1 /*
2 * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2015, Linaro Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.hpp"
29 #include "asm/macroAssembler.inline.hpp"
30 #include "code/debugInfoRec.hpp"
31 #include "code/icBuffer.hpp"
32 #include "code/vtableStubs.hpp"
33 #include "interp_masm_aarch32.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "oops/compiledICHolder.hpp"
36 #include "prims/jvmtiRedefineClassesTrace.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/vframeArray.hpp"
39 #include "vmreg_aarch32.inline.hpp"
40 #ifdef COMPILER1
41 #include "c1/c1_Runtime1.hpp"
42 #endif
43 #ifdef COMPILER2
44 #include "adfiles/ad_aarch32.hpp"
45 #include "opto/runtime.hpp"
46 #endif
47
48
49 #define __ masm->
50
51 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
52
53 class SimpleRuntimeFrame {
54
55 public:
56
57 // Most of the runtime stubs have this simple frame layout.
58 // This class exists to make the layout shared in one place.
59 // Offsets are for compiler stack slots, which are jints.
60 enum layout {
61 // The frame sender code expects that rbp will be in the "natural" place and
62 // will override any oopMap setting for it. We must therefore force the layout
63 // so that it agrees with the frame sender code.
64 // we don't expect any arg reg save area so aarch32 asserts that
65 // frame::arg_reg_save_area_bytes == 0
66 rbp_off = 0,
67 rbp_off2,
68 return_off, return_off2,
69 framesize
70 };
71 };
72
73 // FIXME -- this is used by C1
74 class RegisterSaver {
75 public:
76 static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
77 static void restore_live_registers(MacroAssembler* masm);
78
79 // Capture info about frame layout
80 enum layout {
81 fpu_state_off = 0,
82 fpu_state_end = fpu_state_off+FPUStateSizeInWords-1,
83 // The frame sender code expects that rfp will be in
84 // the "natural" place and will override any oopMap
85 // setting for it. We must therefore force the layout
86 // so that it agrees with the frame sender code.
87 //
88 // FIXME there are extra saved register (from `push_CPU_state`) note that r11 == rfp
89 r0_off,
90 r1_off,
91 r2_off,
92 r3_off,
93 r4_off,
94 r5_off,
95 r6_off,
96 r7_off,
97 r8_off,
98 r9_off, rscratch1_off = r9_off,
99 r10_off, rmethod_off = r10_off,
100 r11_off,
101 r12_off,
102 reg_save_pad, // align area to 8-bytes to simplify stack alignment to 8
103 rfp_off,
104 return_off,
105 reg_save_size,
106 };
107
108
109 // Offsets into the register save area
110 // Used by deoptimization when it is managing result register
111 // values on its own
112
113 static int offset_in_bytes(int offset) { return offset * wordSize; }
114
115 // During deoptimization only the result registers need to be restored,
116 // all the other values have already been extracted.
117 static void restore_result_registers(MacroAssembler* masm);
118
119 };
120
121 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
122 int frame_size_in_bytes = additional_frame_words*wordSize + reg_save_size*BytesPerInt;
123 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
124 int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt;
125 *total_frame_words = frame_size_in_bytes / wordSize;;
126
127 __ enter();
128 __ push_CPU_state();
129
130 // Set an oopmap for the call site. This oopmap will map all
131 // oop-registers and debug-info registers as callee-saved. This
132 // will allow deoptimization at this safepoint to find all possible
133 // debug-info recordings, as well as let GC find all oops.
