restored SP | |
947 // O7 - Valid return address | |
948 // L0-L7, I0-I7 - Caller's temps (no frame pushed yet) | |
949 // F0-F7 - more outgoing args | |
950 | |
951 | |
952 // O4 is about to get loaded up with compiled callee's args | |
953 __ sub(Gargs, BytesPerWord, Gargs); | |
954 | |
955 #ifdef ASSERT | |
956 { | |
957 // on entry OsavedSP and SP should be equal | |
958 Label ok; | |
959 __ cmp(O5_savedSP, SP); | |
960 __ br(Assembler::equal, false, Assembler::pt, ok); | |
961 __ delayed()->nop(); | |
962 __ stop("I5_savedSP not set"); | |
963 __ should_not_reach_here(); | |
964 __ bind(ok); | |
965 } | |
966 #endif | |
967 | |
968 // ON ENTRY TO THE CODE WE ARE MAKING, WE HAVE AN INTERPRETED FRAME | |
969 // WITH O7 HOLDING A VALID RETURN PC | |
970 // | |
971 // | | | |
972 // : java stack : | |
973 // | | | |
974 // +--------------+ <--- start of outgoing args | |
975 // | receiver | | | |
976 // : rest of args : |---size is java-arg-words | |
977 // | | | | |
978 // +--------------+ <--- O4_args (misaligned) and Lesp if prior is not C2I | |
979 // | | | | |
980 // : unused : |---Space for max Java stack, plus stack alignment | |
981 // | | | | |
982 // +--------------+ <--- SP + 16*wordsize | |
983 // | | | |
984 // : window : | |
985 // | | | |
986 // +--------------+ <--- SP | |
987 | |
988 // WE REPACK THE STACK. We use the common calling convention layout as | |
989 // discovered by calling SharedRuntime::calling_convention. We assume it | |
990 // causes an arbitrary shuffle of memory, which may require some register | |
991 // temps to do the shuffle. We hope for (and optimize for) the case where | |
992 // temps are not needed. We may have to resize the stack slightly, in case | |
993 // we need alignment padding (32-bit interpreter can pass longs & doubles | |
994 // misaligned, but the compilers expect them aligned). | |
995 // | |
996 // | | | |
997 // : java stack : | |
998 // | | | |
999 // +--------------+ <--- start of outgoing args | |
1000 // | pad, align | | | |
1001 // +--------------+ | | |
1002 // | ints, floats | |---Outgoing stack args, packed low. | |
1003 // +--------------+ | First few args in registers. | |
1004 // : doubles : | | |
1005 // | longs | | | |
1006 // +--------------+ <--- SP' + 16*wordsize | |
1007 // | | | |
1008 // : window : | |
1009 // | | | |
1010 // +--------------+ <--- SP' | |
1011 | |
1012 // ON EXIT FROM THE CODE WE ARE MAKING, WE STILL HAVE AN INTERPRETED FRAME | |
1013 // WITH O7 HOLDING A VALID RETURN PC - ITS JUST THAT THE ARGS ARE NOW SETUP | |
1014 // FOR COMPILED CODE AND THE FRAME SLIGHTLY GROWN. | |
1015 | |
1016 // Cut-out for having no stack args. Since up to 6 args are passed | |
1017 // in registers, we will commonly have no stack args. | |
1018 if (comp_args_on_stack > 0) { | |
1019 | |
1020 // Convert VMReg stack slots to words. | |
1021 int comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord; | |
1022 // Round up to miminum stack alignment, in wordSize | |
1023 comp_words_on_stack = round_to(comp_words_on_stack, 2); | |
1024 // Now compute the distance from Lesp to SP. This calculation does not | |
1025 // include the space for total_args_passed because Lesp has not yet popped | |
1026 // the arguments. | |
1027 __ sub(SP, (comp_words_on_stack)*wordSize, SP); | |
1028 } | |
1029 | |
1030 // Will jump to the compiled code just as if compiled code was doing it. | |
1031 // Pre-load the register-jump target early, to schedule it better. | |
1032 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3); | |
1033 | |
1034 // Now generate the shuffle code. Pick up all register args and move the | |
1035 // rest through G1_scratch. | |
1036 for (int i=0; i<total_args_passed; i++) { | |
1037 if (sig_bt[i] == T_VOID) { | |
1038 // Longs and doubles are passed in native word order, but misaligned | |
1039 // in the 32-bit build. | |
1040 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); | |
1041 continue; | |
1042 } | |
1043 | |
1044 // Pick up 0, 1 or 2 words from Lesp+offset. Assume mis-aligned in the | |
1045 // 32-bit build and aligned in the 64-bit build. Look for the obvious | |
1046 // ldx/lddf optimizations. | |
1047 | |
1048 // Load in argument order going down. | |
1049 const int ld_off = (total_args_passed-i)*Interpreter::stackElementSize(); | |
1050 #ifdef _LP64 | |
1051 set_Rdisp(G1_scratch); | |
1052 #endif // _LP64 | |
1053 | |
1054 VMReg r_1 = regs[i].first(); | |
1055 VMReg r_2 = regs[i].second(); | |
1056 if (!r_1->is_valid()) { | |
1057 assert(!r_2->is_valid(), ""); | |
1058 continue; | |
1059 } | |
1060 if (r_1->is_stack()) { // Pretend stack targets are loaded into F8/F9 | |
1061 r_1 = F8->as_VMReg(); // as part of the load/store shuffle | |
1062 if (r_2->is_valid()) r_2 = r_1->next(); | |
1063 } | |
1064 if (r_1->is_Register()) { // Register argument | |
1065 Register r = r_1->as_Register()->after_restore(); | |
1066 if (!r_2->is_valid()) { | |
1067 __ ld(Gargs, arg_slot(ld_off), r); | |
1068 } else { | |
1069 #ifdef _LP64 | |
1070 // In V9, longs are given 2 64-bit slots in the interpreter, but the | |
1071 // data is passed in only 1 slot. | |
1072 Register slot = (sig_bt[i]==T_LONG) ? | |
1073 next_arg_slot(ld_off) : arg_slot(ld_off); | |
1074 __ ldx(Gargs, slot, r); | |
1075 #else | |
1076 // Need to load a 64-bit value into G1/G4, but G1/G4 is being used in the | |
1077 // stack shuffle. Load the first 2 longs into G1/G4 later. | |
1078 #endif | |
1079 } | |
1080 } else { | |
1081 assert(r_1->is_FloatRegister(), ""); | |
1082 if (!r_2->is_valid()) { | |
1083 __ ldf(FloatRegisterImpl::S, Gargs, arg_slot(ld_off), r_1->as_FloatRegister()); | |
1084 } else { | |
1085 #ifdef _LP64 | |
1086 // In V9, doubles are given 2 64-bit slots in the interpreter, but the | |
1087 // data is passed in only 1 slot. This code also handles longs that | |
1088 // are passed on the stack, but need a stack-to-stack move through a | |
1089 // spare float register. | |
1090 Register slot = (sig_bt[i]==T_LONG || sig_bt[i] == T_DOUBLE) ? | |
1091 next_arg_slot(ld_off) : arg_slot(ld_off); | |
1092 __ ldf(FloatRegisterImpl::D, Gargs, slot, r_1->as_FloatRegister()); | |
1093 #else | |
1094 // Need to marshal 64-bit value from misaligned Lesp loads | |
1095 __ ldf(FloatRegisterImpl::S, Gargs, next_arg_slot(ld_off), r_1->as_FloatRegister()); | |
1096 __ ldf(FloatRegisterImpl::S, Gargs, arg_slot(ld_off), r_2->as_FloatRegister()); | |
1097 #endif | |
1098 } | |
1099 } | |
1100 // Was the argument really intended to be on the stack, but was loaded | |
1101 // into F8/F9? | |
1102 if (regs[i].first()->is_stack()) { | |
1103 assert(r_1->as_FloatRegister() == F8, "fix this code"); | |
1104 // Convert stack slot to an SP offset | |
1105 int st_off = reg2offset(regs[i].first()) + STACK_BIAS; | |
1106 // Store down the shuffled stack word. Target address _is_ aligned. | |
1107 if (!r_2->is_valid()) __ stf(FloatRegisterImpl::S, r_1->as_FloatRegister(), SP, st_off); | |
1108 else __ stf(FloatRegisterImpl::D, r_1->as_FloatRegister(), SP, st_off); | |
1109 } | |
1110 } | |
1111 bool made_space = false; | |
1112 #ifndef _LP64 | |
1113 // May need to pick up a few long args in G1/G4 | |
1114 bool g4_crushed = false; | |
1115 bool g3_crushed = false; | |
1116 for (int i=0; i<total_args_passed; i++) { | |
1117 if (regs[i].first()->is_Register() && regs[i].second()->is_valid()) { | |
1118 // Load in argument order going down | |
1119 int ld_off = (total_args_passed-i)*Interpreter::stackElementSize(); | |
1120 // Need to marshal 64-bit value from misaligned Lesp loads | |
1121 Register r = regs[i].first()->as_Register()->after_restore(); | |
1122 if (r == G1 || r == G4) { | |
1123 assert(!g4_crushed, "ordering problem"); | |
1124 if (r == G4){ | |
1125 g4_crushed = true; | |
1126 __ lduw(Gargs, arg_slot(ld_off) , G3_scratch); // Load lo bits | |
1127 __ ld (Gargs, next_arg_slot(ld_off), r); // Load hi bits | |
1128 } else { | |
1129 // better schedule this way | |
1130 __ ld (Gargs, next_arg_slot(ld_off), r); // Load hi bits | |
1131 __ lduw(Gargs, arg_slot(ld_off) , G3_scratch); // Load lo bits | |
1132 } | |
1133 g3_crushed = true; | |
1134 __ sllx(r, 32, r); | |
1135 __ or3(G3_scratch, r, r); | |
1136 } else { | |
1137 assert(r->is_out(), "longs passed in two O registers"); | |
1138 __ ld (Gargs, arg_slot(ld_off) , r->successor()); // Load lo bits | |
1139 __ ld (Gargs, next_arg_slot(ld_off), r); // Load hi bits | |
1140 } | |
1141 } | |
1142 } | |
1143 #endif | |
1144 | |
1145 // Jump to the compiled code just as if compiled code was doing it. | |
1146 // | |
1147 #ifndef _LP64 | |
1148 if (g3_crushed) { | |
1149 // Rats load was wasted, at least it is in cache... | |
1150 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3); | |
1151 } | |
1152 #endif /* _LP64 */ | |
1153 | |
1154 // 6243940 We might end up in handle_wrong_method if | |
1155 // the callee is deoptimized as we race thru here. If that | |
1156 // happens we don't want to take a safepoint because the | |
1157 // caller frame will look interpreted and arguments are now | |
1158 // "compiled" so it is much better to make this transition | |
1159 // invisible to the stack walking code. Unfortunately if | |
1160 // we try and find the callee by normal means a safepoint | |
1161 // is possible. So we stash the desired callee in the thread | |
1162 // and the vm will find there should this case occur. | |
1163 Address callee_target_addr(G2_thread, 0, in_bytes(JavaThread::callee_target_offset())); | |
1164 __ st_ptr(G5_method, callee_target_addr); | |
1165 | |
1166 if (StressNonEntrant) { | |
1167 // Open a big window for deopt failure | |
1168 __ save_frame(0); | |
1169 __ mov(G0, L0); | |
1170 Label loop; | |
1171 __ bind(loop); | |
1172 __ sub(L0, 1, L0); | |
1173 __ br_null(L0, false, Assembler::pt, loop); | |
1174 __ delayed()->nop(); | |
1175 | |
1176 __ restore(); | |
1177 } | |
1178 | |
1179 | |
1180 __ jmpl(G3, 0, G0); | |
1181 __ delayed()->nop(); | |
1182 } | |
1183 | |
1184 // --------------------------------------------------------------- | |
1185 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, | |
1186 int total_args_passed, | |
1187 // VMReg max_arg, | |
1188 int comp_args_on_stack, // VMRegStackSlots | |
1189 const BasicType *sig_bt, | |
1190 const VMRegPair *regs) { | |
1191 address i2c_entry = __ pc(); | |
1192 | |
1193 AdapterGenerator agen(masm); | |
1194 | |
1195 agen.gen_i2c_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs); | |
1196 | |
1197 | |
1198 // ------------------------------------------------------------------------- | |
1199 // Generate a C2I adapter. On entry we know G5 holds the methodOop. The | |
1200 // args start out packed in the compiled layout. They need to be unpacked | |
1201 // into the interpreter layout. This will almost always require some stack | |
1202 // space. We grow the current (compiled) stack, then repack the args. We | |
1203 // finally end in a jump to the generic interpreter entry point. On exit | |
1204 // from the interpreter, the interpreter will restore our SP (lest the | |
1205 // compiled code, which relys solely on SP and not FP, get sick). | |
1206 | |
1207 address c2i_unverified_entry = __ pc(); | |
1208 Label skip_fixup; | |
1209 { | |
1210 #if !defined(_LP64) && defined(COMPILER2) | |
1211 Register R_temp = L0; // another scratch register | |
1212 #else | |
1213 Register R_temp = G1; // another scratch register | |
1214 #endif | |
1215 | |
1216 Address ic_miss(G3_scratch, SharedRuntime::get_ic_miss_stub()); | |
1217 | |
1218 __ verify_oop(O0); | |
1219 __ verify_oop(G5_method); | |
1220 __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), G3_scratch); | |
1221 __ verify_oop(G3_scratch); | |
1222 | |
1223 #if !defined(_LP64) && defined(COMPILER2) | |
1224 __ save(SP, -frame::register_save_words*wordSize, SP); | |
1225 __ ld_ptr(G5_method, compiledICHolderOopDesc::holder_klass_offset(), R_temp); | |
1226 __ verify_oop(R_temp); | |
1227 __ cmp(G3_scratch, R_temp); | |
1228 __ restore(); | |
1229 #else | |
1230 __ ld_ptr(G5_method, compiledICHolderOopDesc::holder_klass_offset(), R_temp); | |
1231 __ verify_oop(R_temp); | |
1232 __ cmp(G3_scratch, R_temp); | |
1233 #endif | |
1234 | |
1235 Label ok, ok2; | |
1236 __ brx(Assembler::equal, false, Assembler::pt, ok); | |
1237 __ delayed()->ld_ptr(G5_method, compiledICHolderOopDesc::holder_method_offset(), G5_method); | |
1238 __ jump_to(ic_miss); | |
1239 __ delayed()->nop(); | |
1240 | |
1241 __ bind(ok); | |
1242 // Method might have been compiled since the call site was patched to | |
1243 // interpreted if that is the case treat it as a miss so we can get | |
1244 // the call site corrected. | |
1245 __ ld_ptr(G5_method, in_bytes(methodOopDesc::code_offset()), G3_scratch); | |
1246 __ bind(ok2); | |
1247 __ br_null(G3_scratch, false, __ pt, skip_fixup); | |
1248 __ delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), G3_scratch); | |
1249 __ jump_to(ic_miss); | |
1250 __ delayed()->nop(); | |
1251 | |
1252 } | |
1253 | |
1254 address c2i_entry = __ pc(); | |
1255 | |
1256 agen.gen_c2i_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); | |
1257 | |
1258 __ flush(); | |
1259 return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry); | |
1260 | |
1261 } | |
1262 | |
1263 // Helper function for native calling conventions | |
1264 static VMReg int_stk_helper( int i ) { | |
1265 // Bias any stack based VMReg we get by ignoring the window area | |
1266 // but not the register parameter save area. | |
1267 // | |
1268 // This is strange for the following reasons. We'd normally expect | |
1269 // the calling convention to return an VMReg for a stack slot | |
1270 // completely ignoring any abi reserved area. C2 thinks of that | |
1271 // abi area as only out_preserve_stack_slots. This does not include | |
1272 // the area allocated by the C abi to store down integer arguments | |
1273 // because the java calling convention does not use it. So | |
1274 // since c2 assumes that there are only out_preserve_stack_slots | |
1275 // to bias the optoregs (which impacts VMRegs) when actually referencing any actual stack | |
1276 // location the c calling convention must add in this bias amount | |
1277 // to make up for the fact that the out_preserve_stack_slots is | |
1278 // insufficient for C calls. What a mess. I sure hope those 6 | |
1279 // stack words were worth it on every java call! | |
1280 | |
1281 // Another way of cleaning this up would be for out_preserve_stack_slots | |
1282 // to take a parameter to say whether it was C or java calling conventions. | |
1283 // Then things might look a little better (but not much). | |
1284 | |
1285 int mem_parm_offset = i - SPARC_ARGS_IN_REGS_NUM; | |
1286 if( mem_parm_offset < 0 ) { | |
1287 return as_oRegister(i)->as_VMReg(); | |
1288 } else { | |
1289 int actual_offset = (mem_parm_offset + frame::memory_parameter_word_sp_offset) * VMRegImpl::slots_per_word; | |
1290 // Now return a biased offset that will be correct when out_preserve_slots is added back in | |
1291 return VMRegImpl::stack2reg(actual_offset - SharedRuntime::out_preserve_stack_slots()); | |
1292 } | |
1293 } | |
1294 | |
1295 | |
1296 int SharedRuntime::c_calling_convention(const BasicType *sig_bt, | |
1297 VMRegPair *regs, | |
1298 int total_args_passed) { | |
1299 | |
1300 // Return the number of VMReg stack_slots needed for the args. | |
1301 // This value does not include an abi space (like register window | |
1302 // save area). | |
1303 | |
1304 // The native convention is V8 if !LP64 | |
1305 // The LP64 convention is the V9 convention which is slightly more sane. | |
1306 | |
1307 // We return the amount of VMReg stack slots we need to reserve for all | |
1308 // the arguments NOT counting out_preserve_stack_slots. Since we always | |
1309 // have space for storing at least 6 registers to memory we start with that. | |
1310 // See int_stk_helper for a further discussion. | |
1311 int max_stack_slots = (frame::varargs_offset * VMRegImpl::slots_per_word) - SharedRuntime::out_preserve_stack_slots(); | |
1312 | |
1313 #ifdef _LP64 | |
1314 // V9 convention: All things "as-if" on double-wide stack slots. | |
1315 // Hoist any int/ptr/long's in the first 6 to int regs. | |
1316 // Hoist any flt/dbl's in the first 16 dbl regs. | |
1317 int j = 0; // Count of actual args, not HALVES | |
1318 for( int i=0; i<total_args_passed; i++, j++ ) { | |
1319 switch( sig_bt[i] ) { | |
1320 case T_BOOLEAN: | |
1321 case T_BYTE: | |
1322 case T_CHAR: | |
1323 case T_INT: | |
1324 case T_SHORT: | |
1325 regs[i].set1( int_stk_helper( j ) ); break; | |
1326 case T_LONG: | |
1327 assert( sig_bt[i+1] == T_VOID, "expecting half" ); | |
1328 case T_ADDRESS: // raw pointers, like current thread, for VM calls | |
1329 case T_ARRAY: | |
1330 case T_OBJECT: | |
1331 regs[i].set2( int_stk_helper( j ) ); | |
1332 break; | |
1333 case T_FLOAT: | |
1334 if ( j < 16 ) { | |
1335 // V9ism: floats go in ODD registers | |
1336 regs[i].set1(as_FloatRegister(1 + (j<<1))->as_VMReg()); | |
1337 } else { | |
1338 // V9ism: floats go in ODD stack slot | |
1339 regs[i].set1(VMRegImpl::stack2reg(1 + (j<<1))); | |
1340 } | |
1341 break; | |
1342 case T_DOUBLE: | |
1343 assert( sig_bt[i+1] == T_VOID, "expecting half" ); | |
1344 if ( j < 16 ) { | |
1345 // V9ism: doubles go in EVEN/ODD regs | |
1346 regs[i].set2(as_FloatRegister(j<<1)->as_VMReg()); | |
1347 } else { | |
1348 // V9ism: doubles go in EVEN/ODD stack slots | |
1349 regs[i].set2(VMRegImpl::stack2reg(j<<1)); | |
1350 } | |
1351 break; | |
1352 case T_VOID: regs[i].set_bad(); j--; break; // Do not count HALVES | |
1353 default: | |
1354 ShouldNotReachHere(); | |
1355 } | |
1356 if (regs[i].first()->is_stack()) { | |
1357 int off = regs[i].first()->reg2stack(); | |
1358 if (off > max_stack_slots) max_stack_slots = off; | |
1359 } | |
1360 if (regs[i].second()->is_stack()) { | |
1361 int off = regs[i].second()->reg2stack(); | |
1362 if (off > max_stack_slots) max_stack_slots = off; | |
1363 } | |
1364 } | |
1365 | |
1366 #else // _LP64 | |
1367 // V8 convention: first 6 things in O-regs, rest on stack. | |
1368 // Alignment is willy-nilly. | |
1369 for( int i=0; i<total_args_passed; i++ ) { | |
1370 switch( sig_bt[i] ) { | |
1371 case T_ADDRESS: // raw pointers, like current thread, for VM calls | |
1372 case T_ARRAY: | |
1373 case T_BOOLEAN: | |
1374 case T_BYTE: | |
1375 case T_CHAR: | |
1376 case T_FLOAT: | |
1377 case T_INT: | |
1378 case T_OBJECT: | |
1379 case T_SHORT: | |
1380 regs[i].set1( int_stk_helper( i ) ); | |
1381 break; | |
1382 case T_DOUBLE: | |
1383 case T_LONG: | |
1384 assert( sig_bt[i+1] == T_VOID, "expecting half" ); | |
1385 regs[i].set_pair( int_stk_helper( i+1 ), int_stk_helper( i ) ); | |
1386 break; | |
1387 case T_VOID: regs[i].set_bad(); break; | |
1388 default: | |
1389 ShouldNotReachHere(); | |
1390 } | |
1391 if (regs[i].first()->is_stack()) { | |
1392 int off = regs[i].first()->reg2stack(); | |
1393 if (off > max_stack_slots) max_stack_slots = off; | |
1394 } | |
1395 if (regs[i].second()->is_stack()) { | |
1396 int off = regs[i].second()->reg2stack(); | |
1397 if (off > max_stack_slots) max_stack_slots = off; | |
1398 } | |
1399 } | |
1400 #endif // _LP64 | |
1401 | |
1402 return round_to(max_stack_slots + 1, 2); | |
1403 | |
1404 } | |
1405 | |
1406 | |
1407 // --------------------------------------------------------------------------- | |
1408 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { | |
1409 switch (ret_type) { | |
1410 case T_FLOAT: | |
1411 __ stf(FloatRegisterImpl::S, F0, SP, frame_slots*VMRegImpl::stack_slot_size - 4+STACK_BIAS); | |
1412 break; | |
1413 case T_DOUBLE: | |
1414 __ stf(FloatRegisterImpl::D, F0, SP, frame_slots*VMRegImpl::stack_slot_size - 8+STACK_BIAS); | |
1415 break; | |
1416 } | |
1417 } | |
1418 | |
1419 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { | |
1420 switch (ret_type) { | |
1421 case T_FLOAT: | |
1422 __ ldf(FloatRegisterImpl::S, SP, frame_slots*VMRegImpl::stack_slot_size - 4+STACK_BIAS, F0); | |
1423 break; | |
1424 case T_DOUBLE: | |
1425 __ ldf(FloatRegisterImpl::D, SP, frame_slots*VMRegImpl::stack_slot_size - 8+STACK_BIAS, F0); | |
1426 break; | |
1427 } | |
1428 } | |
1429 | |
1430 // Check and forward and pending exception. Thread is stored in | |
1431 // L7_thread_cache and possibly NOT in G2_thread. Since this is a native call, there | |
1432 // is no exception handler. We merely pop this frame off and throw the | |
1433 // exception in the caller's frame. | |
1434 static void check_forward_pending_exception(MacroAssembler *masm, Register Rex_oop) { | |
1435 Label L; | |
1436 __ br_null(Rex_oop, false, Assembler::pt, L); | |
1437 __ delayed()->mov(L7_thread_cache, G2_thread); // restore in case we have exception | |
1438 // Since this is a native call, we *know* the proper exception handler | |
1439 // without calling into the VM: it's the empty function. Just pop this | |
1440 // frame and then jump to forward_exception_entry; O7 will contain the | |
1441 // native caller's return PC. | |
1442 Address exception_entry(G3_scratch, StubRoutines::forward_exception_entry()); | |
1443 __ jump_to(exception_entry); | |
1444 __ delayed()->restore(); // Pop this frame off. | |
1445 __ bind(L); | |
1446 } | |
1447 | |
1448 // A simple move of integer like type | |
1449 static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1450 if (src.first()->is_stack()) { | |
1451 if (dst.first()->is_stack()) { | |
1452 // stack to stack | |
1453 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); | |
1454 __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1455 } else { | |
1456 // stack to reg | |
1457 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1458 } | |
1459 } else if (dst.first()->is_stack()) { | |
1460 // reg to stack | |
1461 __ st(src.first()->as_Register(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1462 } else { | |
1463 __ mov(src.first()->as_Register(), dst.first()->as_Register()); | |
1464 } | |
1465 } | |
1466 | |
1467 // On 64 bit we will store integer like items to the stack as | |
1468 // 64 bits items (sparc abi) even though java would only store | |
1469 // 32bits for a parameter. On 32bit it will simply be 32 bits | |
1470 // So this routine will do 32->32 on 32bit and 32->64 on 64bit | |
1471 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1472 if (src.first()->is_stack()) { | |
1473 if (dst.first()->is_stack()) { | |
1474 // stack to stack | |
1475 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); | |
1476 __ st_ptr(L5, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1477 } else { | |
1478 // stack to reg | |
1479 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1480 } | |
1481 } else if (dst.first()->is_stack()) { | |
1482 // reg to stack | |
1483 __ st_ptr(src.first()->as_Register(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1484 } else { | |
1485 __ mov(src.first()->as_Register(), dst.first()->as_Register()); | |
1486 } | |
1487 } | |
1488 | |
1489 | |
1490 // An oop arg. Must pass a handle not the oop itself | |
1491 static void object_move(MacroAssembler* masm, | |
1492 OopMap* map, | |
1493 int oop_handle_offset, | |
1494 int framesize_in_slots, | |
1495 VMRegPair src, | |
1496 VMRegPair dst, | |
1497 bool is_receiver, | |
1498 int* receiver_offset) { | |
1499 | |
1500 // must pass a handle. First figure out the location we use as a handle | |
1501 | |
1502 if (src.first()->is_stack()) { | |
1503 // Oop is already on the stack | |
1504 Register rHandle = dst.first()->is_stack() ? L5 : dst.first()->as_Register(); | |
1505 __ add(FP, reg2offset(src.first()) + STACK_BIAS, rHandle); | |
1506 __ ld_ptr(rHandle, 0, L4); | |
1507 #ifdef _LP64 | |