134
135 OopMapSet *oop_maps = new OopMapSet();
136 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
137
138 oop_map->set_callee_saved(VMRegImpl::stack2reg(r0_off + additional_frame_slots), r0->as_VMReg());
139 oop_map->set_callee_saved(VMRegImpl::stack2reg(r1_off + additional_frame_slots), r1->as_VMReg());
140 oop_map->set_callee_saved(VMRegImpl::stack2reg(r2_off + additional_frame_slots), r2->as_VMReg());
141 oop_map->set_callee_saved(VMRegImpl::stack2reg(r3_off + additional_frame_slots), r3->as_VMReg());
142 oop_map->set_callee_saved(VMRegImpl::stack2reg(r4_off + additional_frame_slots), r4->as_VMReg());
143 oop_map->set_callee_saved(VMRegImpl::stack2reg(r5_off + additional_frame_slots), r5->as_VMReg());
144 oop_map->set_callee_saved(VMRegImpl::stack2reg(r6_off + additional_frame_slots), r6->as_VMReg());
145 oop_map->set_callee_saved(VMRegImpl::stack2reg(r7_off + additional_frame_slots), r7->as_VMReg());
146 oop_map->set_callee_saved(VMRegImpl::stack2reg(r8_off + additional_frame_slots), r8->as_VMReg());
147 oop_map->set_callee_saved(VMRegImpl::stack2reg(r10_off + additional_frame_slots), r10->as_VMReg());
148 // r11 saved in frame header as rfp, not map it here
149 // r11 & r14 have special meaning (can't hold oop), so not map them
150
151 for (int i = 0; i < 31; ++i) {
152 oop_map->set_callee_saved(VMRegImpl::stack2reg(fpu_state_off + i + additional_frame_slots),
153 as_FloatRegister(i)->as_VMReg());
154 }
155
156 return oop_map;
157 }
158
159 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
160 __ pop_CPU_state();
161 __ leave();
162 }
163
164 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
165
166 // Just restore result register. Only used by deoptimization. By
167 // now any callee save register that needs to be restored to a c2
168 // caller of the deoptee has been extracted into the vframeArray
169 // and will be stuffed into the c2i adapter we create for later
170 // restoration so only result registers need to be restored here.
171
172
173
174 // Restore fp result register
175 __ vldr_f64(d0, Address(sp, offset_in_bytes(fpu_state_off)));
176 // Restore integer result register
177 __ ldr(r0, Address(sp, offset_in_bytes(r0_off)));
178 __ ldr(r1, Address(sp, offset_in_bytes(r1_off)));
179
180 // Pop all of the register save are off the stack
181 __ add(sp, sp, reg_save_size * wordSize);
182 }
183
184 // Is vector's size (in bytes) bigger than a size saved by default?
185 // 16 bytes XMM registers are saved by default using fxsave/fxrstor instructions.
186 bool SharedRuntime::is_wide_vector(int size) {
187 return size > 16;
188 }
189
190 // This functions returns offset from fp to java arguments on stack.
191 //
192 // The java_calling_convention describes stack locations as ideal slots on
193 // a frame with no abi restrictions. Since we must observe abi restrictions
194 // (like the placement of the register window) the slots must be biased by
195 // the following value.
196 static int reg2offset_in(VMReg r) {
197 // After stack frame created, fp points to 1 slot after previous sp value.
198 return (r->reg2stack() + 1) * VMRegImpl::stack_slot_size;
199 }
200
201 static int reg2offset_out(VMReg r) {
202 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
203 }
204
205 // ---------------------------------------------------------------------------
206 // Read the array of BasicTypes from a signature, and compute where the
207 // arguments should go. Values in the VMRegPair regs array refer to 4-byte
208 // quantities. Values less than VMRegImpl::stack0 are registers, those above
209 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer
210 // as framesizes are fixed.
211 // VMRegImpl::stack0 refers to the first slot 0(sp).
212 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register
213 // up to RegisterImpl::number_of_registers) are the 64-bit
214 // integer registers.
215
216 // Note: the INPUTS in sig_bt are in units of Java argument words,
217 // which are 64-bit. The OUTPUTS are in 32-bit units.
218
219 // The Java calling convention is a "shifted" version of the C ABI.
220 // By skipping the first C ABI register we can call non-static jni
221 // methods with small numbers of arguments without having to shuffle
222 // the arguments at all. Since we control the java ABI we ought to at
223 // least get some advantage out of it.
224
225 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
226 VMRegPair *regs,
227 int total_args_passed,
228 int is_outgoing) {
229
230 // Create the mapping between argument positions and
231 // registers.