1508 __ movr( Assembler::rc_z, L4, G0, rHandle ); | |
1509 #else | |
1510 __ tst( L4 ); | |
1511 __ movcc( Assembler::zero, false, Assembler::icc, G0, rHandle ); | |
1512 #endif | |
1513 if (dst.first()->is_stack()) { | |
1514 __ st_ptr(rHandle, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1515 } | |
1516 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots(); | |
1517 if (is_receiver) { | |
1518 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size; | |
1519 } | |
1520 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); | |
1521 } else { | |
1522 // Oop is in an input register pass we must flush it to the stack | |
1523 const Register rOop = src.first()->as_Register(); | |
1524 const Register rHandle = L5; | |
1525 int oop_slot = rOop->input_number() * VMRegImpl::slots_per_word + oop_handle_offset; | |
1526 int offset = oop_slot*VMRegImpl::stack_slot_size; | |
1527 Label skip; | |
1528 __ st_ptr(rOop, SP, offset + STACK_BIAS); | |
1529 if (is_receiver) { | |
1530 *receiver_offset = oop_slot * VMRegImpl::stack_slot_size; | |
1531 } | |
1532 map->set_oop(VMRegImpl::stack2reg(oop_slot)); | |
1533 __ add(SP, offset + STACK_BIAS, rHandle); | |
1534 #ifdef _LP64 | |
1535 __ movr( Assembler::rc_z, rOop, G0, rHandle ); | |
1536 #else | |
1537 __ tst( rOop ); | |
1538 __ movcc( Assembler::zero, false, Assembler::icc, G0, rHandle ); | |
1539 #endif | |
1540 | |
1541 if (dst.first()->is_stack()) { | |
1542 __ st_ptr(rHandle, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1543 } else { | |
1544 __ mov(rHandle, dst.first()->as_Register()); | |
1545 } | |
1546 } | |
1547 } | |
1548 | |
1549 // A float arg may have to do float reg int reg conversion | |
1550 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1551 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); | |
1552 | |
1553 if (src.first()->is_stack()) { | |
1554 if (dst.first()->is_stack()) { | |
1555 // stack to stack the easiest of the bunch | |
1556 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); | |
1557 __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1558 } else { | |
1559 // stack to reg | |
1560 if (dst.first()->is_Register()) { | |
1561 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1562 } else { | |
1563 __ ldf(FloatRegisterImpl::S, FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_FloatRegister()); | |
1564 } | |
1565 } | |
1566 } else if (dst.first()->is_stack()) { | |
1567 // reg to stack | |
1568 if (src.first()->is_Register()) { | |
1569 __ st(src.first()->as_Register(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1570 } else { | |
1571 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1572 } | |
1573 } else { | |
1574 // reg to reg | |
1575 if (src.first()->is_Register()) { | |
1576 if (dst.first()->is_Register()) { | |
1577 // gpr -> gpr | |
1578 __ mov(src.first()->as_Register(), dst.first()->as_Register()); | |
1579 } else { | |
1580 // gpr -> fpr | |
1581 __ st(src.first()->as_Register(), FP, -4 + STACK_BIAS); | |
1582 __ ldf(FloatRegisterImpl::S, FP, -4 + STACK_BIAS, dst.first()->as_FloatRegister()); | |
1583 } | |
1584 } else if (dst.first()->is_Register()) { | |
1585 // fpr -> gpr | |
1586 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), FP, -4 + STACK_BIAS); | |
1587 __ ld(FP, -4 + STACK_BIAS, dst.first()->as_Register()); | |
1588 } else { | |
1589 // fpr -> fpr | |
1590 // In theory these overlap but the ordering is such that this is likely a nop | |
1591 if ( src.first() != dst.first()) { | |
1592 __ fmov(FloatRegisterImpl::S, src.first()->as_FloatRegister(), dst.first()->as_FloatRegister()); | |
1593 } | |
1594 } | |
1595 } | |
1596 } | |
1597 | |
1598 static void split_long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1599 VMRegPair src_lo(src.first()); | |
1600 VMRegPair src_hi(src.second()); | |
1601 VMRegPair dst_lo(dst.first()); | |
1602 VMRegPair dst_hi(dst.second()); | |
1603 simple_move32(masm, src_lo, dst_lo); | |
1604 simple_move32(masm, src_hi, dst_hi); | |
1605 } | |
1606 | |
1607 // A long move | |
1608 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1609 | |
1610 // Do the simple ones here else do two int moves | |
1611 if (src.is_single_phys_reg() ) { | |
1612 if (dst.is_single_phys_reg()) { | |
1613 __ mov(src.first()->as_Register(), dst.first()->as_Register()); | |
1614 } else { | |
1615 // split src into two separate registers | |
1616 // Remember hi means hi address or lsw on sparc | |
1617 // Move msw to lsw | |
1618 if (dst.second()->is_reg()) { | |
1619 // MSW -> MSW | |
1620 __ srax(src.first()->as_Register(), 32, dst.first()->as_Register()); | |
1621 // Now LSW -> LSW | |
1622 // this will only move lo -> lo and ignore hi | |
1623 VMRegPair split(dst.second()); | |
1624 simple_move32(masm, src, split); | |
1625 } else { | |
1626 VMRegPair split(src.first(), L4->as_VMReg()); | |
1627 // MSW -> MSW (lo ie. first word) | |
1628 __ srax(src.first()->as_Register(), 32, L4); | |
1629 split_long_move(masm, split, dst); | |
1630 } | |
1631 } | |
1632 } else if (dst.is_single_phys_reg()) { | |
1633 if (src.is_adjacent_aligned_on_stack(2)) { | |
1634 __ ldd(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1635 } else { | |
1636 // dst is a single reg. | |
1637 // Remember lo is low address not msb for stack slots | |
1638 // and lo is the "real" register for registers | |
1639 // src is | |
1640 | |
1641 VMRegPair split; | |
1642 | |
1643 if (src.first()->is_reg()) { | |
1644 // src.lo (msw) is a reg, src.hi is stk/reg | |
1645 // we will move: src.hi (LSW) -> dst.lo, src.lo (MSW) -> src.lo [the MSW is in the LSW of the reg] | |
1646 split.set_pair(dst.first(), src.first()); | |
1647 } else { | |
1648 // msw is stack move to L5 | |
1649 // lsw is stack move to dst.lo (real reg) | |
1650 // we will move: src.hi (LSW) -> dst.lo, src.lo (MSW) -> L5 | |
1651 split.set_pair(dst.first(), L5->as_VMReg()); | |
1652 } | |
1653 | |
1654 // src.lo -> src.lo/L5, src.hi -> dst.lo (the real reg) | |
1655 // msw -> src.lo/L5, lsw -> dst.lo | |
1656 split_long_move(masm, src, split); | |
1657 | |
1658 // So dst now has the low order correct position the | |
1659 // msw half | |
1660 __ sllx(split.first()->as_Register(), 32, L5); | |
1661 | |
1662 const Register d = dst.first()->as_Register(); | |
1663 __ or3(L5, d, d); | |
1664 } | |
1665 } else { | |
1666 // For LP64 we can probably do better. | |
1667 split_long_move(masm, src, dst); | |
1668 } | |
1669 } | |
1670 | |
1671 // A double move | |
1672 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { | |
1673 | |
1674 // The painful thing here is that like long_move a VMRegPair might be | |
1675 // 1: a single physical register | |
1676 // 2: two physical registers (v8) | |
1677 // 3: a physical reg [lo] and a stack slot [hi] (v8) | |
1678 // 4: two stack slots | |
1679 | |
1680 // Since src is always a java calling convention we know that the src pair | |
1681 // is always either all registers or all stack (and aligned?) | |
1682 | |
1683 // in a register [lo] and a stack slot [hi] | |
1684 if (src.first()->is_stack()) { | |
1685 if (dst.first()->is_stack()) { | |
1686 // stack to stack the easiest of the bunch | |
1687 // ought to be a way to do this where if alignment is ok we use ldd/std when possible | |
1688 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); | |
1689 __ ld(FP, reg2offset(src.second()) + STACK_BIAS, L4); | |
1690 __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS); | |
1691 __ st(L4, SP, reg2offset(dst.second()) + STACK_BIAS); | |
1692 } else { | |
1693 // stack to reg | |
1694 if (dst.second()->is_stack()) { | |
1695 // stack -> reg, stack -> stack | |
1696 __ ld(FP, reg2offset(src.second()) + STACK_BIAS, L4); | |
1697 if (dst.first()->is_Register()) { | |
1698 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1699 } else { | |
1700 __ ldf(FloatRegisterImpl::S, FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_FloatRegister()); | |
1701 } | |
1702 // This was missing. (very rare case) | |
1703 __ st(L4, SP, reg2offset(dst.second()) + STACK_BIAS); | |
1704 } else { | |
1705 // stack -> reg | |
1706 // Eventually optimize for alignment QQQ | |
1707 if (dst.first()->is_Register()) { | |
1708 __ ld(FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_Register()); | |
1709 __ ld(FP, reg2offset(src.second()) + STACK_BIAS, dst.second()->as_Register()); | |
1710 } else { | |
1711 __ ldf(FloatRegisterImpl::S, FP, reg2offset(src.first()) + STACK_BIAS, dst.first()->as_FloatRegister()); | |
1712 __ ldf(FloatRegisterImpl::S, FP, reg2offset(src.second()) + STACK_BIAS, dst.second()->as_FloatRegister()); | |
1713 } | |
1714 } | |
1715 } | |
1716 } else if (dst.first()->is_stack()) { | |
1717 // reg to stack | |
1718 if (src.first()->is_Register()) { | |
1719 // Eventually optimize for alignment QQQ | |
1720 __ st(src.first()->as_Register(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1721 if (src.second()->is_stack()) { | |
1722 __ ld(FP, reg2offset(src.second()) + STACK_BIAS, L4); | |
1723 __ st(L4, SP, reg2offset(dst.second()) + STACK_BIAS); | |
1724 } else { | |
1725 __ st(src.second()->as_Register(), SP, reg2offset(dst.second()) + STACK_BIAS); | |
1726 } | |
1727 } else { | |
1728 // fpr to stack | |
1729 if (src.second()->is_stack()) { | |
1730 ShouldNotReachHere(); | |
1731 } else { | |
1732 // Is the stack aligned? | |
1733 if (reg2offset(dst.first()) & 0x7) { | |
1734 // No do as pairs | |
1735 __ stf(FloatRegisterImpl::S, src.first()->as_FloatRegister(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1736 __ stf(FloatRegisterImpl::S, src.second()->as_FloatRegister(), SP, reg2offset(dst.second()) + STACK_BIAS); | |
1737 } else { | |
1738 __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), SP, reg2offset(dst.first()) + STACK_BIAS); | |
1739 } | |
1740 } | |
1741 } | |
1742 } else { | |
1743 // reg to reg | |
1744 if (src.first()->is_Register()) { | |
1745 if (dst.first()->is_Register()) { | |
1746 // gpr -> gpr | |
1747 __ mov(src.first()->as_Register(), dst.first()->as_Register()); | |
1748 __ mov(src.second()->as_Register(), dst.second()->as_Register()); | |
1749 } else { | |
1750 // gpr -> fpr | |
1751 // ought to be able to do a single store | |
1752 __ stx(src.first()->as_Register(), FP, -8 + STACK_BIAS); | |
1753 __ stx(src.second()->as_Register(), FP, -4 + STACK_BIAS); | |
1754 // ought to be able to do a single load | |
1755 __ ldf(FloatRegisterImpl::S, FP, -8 + STACK_BIAS, dst.first()->as_FloatRegister()); | |
1756 __ ldf(FloatRegisterImpl::S, FP, -4 + STACK_BIAS, dst.second()->as_FloatRegister()); | |
1757 } | |
1758 } else if (dst.first()->is_Register()) { | |
1759 // fpr -> gpr | |
1760 // ought to be able to do a single store | |
1761 __ stf(FloatRegisterImpl::D, src.first()->as_FloatRegister(), FP, -8 + STACK_BIAS); | |
1762 // ought to be able to do a single load | |
1763 // REMEMBER first() is low address not LSB | |
1764 __ ld(FP, -8 + STACK_BIAS, dst.first()->as_Register()); | |
1765 if (dst.second()->is_Register()) { | |
1766 __ ld(FP, -4 + STACK_BIAS, dst.second()->as_Register()); | |
1767 } else { | |
1768 __ ld(FP, -4 + STACK_BIAS, L4); | |
1769 __ st(L4, SP, reg2offset(dst.second()) + STACK_BIAS); | |
1770 } | |
1771 } else { | |
1772 // fpr -> fpr | |
1773 // In theory these overlap but the ordering is such that this is likely a nop | |
1774 if ( src.first() != dst.first()) { | |
1775 __ fmov(FloatRegisterImpl::D, src.first()->as_FloatRegister(), dst.first()->as_FloatRegister()); | |
1776 } | |
1777 } | |
1778 } | |
1779 } | |
1780 | |
1781 // Creates an inner frame if one hasn't already been created, and | |
1782 // saves a copy of the thread in L7_thread_cache | |
1783 static void create_inner_frame(MacroAssembler* masm, bool* already_created) { | |
1784 if (!*already_created) { | |
1785 __ save_frame(0); | |
1786 // Save thread in L7 (INNER FRAME); it crosses a bunch of VM calls below | |
1787 // Don't use save_thread because it smashes G2 and we merely want to save a | |
1788 // copy | |
1789 __ mov(G2_thread, L7_thread_cache); | |
1790 *already_created = true; | |
1791 } | |
1792 } | |
1793 | |
1794 // --------------------------------------------------------------------------- | |
1795 // Generate a native wrapper for a given method. The method takes arguments | |
1796 // in the Java compiled code convention, marshals them to the native | |
1797 // convention (handlizes oops, etc), transitions to native, makes the call, | |
1798 // returns to java state (possibly blocking), unhandlizes any result and | |
1799 // returns. | |
1800 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler* masm, | |
1801 methodHandle method, | |
1802 int total_in_args, | |
1803 int comp_args_on_stack, // in VMRegStackSlots | |
1804 BasicType *in_sig_bt, | |
1805 VMRegPair *in_regs, | |
1806 BasicType ret_type) { | |
1807 | |
1808 | |
1809 // Native nmethod wrappers never take possesion of the oop arguments. | |
1810 // So the caller will gc the arguments. The only thing we need an | |
1811 // oopMap for is if the call is static | |
1812 // | |
1813 // An OopMap for lock (and class if static), and one for the VM call itself | |
1814 OopMapSet *oop_maps = new OopMapSet(); | |
1815 intptr_t start = (intptr_t)__ pc(); | |
1816 | |
1817 // First thing make an ic check to see if we should even be here | |
1818 { | |
1819 Label L; | |
1820 const Register temp_reg = G3_scratch; | |
1821 Address ic_miss(temp_reg, SharedRuntime::get_ic_miss_stub()); | |
1822 __ verify_oop(O0); | |
1823 __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), temp_reg); | |
1824 __ cmp(temp_reg, G5_inline_cache_reg); | |
1825 __ brx(Assembler::equal, true, Assembler::pt, L); | |
1826 __ delayed()->nop(); | |
1827 | |
1828 __ jump_to(ic_miss, 0); | |
1829 __ delayed()->nop(); | |
1830 __ align(CodeEntryAlignment); | |
1831 __ bind(L); | |
1832 } | |
1833 | |
1834 int vep_offset = ((intptr_t)__ pc()) - start; | |
1835 | |
1836 #ifdef COMPILER1 | |
1837 if (InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) { | |
1838 // Object.hashCode can pull the hashCode from the header word | |
1839 // instead of doing a full VM transition once it's been computed. | |
1840 // Since hashCode is usually polymorphic at call sites we can't do | |
1841 // this optimization at the call site without a lot of work. | |
1842 Label slowCase; | |
1843 Register receiver = O0; | |
1844 Register result = O0; | |
1845 Register header = G3_scratch; | |
1846 Register hash = G3_scratch; // overwrite header value with hash value | |
1847 Register mask = G1; // to get hash field from header | |
1848 | |
1849 // Read the header and build a mask to get its hash field. Give up if the object is not unlocked. | |
1850 // We depend on hash_mask being at most 32 bits and avoid the use of | |
1851 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit | |
1852 // vm: see markOop.hpp. | |
1853 __ ld_ptr(receiver, oopDesc::mark_offset_in_bytes(), header); | |
1854 __ sethi(markOopDesc::hash_mask, mask); | |
1855 __ btst(markOopDesc::unlocked_value, header); | |
1856 __ br(Assembler::zero, false, Assembler::pn, slowCase); | |
1857 if (UseBiasedLocking) { | |
1858 // Check if biased and fall through to runtime if so | |
1859 __ delayed()->nop(); | |
1860 __ btst(markOopDesc::biased_lock_bit_in_place, header); | |
1861 __ br(Assembler::notZero, false, Assembler::pn, slowCase); | |
1862 } | |
1863 __ delayed()->or3(mask, markOopDesc::hash_mask & 0x3ff, mask); | |
1864 | |
1865 // Check for a valid (non-zero) hash code and get its value. | |
1866 #ifdef _LP64 | |
1867 __ srlx(header, markOopDesc::hash_shift, hash); | |
1868 #else | |
1869 __ srl(header, markOopDesc::hash_shift, hash); | |
1870 #endif | |
1871 __ andcc(hash, mask, hash); | |
1872 __ br(Assembler::equal, false, Assembler::pn, slowCase); | |
1873 __ delayed()->nop(); | |
1874 | |
1875 // leaf return. | |
1876 __ retl(); | |
1877 __ delayed()->mov(hash, result); | |
1878 __ bind(slowCase); | |
1879 } | |
1880 #endif // COMPILER1 | |
1881 | |
1882 | |
1883 // We have received a description of where all the java arg are located | |
1884 // on entry to the wrapper. We need to convert these args to where | |
1885 // the jni function will expect them. To figure out where they go | |
1886 // we convert the java signature to a C signature by inserting | |
1887 // the hidden arguments as arg[0] and possibly arg[1] (static method) | |
1888 | |
1889 int total_c_args = total_in_args + 1; | |
1890 if (method->is_static()) { | |
1891 total_c_args++; | |
1892 } | |
1893 | |
1894 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); | |
1895 VMRegPair * out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); | |
1896 | |
1897 int argc = 0; | |
1898 out_sig_bt[argc++] = T_ADDRESS; | |
1899 if (method->is_static()) { | |
1900 out_sig_bt[argc++] = T_OBJECT; | |
1901 } | |
1902 | |
1903 for (int i = 0; i < total_in_args ; i++ ) { | |
1904 out_sig_bt[argc++] = in_sig_bt[i]; | |
1905 } | |
1906 | |
1907 // Now figure out where the args must be stored and how much stack space | |
1908 // they require (neglecting out_preserve_stack_slots but space for storing | |
1909 // the 1st six register arguments). It's weird see int_stk_helper. | |
1910 // | |
1911 int out_arg_slots; | |
1912 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); | |
1913 | |
1914 // Compute framesize for the wrapper. We need to handlize all oops in | |
1915 // registers. We must create space for them here that is disjoint from | |
1916 // the windowed save area because we have no control over when we might | |
1917 // flush the window again and overwrite values that gc has since modified. | |
1918 // (The live window race) | |
1919 // | |
1920 // We always just allocate 6 word for storing down these object. This allow | |
1921 // us to simply record the base and use the Ireg number to decide which | |
1922 // slot to use. (Note that the reg number is the inbound number not the | |
1923 // outbound number). | |
1924 // We must shuffle args to match the native convention, and include var-args space. | |
1925 | |
1926 // Calculate the total number of stack slots we will need. | |
1927 | |
1928 // First count the abi requirement plus all of the outgoing args | |
1929 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; | |
1930 | |
1931 // Now the space for the inbound oop handle area | |
1932 | |
1933 int oop_handle_offset = stack_slots; | |
1934 stack_slots += 6*VMRegImpl::slots_per_word; | |
1935 | |
1936 // Now any space we need for handlizing a klass if static method | |
1937 | |
1938 int oop_temp_slot_offset = 0; | |
1939 int klass_slot_offset = 0; | |
1940 int klass_offset = -1; | |
1941 int lock_slot_offset = 0; | |
1942 bool is_static = false; | |
1943 | |
1944 if (method->is_static()) { | |
1945 klass_slot_offset = stack_slots; | |
1946 stack_slots += VMRegImpl::slots_per_word; | |
1947 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; | |
1948 is_static = true; | |
1949 } | |
1950 | |
1951 // Plus a lock if needed | |
1952 | |
1953 if (method->is_synchronized()) { | |
1954 lock_slot_offset = stack_slots; | |
1955 stack_slots += VMRegImpl::slots_per_word; | |
1956 } | |
1957 | |
1958 // Now a place to save return value or as a temporary for any gpr -> fpr moves | |
1959 stack_slots += 2; | |
1960 | |
1961 // Ok The space we have allocated will look like: | |
1962 // | |
1963 // | |
1964 // FP-> | | | |
1965 // |---------------------| | |
1966 // | 2 slots for moves | | |
1967 // |---------------------| | |
1968 // | lock box (if sync) | | |
1969 // |---------------------| <- lock_slot_offset | |
1970 // | klass (if static) | | |
1971 // |---------------------| <- klass_slot_offset | |
1972 // | oopHandle area | | |
1973 // |---------------------| <- oop_handle_offset | |
1974 // | outbound memory | | |
1975 // | based arguments | | |
1976 // | | | |
1977 // |---------------------| | |
1978 // | vararg area | | |
1979 // |---------------------| | |
1980 // | | | |
1981 // SP-> | out_preserved_slots | | |
1982 // | |
1983 // | |
1984 | |
1985 | |
1986 // Now compute actual number of stack words we need rounding to make | |
1987 // stack properly aligned. | |
1988 stack_slots = round_to(stack_slots, 2 * VMRegImpl::slots_per_word); | |
1989 | |
1990 int stack_size = stack_slots * VMRegImpl::stack_slot_size; | |
1991 | |
1992 // Generate stack overflow check before creating frame | |
1993 __ generate_stack_overflow_check(stack_size); | |
1994 | |
1995 // Generate a new frame for the wrapper. | |
1996 __ save(SP, -stack_size, SP); | |
1997 | |
1998 int frame_complete = ((intptr_t)__ pc()) - start; | |
1999 | |
2000 __ verify_thread(); | |
2001 | |
2002 | |
2003 // | |
2004 // We immediately shuffle the arguments so that any vm call we have to | |
2005 // make from here on out (sync slow path, jvmti, etc.) we will have | |
2006 // captured the oops from our caller and have a valid oopMap for | |
2007 // them. | |
2008 | |
2009 // ----------------- | |
2010 // The Grand Shuffle | |
2011 // | |
2012 // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv* | |
2013 // (derived from JavaThread* which is in L7_thread_cache) and, if static, | |
2014 // the class mirror instead of a receiver. This pretty much guarantees that | |
2015 // register layout will not match. We ignore these extra arguments during | |
2016 // the shuffle. The shuffle is described by the two calling convention | |
2017 // vectors we have in our possession. We simply walk the java vector to | |
2018 // get the source locations and the c vector to get the destinations. | |
2019 // Because we have a new window and the argument registers are completely | |
2020 // disjoint ( I0 -> O1, I1 -> O2, ...) we have nothing to worry about | |
2021 // here. | |
2022 | |
2023 // This is a trick. We double the stack slots so we can claim | |
2024 // the oops in the caller's frame. Since we are sure to have | |
2025 // more args than the caller doubling is enough to make | |
2026 // sure we can capture all the incoming oop args from the | |
2027 // caller. | |
2028 // | |
2029 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); | |
2030 int c_arg = total_c_args - 1; | |
2031 // Record sp-based slot for receiver on stack for non-static methods | |
2032 int receiver_offset = -1; | |
2033 | |
2034 // We move the arguments backward because the floating point registers | |
2035 // destination will always be to a register with a greater or equal register | |
2036 // number or the stack. | |
2037 | |
2038 #ifdef ASSERT | |
2039 bool reg_destroyed[RegisterImpl::number_of_registers]; | |
2040 bool freg_destroyed[FloatRegisterImpl::number_of_registers]; | |
2041 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { | |
2042 reg_destroyed[r] = false; | |
2043 } | |
2044 for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) { | |
2045 freg_destroyed[f] = false; | |
2046 } | |
2047 | |
2048 #endif /* ASSERT */ | |
2049 | |
2050 for ( int i = total_in_args - 1; i >= 0 ; i--, c_arg-- ) { | |
2051 | |
2052 #ifdef ASSERT | |
2053 if (in_regs[i].first()->is_Register()) { | |
2054 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "ack!"); | |
2055 } else if (in_regs[i].first()->is_FloatRegister()) { | |
2056 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding(FloatRegisterImpl::S)], "ack!"); | |
2057 } | |
2058 if (out_regs[c_arg].first()->is_Register()) { | |
2059 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; | |
2060 } else if (out_regs[c_arg].first()->is_FloatRegister()) { | |
2061 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding(FloatRegisterImpl::S)] = true; | |
2062 } | |
2063 #endif /* ASSERT */ | |
2064 | |
2065 switch (in_sig_bt[i]) { | |
2066 case T_ARRAY: | |