232 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
233 j_rarg0, j_rarg1, j_rarg2, j_rarg3
234 };
235 const int FP_ArgReg_N = 16;
236 static const FloatRegister FP_ArgReg[] = {
237 f0, f1, f2, f3,
238 f4, f5, f6, f7,
239 f8, f9, f10, f11,
240 f12, f13, f14, f15,
241 };
242
243 uint int_args = 0;
244 uint fp_args = 0;
245 uint stk_args = 0;
246
247 for (int i = 0; i < total_args_passed; i++) {
248 switch (sig_bt[i]) {
249 case T_BOOLEAN:
250 case T_CHAR:
251 case T_BYTE:
252 case T_SHORT:
253 case T_INT:
254 case T_OBJECT:
255 case T_ARRAY:
256 case T_ADDRESS:
257 if (int_args < Argument::n_int_register_parameters_j) {
258 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
259 } else {
260 regs[i].set1(VMRegImpl::stack2reg(stk_args));
261 stk_args += 1;
262 }
263 break;
264 case T_VOID:
265 // halves of T_LONG or T_DOUBLE
266 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
267 regs[i].set_bad();
268 break;
269 case T_LONG:
270 assert(sig_bt[i + 1] == T_VOID, "expecting half");
271 if (int_args + 1 < Argument::n_int_register_parameters_j) {
272 regs[i].set_pair(INT_ArgReg[int_args + 1]->as_VMReg(), INT_ArgReg[int_args]->as_VMReg());
273 int_args += 2;
274 } else {
275 regs[i].set2(VMRegImpl::stack2reg(stk_args));
276 stk_args += 2;
277 }
278 break;
279 case T_FLOAT:
280 if (fp_args < FP_ArgReg_N) {
281 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
282 } else {
283 regs[i].set1(VMRegImpl::stack2reg(stk_args));
284 stk_args += 1;
285 }
286 break;
287 case T_DOUBLE:
288 assert(sig_bt[i + 1] == T_VOID, "expecting half");
289 fp_args = round_to(fp_args, 2);
290 if (fp_args < FP_ArgReg_N) {
291 regs[i].set2(FP_ArgReg[fp_args]->as_VMReg());
292 fp_args += 2;
293 } else {
294 regs[i].set2(VMRegImpl::stack2reg(stk_args));
295 stk_args += 2;
296 }
297 break;
298 default:
299 ShouldNotReachHere();
300 break;
301 }
302 }
303
304 return round_to(stk_args, StackAlignmentInBytes/wordSize);
305 }
306
307 // Patch the callers callsite with entry to compiled code if it exists.
308 static void patch_callers_callsite(MacroAssembler *masm) {
309 Label L;
310 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
311 __ cbz(rscratch1, L);
312
313 __ enter();
314 __ push_CPU_state();
315
316 // VM needs caller's callsite
317 // VM needs target method
318 // This needs to be a long call since we will relocate this adapter to
319 // the codeBuffer and it may not reach
320
321 #ifndef PRODUCT
322 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
323 #endif
324
325 __ mov(c_rarg0, rmethod);
326 __ mov(c_rarg1, lr);
327 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
328 __ bl(rscratch1);
329 __ maybe_isb();
330
331 __ pop_CPU_state();
332 // restore sp
333 __ leave();
334 __ bind(L);
335 }
336
337 static void gen_c2i_adapter(MacroAssembler *masm,
338 int total_args_passed,
339 int comp_args_on_stack,
340 const BasicType *sig_bt,
341 const VMRegPair *regs,
342 Label& skip_fixup) {
343 // Before we get into the guts of the C2I adapter, see if we should be here
344 // at all. We've come from compiled code and are attempting to jump to the
345 // interpreter, which means the caller made a static call to get here
346 // (vcalls always get a compiled target if there is one). Check for a
347 // compiled target. If there is one, we need to patch the caller's call.
348 patch_callers_callsite(masm);
349
350 __ bind(skip_fixup);
351
352 // Since all args are passed on the stack, total_args_passed *
353 // Interpreter::stackElementSize is the space we need.