2067 case T_OBJECT: | |
2068 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], | |
2069 ((i == 0) && (!is_static)), | |
2070 &receiver_offset); | |
2071 break; | |
2072 case T_VOID: | |
2073 break; | |
2074 | |
2075 case T_FLOAT: | |
2076 float_move(masm, in_regs[i], out_regs[c_arg]); | |
2077 break; | |
2078 | |
2079 case T_DOUBLE: | |
2080 assert( i + 1 < total_in_args && | |
2081 in_sig_bt[i + 1] == T_VOID && | |
2082 out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); | |
2083 double_move(masm, in_regs[i], out_regs[c_arg]); | |
2084 break; | |
2085 | |
2086 case T_LONG : | |
2087 long_move(masm, in_regs[i], out_regs[c_arg]); | |
2088 break; | |
2089 | |
2090 case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); | |
2091 | |
2092 default: | |
2093 move32_64(masm, in_regs[i], out_regs[c_arg]); | |
2094 } | |
2095 } | |
2096 | |
2097 // Pre-load a static method's oop into O1. Used both by locking code and | |
2098 // the normal JNI call code. | |
2099 if (method->is_static()) { | |
2100 __ set_oop_constant(JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror()), O1); | |
2101 | |
2102 // Now handlize the static class mirror in O1. It's known not-null. | |
2103 __ st_ptr(O1, SP, klass_offset + STACK_BIAS); | |
2104 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); | |
2105 __ add(SP, klass_offset + STACK_BIAS, O1); | |
2106 } | |
2107 | |
2108 | |
2109 const Register L6_handle = L6; | |
2110 | |
2111 if (method->is_synchronized()) { | |
2112 __ mov(O1, L6_handle); | |
2113 } | |
2114 | |
2115 // We have all of the arguments setup at this point. We MUST NOT touch any Oregs | |
2116 // except O6/O7. So if we must call out we must push a new frame. We immediately | |
2117 // push a new frame and flush the windows. | |
2118 | |
2119 #ifdef _LP64 | |
2120 intptr_t thepc = (intptr_t) __ pc(); | |
2121 { | |
2122 address here = __ pc(); | |
2123 // Call the next instruction | |
2124 __ call(here + 8, relocInfo::none); | |
2125 __ delayed()->nop(); | |
2126 } | |
2127 #else | |
2128 intptr_t thepc = __ load_pc_address(O7, 0); | |
2129 #endif /* _LP64 */ | |
2130 | |
2131 // We use the same pc/oopMap repeatedly when we call out | |
2132 oop_maps->add_gc_map(thepc - start, map); | |
2133 | |
2134 // O7 now has the pc loaded that we will use when we finally call to native. | |
2135 | |
2136 // Save thread in L7; it crosses a bunch of VM calls below | |
2137 // Don't use save_thread because it smashes G2 and we merely | |
2138 // want to save a copy | |
2139 __ mov(G2_thread, L7_thread_cache); | |
2140 | |
2141 | |
2142 // If we create an inner frame once is plenty | |
2143 // when we create it we must also save G2_thread | |
2144 bool inner_frame_created = false; | |
2145 | |
2146 // dtrace method entry support | |
2147 { | |
2148 SkipIfEqual skip_if( | |
2149 masm, G3_scratch, &DTraceMethodProbes, Assembler::zero); | |
2150 // create inner frame | |
2151 __ save_frame(0); | |
2152 __ mov(G2_thread, L7_thread_cache); | |
2153 __ set_oop_constant(JNIHandles::make_local(method()), O1); | |
2154 __ call_VM_leaf(L7_thread_cache, | |
2155 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), | |
2156 G2_thread, O1); | |
2157 __ restore(); | |
2158 } | |
2159 | |
2160 // We are in the jni frame unless saved_frame is true in which case | |
2161 // we are in one frame deeper (the "inner" frame). If we are in the | |
2162 // "inner" frames the args are in the Iregs and if the jni frame then | |
2163 // they are in the Oregs. | |
2164 // If we ever need to go to the VM (for locking, jvmti) then | |
2165 // we will always be in the "inner" frame. | |
2166 | |
2167 // Lock a synchronized method | |
2168 int lock_offset = -1; // Set if locked | |
2169 if (method->is_synchronized()) { | |
2170 Register Roop = O1; | |
2171 const Register L3_box = L3; | |
2172 | |
2173 create_inner_frame(masm, &inner_frame_created); | |
2174 | |
2175 __ ld_ptr(I1, 0, O1); | |
2176 Label done; | |
2177 | |
2178 lock_offset = (lock_slot_offset * VMRegImpl::stack_slot_size); | |
2179 __ add(FP, lock_offset+STACK_BIAS, L3_box); | |
2180 #ifdef ASSERT | |
2181 if (UseBiasedLocking) { | |
2182 // making the box point to itself will make it clear it went unused | |
2183 // but also be obviously invalid | |
2184 __ st_ptr(L3_box, L3_box, 0); | |
2185 } | |
2186 #endif // ASSERT | |
2187 // | |
2188 // Compiler_lock_object (Roop, Rmark, Rbox, Rscratch) -- kills Rmark, Rbox, Rscratch | |
2189 // | |
2190 __ compiler_lock_object(Roop, L1, L3_box, L2); | |
2191 __ br(Assembler::equal, false, Assembler::pt, done); | |
2192 __ delayed() -> add(FP, lock_offset+STACK_BIAS, L3_box); | |
2193 | |
2194 | |
2195 // None of the above fast optimizations worked so we have to get into the | |
2196 // slow case of monitor enter. Inline a special case of call_VM that | |
2197 // disallows any pending_exception. | |
2198 __ mov(Roop, O0); // Need oop in O0 | |
2199 __ mov(L3_box, O1); | |
2200 | |
2201 // Record last_Java_sp, in case the VM code releases the JVM lock. | |
2202 | |
2203 __ set_last_Java_frame(FP, I7); | |
2204 | |
2205 // do the call | |
2206 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), relocInfo::runtime_call_type); | |
2207 __ delayed()->mov(L7_thread_cache, O2); | |
2208 | |
2209 __ restore_thread(L7_thread_cache); // restore G2_thread | |
2210 __ reset_last_Java_frame(); | |
2211 | |
2212 #ifdef ASSERT | |
2213 { Label L; | |
2214 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), O0); | |
2215 __ br_null(O0, false, Assembler::pt, L); | |
2216 __ delayed()->nop(); | |
2217 __ stop("no pending exception allowed on exit from IR::monitorenter"); | |
2218 __ bind(L); | |
2219 } | |
2220 #endif | |
2221 __ bind(done); | |
2222 } | |
2223 | |
2224 | |
2225 // Finally just about ready to make the JNI call | |
2226 | |
2227 __ flush_windows(); | |
2228 if (inner_frame_created) { | |
2229 __ restore(); | |
2230 } else { | |
2231 // Store only what we need from this frame | |
2232 // QQQ I think that non-v9 (like we care) we don't need these saves | |
2233 // either as the flush traps and the current window goes too. | |
2234 __ st_ptr(FP, SP, FP->sp_offset_in_saved_window()*wordSize + STACK_BIAS); | |
2235 __ st_ptr(I7, SP, I7->sp_offset_in_saved_window()*wordSize + STACK_BIAS); | |
2236 } | |
2237 | |
2238 // get JNIEnv* which is first argument to native | |
2239 | |
2240 __ add(G2_thread, in_bytes(JavaThread::jni_environment_offset()), O0); | |
2241 | |
2242 // Use that pc we placed in O7 a while back as the current frame anchor | |
2243 | |
2244 __ set_last_Java_frame(SP, O7); | |
2245 | |
2246 // Transition from _thread_in_Java to _thread_in_native. | |
2247 __ set(_thread_in_native, G3_scratch); | |
2248 __ st(G3_scratch, G2_thread, in_bytes(JavaThread::thread_state_offset())); | |
2249 | |
2250 // We flushed the windows ages ago now mark them as flushed | |
2251 | |
2252 // mark windows as flushed | |
2253 __ set(JavaFrameAnchor::flushed, G3_scratch); | |
2254 | |
2255 Address flags(G2_thread, | |
2256 0, | |
2257 in_bytes(JavaThread::frame_anchor_offset()) + in_bytes(JavaFrameAnchor::flags_offset())); | |
2258 | |
2259 #ifdef _LP64 | |
2260 Address dest(O7, method->native_function()); | |
2261 __ relocate(relocInfo::runtime_call_type); | |
2262 __ jumpl_to(dest, O7); | |
2263 #else | |
2264 __ call(method->native_function(), relocInfo::runtime_call_type); | |
2265 #endif | |
2266 __ delayed()->st(G3_scratch, flags); | |
2267 | |
2268 __ restore_thread(L7_thread_cache); // restore G2_thread | |
2269 | |
2270 // Unpack native results. For int-types, we do any needed sign-extension | |
2271 // and move things into I0. The return value there will survive any VM | |
2272 // calls for blocking or unlocking. An FP or OOP result (handle) is done | |
2273 // specially in the slow-path code. | |
2274 switch (ret_type) { | |
2275 case T_VOID: break; // Nothing to do! | |
2276 case T_FLOAT: break; // Got it where we want it (unless slow-path) | |
2277 case T_DOUBLE: break; // Got it where we want it (unless slow-path) | |
2278 // In 64 bits build result is in O0, in O0, O1 in 32bit build | |
2279 case T_LONG: | |
2280 #ifndef _LP64 | |
2281 __ mov(O1, I1); | |
2282 #endif | |
2283 // Fall thru | |
2284 case T_OBJECT: // Really a handle | |
2285 case T_ARRAY: | |
2286 case T_INT: | |
2287 __ mov(O0, I0); | |
2288 break; | |
2289 case T_BOOLEAN: __ subcc(G0, O0, G0); __ addc(G0, 0, I0); break; // !0 => true; 0 => false | |
2290 case T_BYTE : __ sll(O0, 24, O0); __ sra(O0, 24, I0); break; | |
2291 case T_CHAR : __ sll(O0, 16, O0); __ srl(O0, 16, I0); break; // cannot use and3, 0xFFFF too big as immediate value! | |
2292 case T_SHORT : __ sll(O0, 16, O0); __ sra(O0, 16, I0); break; | |
2293 break; // Cannot de-handlize until after reclaiming jvm_lock | |
2294 default: | |
2295 ShouldNotReachHere(); | |
2296 } | |
2297 | |
2298 // must we block? | |
2299 | |
2300 // Block, if necessary, before resuming in _thread_in_Java state. | |
2301 // In order for GC to work, don't clear the last_Java_sp until after blocking. | |
2302 { Label no_block; | |
2303 Address sync_state(G3_scratch, SafepointSynchronize::address_of_state()); | |
2304 | |
2305 // Switch thread to "native transition" state before reading the synchronization state. | |
2306 // This additional state is necessary because reading and testing the synchronization | |
2307 // state is not atomic w.r.t. GC, as this scenario demonstrates: | |
2308 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. | |
2309 // VM thread changes sync state to synchronizing and suspends threads for GC. | |
2310 // Thread A is resumed to finish this native method, but doesn't block here since it | |
2311 // didn't see any synchronization is progress, and escapes. | |
2312 __ set(_thread_in_native_trans, G3_scratch); | |
2313 __ st(G3_scratch, G2_thread, in_bytes(JavaThread::thread_state_offset())); | |
2314 if(os::is_MP()) { | |
2315 if (UseMembar) { | |
2316 // Force this write out before the read below | |
2317 __ membar(Assembler::StoreLoad); | |
2318 } else { | |
2319 // Write serialization page so VM thread can do a pseudo remote membar. | |
2320 // We use the current thread pointer to calculate a thread specific | |
2321 // offset to write to within the page. This minimizes bus traffic | |
2322 // due to cache line collision. | |
2323 __ serialize_memory(G2_thread, G1_scratch, G3_scratch); | |
2324 } | |
2325 } | |
2326 __ load_contents(sync_state, G3_scratch); | |
2327 __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized); | |
2328 | |
2329 Label L; | |
2330 Address suspend_state(G2_thread, 0, in_bytes(JavaThread::suspend_flags_offset())); | |
2331 __ br(Assembler::notEqual, false, Assembler::pn, L); | |
2332 __ delayed()-> | |
2333 ld(suspend_state, G3_scratch); | |
2334 __ cmp(G3_scratch, 0); | |
2335 __ br(Assembler::equal, false, Assembler::pt, no_block); | |
2336 __ delayed()->nop(); | |
2337 __ bind(L); | |
2338 | |
2339 // Block. Save any potential method result value before the operation and | |
2340 // use a leaf call to leave the last_Java_frame setup undisturbed. Doing this | |
2341 // lets us share the oopMap we used when we went native rather the create | |
2342 // a distinct one for this pc | |
2343 // | |
2344 save_native_result(masm, ret_type, stack_slots); | |
2345 __ call_VM_leaf(L7_thread_cache, | |
2346 CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), | |
2347 G2_thread); | |
2348 | |
2349 // Restore any method result value | |
2350 restore_native_result(masm, ret_type, stack_slots); | |
2351 __ bind(no_block); | |
2352 } | |
2353 | |
2354 // thread state is thread_in_native_trans. Any safepoint blocking has already | |
2355 // happened so we can now change state to _thread_in_Java. | |
2356 | |
2357 | |
2358 __ set(_thread_in_Java, G3_scratch); | |
2359 __ st(G3_scratch, G2_thread, in_bytes(JavaThread::thread_state_offset())); | |
2360 | |
2361 | |
2362 Label no_reguard; | |
2363 __ ld(G2_thread, in_bytes(JavaThread::stack_guard_state_offset()), G3_scratch); | |
2364 __ cmp(G3_scratch, JavaThread::stack_guard_yellow_disabled); | |
2365 __ br(Assembler::notEqual, false, Assembler::pt, no_reguard); | |
2366 __ delayed()->nop(); | |
2367 | |
2368 save_native_result(masm, ret_type, stack_slots); | |
2369 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)); | |
2370 __ delayed()->nop(); | |
2371 | |
2372 __ restore_thread(L7_thread_cache); // restore G2_thread | |
2373 restore_native_result(masm, ret_type, stack_slots); | |
2374 | |
2375 __ bind(no_reguard); | |
2376 | |
2377 // Handle possible exception (will unlock if necessary) | |
2378 | |
2379 // native result if any is live in freg or I0 (and I1 if long and 32bit vm) | |
2380 | |
2381 // Unlock | |
2382 if (method->is_synchronized()) { | |
2383 Label done; | |
2384 Register I2_ex_oop = I2; | |
2385 const Register L3_box = L3; | |
2386 // Get locked oop from the handle we passed to jni | |
2387 __ ld_ptr(L6_handle, 0, L4); | |
2388 __ add(SP, lock_offset+STACK_BIAS, L3_box); | |
2389 // Must save pending exception around the slow-path VM call. Since it's a | |
2390 // leaf call, the pending exception (if any) can be kept in a register. | |
2391 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), I2_ex_oop); | |
2392 // Now unlock | |
2393 // (Roop, Rmark, Rbox, Rscratch) | |
2394 __ compiler_unlock_object(L4, L1, L3_box, L2); | |
2395 __ br(Assembler::equal, false, Assembler::pt, done); | |
2396 __ delayed()-> add(SP, lock_offset+STACK_BIAS, L3_box); | |
2397 | |
2398 // save and restore any potential method result value around the unlocking | |
2399 // operation. Will save in I0 (or stack for FP returns). | |
2400 save_native_result(masm, ret_type, stack_slots); | |
2401 | |
2402 // Must clear pending-exception before re-entering the VM. Since this is | |
2403 // a leaf call, pending-exception-oop can be safely kept in a register. | |
2404 __ st_ptr(G0, G2_thread, in_bytes(Thread::pending_exception_offset())); | |
2405 | |
2406 // slow case of monitor enter. Inline a special case of call_VM that | |
2407 // disallows any pending_exception. | |
2408 __ mov(L3_box, O1); | |
2409 | |
2410 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C), relocInfo::runtime_call_type); | |
2411 __ delayed()->mov(L4, O0); // Need oop in O0 | |
2412 | |
2413 __ restore_thread(L7_thread_cache); // restore G2_thread | |
2414 | |
2415 #ifdef ASSERT | |
2416 { Label L; | |
2417 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), O0); | |
2418 __ br_null(O0, false, Assembler::pt, L); | |
2419 __ delayed()->nop(); | |
2420 __ stop("no pending exception allowed on exit from IR::monitorexit"); | |
2421 __ bind(L); | |
2422 } | |
2423 #endif | |
2424 restore_native_result(masm, ret_type, stack_slots); | |
2425 // check_forward_pending_exception jump to forward_exception if any pending | |
2426 // exception is set. The forward_exception routine expects to see the | |
2427 // exception in pending_exception and not in a register. Kind of clumsy, | |
2428 // since all folks who branch to forward_exception must have tested | |
2429 // pending_exception first and hence have it in a register already. | |
2430 __ st_ptr(I2_ex_oop, G2_thread, in_bytes(Thread::pending_exception_offset())); | |
2431 __ bind(done); | |
2432 } | |
2433 | |
2434 // Tell dtrace about this method exit | |
2435 { | |
2436 SkipIfEqual skip_if( | |
2437 masm, G3_scratch, &DTraceMethodProbes, Assembler::zero); | |
2438 save_native_result(masm, ret_type, stack_slots); | |
2439 __ set_oop_constant(JNIHandles::make_local(method()), O1); | |
2440 __ call_VM_leaf(L7_thread_cache, | |
2441 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), | |
2442 G2_thread, O1); | |
2443 restore_native_result(masm, ret_type, stack_slots); | |
2444 } | |
2445 | |
2446 // Clear "last Java frame" SP and PC. | |
2447 __ verify_thread(); // G2_thread must be correct | |
2448 __ reset_last_Java_frame(); | |
2449 | |
2450 // Unpack oop result | |
2451 if (ret_type == T_OBJECT || ret_type == T_ARRAY) { | |
2452 Label L; | |
2453 __ addcc(G0, I0, G0); | |
2454 __ brx(Assembler::notZero, true, Assembler::pt, L); | |
2455 __ delayed()->ld_ptr(I0, 0, I0); | |
2456 __ mov(G0, I0); | |
2457 __ bind(L); | |
2458 __ verify_oop(I0); | |
2459 } | |
2460 | |
2461 // reset handle block | |
2462 __ ld_ptr(G2_thread, in_bytes(JavaThread::active_handles_offset()), L5); | |
2463 __ st_ptr(G0, L5, JNIHandleBlock::top_offset_in_bytes()); | |
2464 | |
2465 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), G3_scratch); | |
2466 check_forward_pending_exception(masm, G3_scratch); | |
2467 | |
2468 | |
2469 // Return | |
2470 | |
2471 #ifndef _LP64 | |
2472 if (ret_type == T_LONG) { | |
2473 | |
2474 // Must leave proper result in O0,O1 and G1 (c2/tiered only) | |
2475 __ sllx(I0, 32, G1); // Shift bits into high G1 | |
2476 __ srl (I1, 0, I1); // Zero extend O1 (harmless?) | |
2477 __ or3 (I1, G1, G1); // OR 64 bits into G1 | |
2478 } | |
2479 #endif | |
2480 | |
2481 __ ret(); | |
2482 __ delayed()->restore(); | |
2483 | |
2484 __ flush(); | |
2485 | |
2486 nmethod *nm = nmethod::new_native_nmethod(method, | |
2487 masm->code(), | |
2488 vep_offset, | |
2489 frame_complete, | |
2490 stack_slots / VMRegImpl::slots_per_word, | |
2491 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), | |
2492 in_ByteSize(lock_offset), | |
2493 oop_maps); | |
2494 return nm; | |
2495 | |
2496 } | |
2497 | |
2498 // this function returns the adjust size (in number of words) to a c2i adapter | |
2499 // activation for use during deoptimization | |
2500 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) { | |
2501 assert(callee_locals >= callee_parameters, | |
2502 "test and remove; got more parms than locals"); | |
2503 if (callee_locals < callee_parameters) | |
2504 return 0; // No adjustment for negative locals | |
2505 int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords(); | |
2506 return round_to(diff, WordsPerLong); | |
2507 } | |
2508 | |
2509 // "Top of Stack" slots that may be unused by the calling convention but must | |
2510 // otherwise be preserved. | |
2511 // On Intel these are not necessary and the value can be zero. | |
2512 // On Sparc this describes the words reserved for storing a register window | |
2513 // when an interrupt occurs. | |
2514 uint SharedRuntime::out_preserve_stack_slots() { | |
2515 return frame::register_save_words * VMRegImpl::slots_per_word; | |
2516 } | |
2517 | |
2518 static void gen_new_frame(MacroAssembler* masm, bool deopt) { | |
2519 // | |
2520 // Common out the new frame generation for deopt and uncommon trap | |
2521 // | |
2522 Register G3pcs = G3_scratch; // Array of new pcs (input) | |
2523 Register Oreturn0 = O0; | |
2524 Register Oreturn1 = O1; | |
2525 Register O2UnrollBlock = O2; | |
2526 Register O3array = O3; // Array of frame sizes (input) | |
2527 Register O4array_size = O4; // number of frames (input) | |
2528 Register O7frame_size = O7; // number of frames (input) | |
2529 | |
2530 __ ld_ptr(O3array, 0, O7frame_size); | |
2531 __ sub(G0, O7frame_size, O7frame_size); | |
2532 __ save(SP, O7frame_size, SP); | |
2533 __ ld_ptr(G3pcs, 0, I7); // load frame's new pc | |
2534 | |
2535 #ifdef ASSERT | |
2536 // make sure that the frames are aligned properly | |
2537 #ifndef _LP64 | |
2538 __ btst(wordSize*2-1, SP); | |
2539 __ breakpoint_trap(Assembler::notZero); | |
2540 #endif | |
2541 #endif | |
2542 | |
2543 // Deopt needs to pass some extra live values from frame to frame | |
2544 | |
2545 if (deopt) { | |
2546 __ mov(Oreturn0->after_save(), Oreturn0); | |
2547 __ mov(Oreturn1->after_save(), Oreturn1); | |
2548 } | |
2549 | |
2550 __ mov(O4array_size->after_save(), O4array_size); | |
2551 __ sub(O4array_size, 1, O4array_size); | |
2552 __ mov(O3array->after_save(), O3array); | |
2553 __ mov(O2UnrollBlock->after_save(), O2UnrollBlock); | |
2554 __ add(G3pcs, wordSize, G3pcs); // point to next pc value | |
2555 | |
2556 #ifdef ASSERT | |
2557 // trash registers to show a clear pattern in backtraces | |
2558 __ set(0xDEAD0000, I0); | |
2559 __ add(I0, 2, I1); | |
2560 __ add(I0, 4, I2); | |
2561 __ add(I0, 6, I3); | |
2562 __ add(I0, 8, I4); | |
2563 // Don't touch I5 could have valuable savedSP | |
2564 __ set(0xDEADBEEF, L0); | |
2565 __ mov(L0, L1); | |
2566 __ mov(L0, L2); | |
2567 __ mov(L0, L3); | |
2568 __ mov(L0, L4); | |
2569 __ mov(L0, L5); | |
2570 | |
2571 // trash the return value as there is nothing to return yet | |
2572 __ set(0xDEAD0001, O7); | |
2573 #endif | |
2574 | |
2575 __ mov(SP, O5_savedSP); | |
2576 } | |
2577 | |
2578 | |
2579 static void make_new_frames(MacroAssembler* masm, bool deopt) { | |
2580 // | |
2581 // loop through the UnrollBlock info and create new frames | |
2582 // | |
2583 Register G3pcs = G3_scratch; | |
2584 Register Oreturn0 = O0; | |
2585 Register Oreturn1 = O1; | |
2586 Register O2UnrollBlock = O2; | |
2587 Register O3array = O3; | |
2588 Register O4array_size = O4; | |
2589 Label loop; | |
2590 | |
2591 // Before we make new frames, check to see if stack is available. | |
2592 // Do this after the caller's return address is on top of stack | |
2593 if (UseStackBanging) { | |
2594 // Get total frame size for interpreted frames | |
2595 __ ld(Address(O2UnrollBlock, 0, | |
2596 Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()), O4); | |
2597 __ bang_stack_size(O4, O3, G3_scratch); | |
2598 } | |
2599 | |
2600 __ ld(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()), O4array_size); | |
2601 __ ld_ptr(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()), G3pcs); | |
2602 | |
2603 __ ld_ptr(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()), O3array); | |
2604 | |
2605 // Adjust old interpreter frame to make space for new frame's extra java locals | |
2606 // | |
2607 // We capture the original sp for the transition frame only because it is needed in | |
2608 // order to properly calculate interpreter_sp_adjustment. Even though in real life | |
2609 // every interpreter frame captures a savedSP it is only needed at the transition | |
2610 // (fortunately). If we had to have it correct everywhere then we would need to | |
2611 // be told the sp_adjustment for each frame we create. If the frame size array | |
2612 // were to have twice the frame count entries then we could have pairs [sp_adjustment, frame_size] | |
2613 // for each frame we create and keep up the illusion every where. | |
2614 // | |
2615 | |
2616 __ ld(Address(O2UnrollBlock, 0, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()), O7); | |
2617 __ mov(SP, O5_savedSP); // remember initial sender's original sp before adjustment | |
2618 __ sub(SP, O7, SP); | |
2619 | |
2620 #ifdef ASSERT | |
2621 // make sure that there is at least one entry in the array | |
2622 __ tst(O4array_size); | |
2623 __ breakpoint_trap(Assembler::zero); | |
2624 #endif | |
2625 | |
2626 // Now push the new interpreter frames | |
2627 __ bind(loop); | |
2628 | |
2629 // allocate a new frame, filling the registers | |
2630 | |
2631 gen_new_frame(masm, deopt); // allocate an interpreter frame | |
2632 | |
2633 __ tst(O4array_size); | |
2634 __ br(Assembler::notZero, false, Assembler::pn, loop); | |
2635 __ delayed()->add(O3array, wordSize, O3array); | |
2636 __ ld_ptr(G3pcs, 0, O7); // load final frame new pc | |
2637 | |
2638 } | |
2639 | |
2640 //------------------------------generate_deopt_blob---------------------------- | |
2641 // Ought to generate an ideal graph & compile, but here's some SPARC ASM | |
2642 // instead. | |
2643 void SharedRuntime::generate_deopt_blob() { | |
2644 // allocate space for the code | |
2645 ResourceMark rm; | |
2646 // setup code generation tools | |
2647 int pad = VerifyThread ? 512 : 0;// Extra slop space for more verify code | |
2648 #ifdef _LP64 | |
2649 CodeBuffer buffer("deopt_blob", 2100+pad, 512); | |
2650 #else | |
2651 // Measured 8/7/03 at 1212 in 32bit debug build (no VerifyThread) | |
2652 // Measured 8/7/03 at 1396 in 32bit debug build (VerifyThread) | |
2653 CodeBuffer buffer("deopt_blob", 1600+pad, 512); | |
2654 #endif /* _LP64 */ | |
2655 MacroAssembler* masm = new MacroAssembler(&buffer); | |
2656 FloatRegister Freturn0 = F0; | |
2657 Register Greturn1 = G1; | |
2658 Register Oreturn0 = O0; | |
2659 Register Oreturn1 = O1; | |
2660 Register O2UnrollBlock = O2; | |
2661 Register O3tmp = O3; | |
2662 Register I5exception_tmp = I5; | |
2663 Register G4exception_tmp = G4_scratch; | |
2664 int frame_size_words; | |
2665 Address saved_Freturn0_addr(FP, 0, -sizeof(double) + STACK_BIAS); | |
2666 #if !defined(_LP64) && defined(COMPILER2) | |
2667 Address saved_Greturn1_addr(FP, 0, -sizeof(double) -sizeof(jlong) + STACK_BIAS); | |
2668 #endif | |
2669 Label cont; | |
2670 | |
2671 OopMapSet *oop_maps = new OopMapSet(); | |
2672 | |
2673 // | |
2674 // This is the entry point for code which is returning to a de-optimized | |
2675 // frame. | |
2676 // The steps taken by this frame are as follows: | |
2677 // - push a dummy "register_save" and save the return values (O0, O1, F0/F1, G1) | |
2678 // and all potentially live registers (at a pollpoint many registers can be live). | |
2679 // | |
2680 // - call the C routine: Deoptimization::fetch_unroll_info (this function | |
2681 // returns information about the number and size of interpreter frames | |
2682 // which are equivalent to the frame which is being deoptimized) | |
2683 // - deallocate the unpack frame, restoring only results values. Other | |
2684 // volatile registers will now be captured in the vframeArray as needed. | |
2685 // - deallocate the deoptimization frame | |
2686 // - in a loop using the information returned in the previous step | |
2687 // push new interpreter frames (take care to propagate the return | |
2688 // values through each new frame pushed) | |
2689 // - create a dummy "unpack_frame" and save the return values (O0, O1, F0) | |
2690 // - call the C routine: Deoptimization::unpack_frames (this function | |
2691 // lays out values on the interpreter frame which was just created) | |
2692 // - deallocate the dummy unpack_frame | |
2693 // - ensure that all the return values are correctly set and then do | |
2694 // a return to the interpreter entry point | |
2695 // | |
2696 // Refer to the following methods for more information: | |
2697 // - Deoptimization::fetch_unroll_info | |
2698 // - Deoptimization::unpack_frames | |
2699 | |
2700 OopMap* map = NULL; | |
2701 | |
2702 int start = __ offset(); | |
2703 | |
2704 // restore G2, the trampoline destroyed it | |
2705 __ get_thread(); | |
2706 | |
2707 // On entry we have been called by the deoptimized nmethod with a call that | |
2708 // replaced the original call (or safepoint polling location) so the deoptimizing | |
2709 // pc is now in O7. Return values are still in the expected places | |
2710 | |
2711 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_words); | |
2712 __ ba(false, cont); | |
2713 __ delayed()->mov(Deoptimization::Unpack_deopt, I5exception_tmp); | |
2714 | |
2715 int exception_offset = __ offset() - start; | |
2716 | |
2717 // restore G2, the trampoline destroyed it | |
2718 __ get_thread(); | |
2719 | |
2720 // On entry we have been jumped to by the exception handler (or exception_blob | |
2721 // for server). O0 contains the exception oop and O7 contains the original | |
2722 // exception pc. So if we push a frame here it will look to the | |
2723 // stack walking code (fetch_unroll_info) just like a normal call so | |
2724 // state will be extracted normally. | |
2725 | |
2726 // save exception oop in JavaThread and fall through into the | |
2727 // exception_in_tls case since they are handled in same way except | |
2728 // for where the pending exception is kept. | |
2729 __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset())); | |
2730 | |
2731 // | |
2732 // Vanilla deoptimization with an exception pending in exception_oop | |
2733 // | |
2734 int exception_in_tls_offset = __ offset() - start; | |
2735 | |
2736 // No need to update oop_map as each call to save_live_registers will produce identical oopmap | |
2737 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_words); | |
2738 | |
2739 // Restore G2_thread | |
2740 __ get_thread(); | |
2741 | |
2742 #ifdef ASSERT | |
2743 { | |
2744 // verify that there is really an exception oop in exception_oop | |
2745 Label has_exception; | |
2746 __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), Oexception); | |
2747 __ br_notnull(Oexception, false, Assembler::pt, has_exception); | |
2748 __ delayed()-> nop(); | |
2749 __ stop("no exception in thread"); | |
2750 __ bind(has_exception); | |
2751 | |
2752 // verify that there is no pending exception | |
2753 Label no_pending_exception; | |
2754 Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset())); | |
2755 __ ld_ptr(exception_addr, Oexception); | |
2756 __ br_null(Oexception, false, Assembler::pt, no_pending_exception); | |
2757 __ delayed()->nop(); | |
2758 __ stop("must not have pending exception here"); | |
2759 __ bind(no_pending_exception); | |
2760 } | |
2761 #endif | |
2762 | |
2763 __ ba(false, cont); | |
2764 __ delayed()->mov(Deoptimization::Unpack_exception, I5exception_tmp);; | |
2765 | |
2766 // | |
2767 // Reexecute entry, similar to c2 uncommon trap | |
2768 // | |
2769 int reexecute_offset = __ offset() - start; | |
2770 | |
2771 // No need to update oop_map as each call to save_live_registers will produce identical oopmap | |
2772 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_words); | |
2773 | |
2774 __ mov(Deoptimization::Unpack_reexecute, I5exception_tmp); | |
2775 | |
2776 __ bind(cont); | |
2777 | |
2778 __ set_last_Java_frame(SP, noreg); | |
2779 | |
2780 // do the call by hand so we can get the oopmap | |
2781 | |
2782 __ mov(G2_thread, L7_thread_cache); | |
2783 __ call(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info), relocInfo::runtime_call_type); | |
2784 __ delayed()->mov(G2_thread, O0); | |
2785 | |
2786 // Set an oopmap for the call site this describes all our saved volatile registers | |
2787 | |
2788 oop_maps->add_gc_map( __ offset()-start, map); | |
2789 | |
2790 __ mov(L7_thread_cache, G2_thread); | |
2791 | |
2792 __ reset_last_Java_frame(); | |
2793 | |
2794 // NOTE: we know that only O0/O1 will be reloaded by restore_result_registers | |
2795 // so this move will survive | |
2796 | |
2797 __ mov(I5exception_tmp, G4exception_tmp); | |
2798 | |
2799 __ mov(O0, O2UnrollBlock->after_save()); | |
2800 | |
2801 RegisterSaver::restore_result_registers(masm); | |
2802 | |
2803 Label noException; | |
2804 __ cmp(G4exception_tmp, Deoptimization::Unpack_exception); // Was exception pending? | |
2805 __ br(Assembler::notEqual, false, Assembler::pt, noException); | |
2806 __ delayed()->nop(); | |
2807 | |
2808 // Move the pending exception from exception_oop to Oexception so | |
2809 // the pending exception will be picked up the interpreter. | |
2810 __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), Oexception); | |
2811 __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset())); | |
2812 __ bind(noException); | |
2813 | |
2814 // deallocate the deoptimization frame taking care to preserve the return values | |
2815 __ mov(Oreturn0, Oreturn0->after_save()); | |
2816 __ mov(Oreturn1, Oreturn1->after_save()); | |
2817 __ mov(O2UnrollBlock, O2UnrollBlock->after_save()); | |
2818 __ restore(); | |
2819 | |
2820 // Allocate new interpreter frame(s) and possible c2i adapter frame | |
2821 | |
2822 make_new_frames(masm, true); | |
2823 | |
2824 // push a dummy "unpack_frame" taking care of float return values and | |
2825 // call Deoptimization::unpack_frames to have the unpacker layout | |
2826 // information in the interpreter frames just created and then return | |
2827 // to the interpreter entry point | |
2828 __ save(SP, -frame_size_words*wordSize, SP); | |
2829 __ stf(FloatRegisterImpl::D, Freturn0, saved_Freturn0_addr); | |
2830 #if !defined(_LP64) | |
2831 #if defined(COMPILER2) | |
2832 if (!TieredCompilation) { | |
2833 // 32-bit 1-register longs return longs in G1 | |
2834 __ stx(Greturn1, saved_Greturn1_addr); | |
2835 } | |
2836 #endif | |
2837 __ set_last_Java_frame(SP, noreg); | |
2838 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames), G2_thread, G4exception_tmp); | |
2839 #else | |
2840 // LP64 uses g4 in set_last_Java_frame | |
2841 __ mov(G4exception_tmp, O1); | |
2842 __ set_last_Java_frame(SP, G0); | |
2843 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames), G2_thread, O1); | |
2844 #endif | |
2845 __ reset_last_Java_frame(); | |
2846 __ ldf(FloatRegisterImpl::D, saved_Freturn0_addr, Freturn0); | |
2847 | |
2848 // In tiered we never use C2 to compile methods returning longs so | |
2849 // the result is where we expect it already. | |
2850 | |
2851 #if !defined(_LP64) && defined(COMPILER2) | |
2852 // In 32 bit, C2 returns longs in G1 so restore the saved G1 into | |
2853 // I0/I1 if the return value is long. In the tiered world there is | |
2854 // a mismatch between how C1 and C2 return longs compiles and so | |
2855 // currently compilation of methods which return longs is disabled | |
2856 // for C2 and so is this code. Eventually C1 and C2 will do the | |
2857 // same thing for longs in the tiered world. | |
2858 if (!TieredCompilation) { | |
2859 Label not_long; | |
2860 __ cmp(O0,T_LONG); | |
2861 __ br(Assembler::notEqual, false, Assembler::pt, not_long); | |
2862 __ delayed()->nop(); | |
2863 __ ldd(saved_Greturn1_addr,I0); | |
2864 __ bind(not_long); | |
2865 } | |
2866 #endif | |
2867 __ ret(); | |
2868 __ delayed()->restore(); | |
2869 | |
2870 masm->flush(); | |
2871 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_words); | |
2872 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); | |
2873 } | |
2874 | |
2875 #ifdef COMPILER2 | |
2876 | |
2877 //------------------------------generate_uncommon_trap_blob-------------------- | |
2878 // Ought to generate an ideal graph & compile, but here's some SPARC ASM | |
2879 // instead. | |
2880 void SharedRuntime::generate_uncommon_trap_blob() { | |
2881 // allocate space for the code | |
2882 ResourceMark rm; | |
2883 // setup code generation tools | |
2884 int pad = VerifyThread ? 512 : 0; | |
2885 #ifdef _LP64 | |
2886 CodeBuffer buffer("uncommon_trap_blob", 2700+pad, 512); | |
2887 #else | |
2888 // Measured 8/7/03 at 660 in 32bit debug build (no VerifyThread) | |
2889 // Measured 8/7/03 at 1028 in 32bit debug build (VerifyThread) | |
2890 CodeBuffer buffer("uncommon_trap_blob", 2000+pad, 512); | |
2891 #endif | |
2892 MacroAssembler* masm = new MacroAssembler(&buffer); | |
2893 Register O2UnrollBlock = O2; | |
2894 Register O3tmp = O3; | |
2895 Register O2klass_index = O2; | |
2896 | |
2897 // | |
2898 // This is the entry point for all traps the compiler takes when it thinks | |
2899 // it cannot handle further execution of compilation code. The frame is | |
2900 // deoptimized in these cases and converted into interpreter frames for | |
2901 // execution | |
2902 // The steps taken by this frame are as follows: | |
2903 // - push a fake "unpack_frame" | |
2904 // - call the C routine Deoptimization::uncommon_trap (this function | |
2905 // packs the current compiled frame into vframe arrays and returns | |
2906 // information about the number and size of interpreter frames which | |
2907 // are equivalent to the frame which is being deoptimized) | |
2908 // - deallocate the "unpack_frame" | |
2909 // - deallocate the deoptimization frame | |
2910 // - in a loop using the information returned in the previous step | |
2911 // push interpreter frames; | |
2912 // - create a dummy "unpack_frame" | |
2913 // - call the C routine: Deoptimization::unpack_frames (this function | |
2914 // lays out values on the interpreter frame which was just created) | |
2915 // - deallocate the dummy unpack_frame | |
2916 // - return to the interpreter entry point | |
2917 // | |
2918 // Refer to the following methods for more information: | |
2919 // - Deoptimization::uncommon_trap | |
2920 // - Deoptimization::unpack_frame | |
2921 | |
2922 // the unloaded class index is in O0 (first parameter to this blob) | |
2923 | |
2924 // push a dummy "unpack_frame" | |
2925 // and call Deoptimization::uncommon_trap to pack the compiled frame into | |
2926 // vframe array and return the UnrollBlock information | |
2927 __ save_frame(0); | |
2928 __ set_last_Java_frame(SP, noreg); | |
2929 __ mov(I0, O2klass_index); | |
2930 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap), G2_thread, O2klass_index); | |
2931 __ reset_last_Java_frame(); | |
2932 __ mov(O0, O2UnrollBlock->after_save()); | |
2933 __ restore(); | |
2934 | |
2935 // deallocate the deoptimized frame taking care to preserve the return values | |
2936 __ mov(O2UnrollBlock, O2UnrollBlock->after_save()); | |
2937 __ restore(); | |
2938 | |
2939 // Allocate new interpreter frame(s) and possible c2i adapter frame | |
2940 | |
2941 make_new_frames(masm, false); | |
2942 | |
2943 // push a dummy "unpack_frame" taking care of float return values and | |
2944 // call Deoptimization::unpack_frames to have the unpacker layout | |
2945 // information in the interpreter frames just created and then return | |
2946 // to the interpreter entry point | |
2947 __ save_frame(0); | |
2948 __ set_last_Java_frame(SP, noreg); | |
2949 __ mov(Deoptimization::Unpack_uncommon_trap, O3); // indicate it is the uncommon trap case | |
2950 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames), G2_thread, O3); | |
2951 __ reset_last_Java_frame(); | |
2952 __ ret(); | |
2953 __ delayed()->restore(); | |
2954 | |
2955 masm->flush(); | |
2956 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, NULL, __ total_frame_size_in_bytes(0)/wordSize); | |
2957 } | |
2958 | |
2959 #endif // COMPILER2 | |
2960 | |
2961 //------------------------------generate_handler_blob------------------- | |
2962 // | |
2963 // Generate a special Compile2Runtime blob that saves all registers, and sets | |
2964 // up an OopMap. | |
2965 // | |
2966 // This blob is jumped to (via a breakpoint and the signal handler) from a | |
2967 // safepoint in compiled code. On entry to this blob, O7 contains the | |
2968 // address in the original nmethod at which we should resume normal execution. | |
2969 // Thus, this blob looks like a subroutine which must preserve lots of | |
2970 // registers and return normally. Note that O7 is never register-allocated, | |
2971 // so it is guaranteed to be free here. | |
2972 // | |
2973 | |
2974 // The hardest part of what this blob must do is to save the 64-bit %o | |
2975 // registers in the 32-bit build. A simple 'save' turn the %o's to %i's and | |
2976 // an interrupt will chop off their heads. Making space in the caller's frame | |
2977 // first will let us save the 64-bit %o's before save'ing, but we cannot hand | |
2978 // the adjusted FP off to the GC stack-crawler: this will modify the caller's | |
2979 // SP and mess up HIS OopMaps. So we first adjust the caller's SP, then save | |
2980 // the 64-bit %o's, then do a save, then fixup the caller's SP (our FP). | |
2981 // Tricky, tricky, tricky... | |
2982 | |
2983 static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) { | |
2984 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); | |
2985 | |
2986 // allocate space for the code | |
2987 ResourceMark rm; | |
2988 // setup code generation tools | |
2989 // Measured 8/7/03 at 896 in 32bit debug build (no VerifyThread) | |
2990 // Measured 8/7/03 at 1080 in 32bit debug build (VerifyThread) | |
2991 // even larger with TraceJumps | |
2992 int pad = TraceJumps ? 512 : 0; | |
2993 CodeBuffer buffer("handler_blob", 1600 + pad, 512); | |
2994 MacroAssembler* masm = new MacroAssembler(&buffer); | |
2995 int frame_size_words; | |
2996 OopMapSet *oop_maps = new OopMapSet(); | |
2997 OopMap* map = NULL; | |
2998 | |
2999 int start = __ offset(); | |
3000 | |
3001 // If this causes a return before the processing, then do a "restore" | |
3002 if (cause_return) { | |
3003 __ restore(); | |
3004 } else { | |
3005 // Make it look like we were called via the poll | |
3006 // so that frame constructor always sees a valid return address | |
3007 __ ld_ptr(G2_thread, in_bytes(JavaThread::saved_exception_pc_offset()), O7); | |
3008 __ sub(O7, frame::pc_return_offset, O7); | |
3009 } | |
3010 | |
3011 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_words); | |
3012 | |
3013 // setup last_Java_sp (blows G4) | |
3014 __ set_last_Java_frame(SP, noreg); | |
3015 | |
3016 // call into the runtime to handle illegal instructions exception | |
3017 // Do not use call_VM_leaf, because we need to make a GC map at this call site. | |
3018 __ mov(G2_thread, O0); | |
3019 __ save_thread(L7_thread_cache); | |
3020 __ call(call_ptr); | |
3021 __ delayed()->nop(); | |
3022 | |
3023 // Set an oopmap for the call site. | |
3024 // We need this not only for callee-saved registers, but also for volatile | |
3025 // registers that the compiler might be keeping live across a safepoint. | |
3026 | |
3027 oop_maps->add_gc_map( __ offset() - start, map); | |
3028 | |
3029 __ restore_thread(L7_thread_cache); | |
3030 // clear last_Java_sp | |
3031 __ reset_last_Java_frame(); | |
3032 | |
3033 // Check for exceptions | |
3034 Label pending; | |
3035 | |
3036 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), O1); | |
3037 __ tst(O1); | |
3038 __ brx(Assembler::notEqual, true, Assembler::pn, pending); | |
3039 __ delayed()->nop(); | |
3040 | |
3041 RegisterSaver::restore_live_registers(masm); | |
3042 | |
3043 // We are back the the original state on entry and ready to go. | |
3044 | |
3045 __ retl(); | |
3046 __ delayed()->nop(); | |
3047 | |
3048 // Pending exception after the safepoint | |
3049 | |
3050 __ bind(pending); | |
3051 | |
3052 RegisterSaver::restore_live_registers(masm); | |
3053 | |
3054 // We are back the the original state on entry. | |
3055 | |
3056 // Tail-call forward_exception_entry, with the issuing PC in O7, | |
3057 // so it looks like the original nmethod called forward_exception_entry. | |
3058 __ set((intptr_t)StubRoutines::forward_exception_entry(), O0); | |
3059 __ JMP(O0, 0); | |
3060 __ delayed()->nop(); | |
3061 | |
3062 // ------------- | |
3063 // make sure all code is generated | |
3064 masm->flush(); | |
3065 | |
3066 // return exception blob | |
3067 return SafepointBlob::create(&buffer, oop_maps, frame_size_words); | |
3068 } | |
3069 | |
3070 // | |
3071 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss | |
3072 // | |
3073 // Generate a stub that calls into vm to find out the proper destination | |
3074 // of a java call. All the argument registers are live at this point | |
3075 // but since this is generic code we don't know what they are and the caller | |
3076 // must do any gc of the args. | |
3077 // | |
3078 static RuntimeStub* generate_resolve_blob(address destination, const char* name) { | |
3079 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); | |
3080 | |
3081 // allocate space for the code | |
3082 ResourceMark rm; | |
3083 // setup code generation tools | |
3084 // Measured 8/7/03 at 896 in 32bit debug build (no VerifyThread) | |
3085 // Measured 8/7/03 at 1080 in 32bit debug build (VerifyThread) | |
3086 // even larger with TraceJumps | |
3087 int pad = TraceJumps ? 512 : 0; | |
3088 CodeBuffer buffer(name, 1600 + pad, 512); | |
3089 MacroAssembler* masm = new MacroAssembler(&buffer); | |
3090 int frame_size_words; | |
3091 OopMapSet *oop_maps = new OopMapSet(); | |
3092 OopMap* map = NULL; | |
3093 | |
3094 int start = __ offset(); | |
3095 | |
3096 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_words); | |
3097 | |
3098 int frame_complete = __ offset(); | |
3099 | |
3100 // setup last_Java_sp (blows G4) | |
3101 __ set_last_Java_frame(SP, noreg); | |
3102 | |
3103 // call into the runtime to handle illegal instructions exception | |
3104 // Do not use call_VM_leaf, because we need to make a GC map at this call site. | |
3105 __ mov(G2_thread, O0); | |
3106 __ save_thread(L7_thread_cache); | |
3107 __ call(destination, relocInfo::runtime_call_type); | |
3108 __ delayed()->nop(); | |
3109 | |
3110 // O0 contains the address we are going to jump to assuming no exception got installed | |
3111 | |
3112 // Set an oopmap for the call site. | |
3113 // We need this not only for callee-saved registers, but also for volatile | |
3114 // registers that the compiler might be keeping live across a safepoint. | |
3115 | |
3116 oop_maps->add_gc_map( __ offset() - start, map); | |
3117 | |
3118 __ restore_thread(L7_thread_cache); | |
3119 // clear last_Java_sp | |
3120 __ reset_last_Java_frame(); | |
3121 | |
3122 // Check for exceptions | |
3123 Label pending; | |
3124 | |
3125 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), O1); | |
3126 __ tst(O1); | |
3127 __ brx(Assembler::notEqual, true, Assembler::pn, pending); | |
3128 __ delayed()->nop(); | |
3129 | |
3130 // get the returned methodOop | |
3131 | |
3132 __ get_vm_result(G5_method); | |
3133 __ stx(G5_method, SP, RegisterSaver::G5_offset()+STACK_BIAS); | |
3134 | |
3135 // O0 is where we want to jump, overwrite G3 which is saved and scratch | |
3136 | |
3137 __ stx(O0, SP, RegisterSaver::G3_offset()+STACK_BIAS); | |
3138 | |
3139 RegisterSaver::restore_live_registers(masm); | |
3140 | |
3141 // We are back the the original state on entry and ready to go. | |
3142 | |
3143 __ JMP(G3, 0); | |
3144 __ delayed()->nop(); | |
3145 | |
3146 // Pending exception after the safepoint | |
3147 | |
3148 __ bind(pending); | |
3149 | |
3150 RegisterSaver::restore_live_registers(masm); | |
3151 | |
3152 // We are back the the original state on entry. | |
3153 | |
3154 // Tail-call forward_exception_entry, with the issuing PC in O7, | |
3155 // so it looks like the original nmethod called forward_exception_entry. | |
3156 __ set((intptr_t)StubRoutines::forward_exception_entry(), O0); | |
3157 __ JMP(O0, 0); | |
3158 __ delayed()->nop(); | |
3159 | |
3160 // ------------- | |
3161 // make sure all code is generated | |
3162 masm->flush(); | |
3163 | |
3164 // return the blob | |
3165 // frame_size_words or bytes?? | |
3166 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true); | |
3167 } | |
3168 | |
3169 void SharedRuntime::generate_stubs() { | |
3170 | |
3171 _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method), | |
3172 "wrong_method_stub"); | |
3173 | |
3174 _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss), | |
3175 "ic_miss_stub"); | |
3176 | |
3177 _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C), | |
3178 "resolve_opt_virtual_call"); | |
3179 | |
3180 _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C), | |
3181 "resolve_virtual_call"); | |
3182 | |
3183 _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C), | |
3184 "resolve_static_call"); | |
3185 | |
3186 _polling_page_safepoint_handler_blob = | |
3187 generate_handler_blob(CAST_FROM_FN_PTR(address, | |
3188 SafepointSynchronize::handle_polling_page_exception), false); | |
3189 | |
3190 _polling_page_return_handler_blob = | |
3191 generate_handler_blob(CAST_FROM_FN_PTR(address, | |
3192 SafepointSynchronize::handle_polling_page_exception), true); | |
3193 | |
3194 generate_deopt_blob(); | |
3195 | |
3196 #ifdef COMPILER2 | |
3197 generate_uncommon_trap_blob(); | |
3198 #endif // COMPILER2 | |
3199 } |