354
355 const int extraspace = total_args_passed * Interpreter::stackElementSize;
356 const Register compArgPos = lr;
357 int ld_shift = 0;
358
359 __ str(compArgPos, Address(sp, -(extraspace + wordSize)));
360 __ mov(compArgPos, sp);
361
362 // Now write the args into the outgoing interpreter space
363 for (int i = 0; i < total_args_passed; i++) {
364
365 if (sig_bt[i] == T_VOID) {
366 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
367 continue;
368 }
369
370 // next stack slot offset
371 const int next_off = -Interpreter::stackElementSize;
372
373 VMReg r_1 = regs[i].first();
374 VMReg r_2 = regs[i].second();
375 if (!r_1->is_valid()) {
376 assert(!r_2->is_valid(), "");
377 continue;
378 }
379
380 if (r_2->is_valid()) {
381 assert(i + 1 < total_args_passed && sig_bt[i + 1] == T_VOID, "going to overrwrite reg_2 value");
382 }
383
384 if (r_1->is_stack()) {
385 // memory to memory use rscratch1
386 int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size - ld_shift;
387 if (!r_2->is_valid()) {
388 __ ldr(rscratch1, Address(compArgPos, ld_off));
389 __ str(rscratch1, Address(sp, next_off, Address::pre));
390 } else {
391 int tmp_off = ld_off;
392 // ldrd accepts only imm8
393 if(abs(ld_off) > (255 << 2)) {
394 if(__ is_valid_for_imm12(ld_off)) {
395 __ add(compArgPos, compArgPos, ld_off);
396 } else {
397 // add operates encoded imm12, NOT plain
398 __ mov(rscratch1, ld_off);
399 __ add(compArgPos, compArgPos, rscratch1);
400 }
401 tmp_off = 0;
402 ld_shift += ld_off;
403 }
404 __ ldrd(rscratch1, rscratch2, Address(compArgPos, tmp_off));
405 __ strd(rscratch1, rscratch2, Address(sp, 2* next_off, Address::pre));
406 }
407 } else if (r_1->is_Register()) {
408 Register r = r_1->as_Register();
409 assert(r != compArgPos, "compArgPos was modified");
410 if (!r_2->is_valid()) {
411 __ str(r, Address(sp, next_off, Address::pre));
412 } else {
413 assert(r_2->as_Register() != compArgPos, "compArgPos was modified");
414 __ strd(r, r_2->as_Register(), Address(sp, 2 * next_off, Address::pre));
415 }
416 } else {
417 assert(r_1->is_FloatRegister(), "");
418 if (!r_2->is_valid()) {
419 // Can't do pre or post addressing for vldr, vstr
420 __ add(sp, sp, next_off);
421 __ vstr_f32(r_1->as_FloatRegister(), Address(sp));
422 } else {
423 // TODO assert(r_2->is_FloatRegister() && r_2->as_FloatRegister() == r_1->as_FloatRegister() + 1, "");
424 // Can't do pre or post addressing for vldr, vstr
425 __ add(sp, sp, 2 * next_off);
426 __ vstr_f64(r_1->as_FloatRegister(), Address(sp));
427 }
428 }
429 }
430
431 // hope, sp is returned to desired value
432 __ ldr(compArgPos, Address(sp, -wordSize));
433
434 // set sender sp
435 if(__ is_valid_for_imm12(extraspace)) {
436 __ add(r4, sp, extraspace);
437 } else {
438 __ mov(rscratch1, extraspace);
439 __ add(r4, sp, rscratch1);
440 }
441
442 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
443 __ b(rscratch1);
444 }
445
446 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
447 address code_start, address code_end,
448 Label& L_ok) {
449 Label L_fail;
450 __ lea(temp_reg, ExternalAddress(code_start));
451 __ cmp(pc_reg, temp_reg);
452 __ b(L_fail, Assembler::LO);
453 __ lea(temp_reg, ExternalAddress(code_end));
454 __ cmp(pc_reg, temp_reg);
455 __ b(L_ok, Assembler::LO);
456 __ bind(L_fail);
457 }
458
459 static void gen_i2c_adapter(MacroAssembler *masm,
460 int total_args_passed,
461 int comp_args_on_stack,
462 const BasicType *sig_bt,
463 const VMRegPair *regs) {
464
465 // Note: r13 contains the senderSP on entry. We must preserve it since
466 // we may do a i2c -> c2i transition if we lose a race where compiled
467 // code goes non-entrant while we get args ready.
468
469 // In addition we use r13 to locate all the interpreter args because
470 // we must align the stack to 16 bytes.
471
472 // Adapters are frameless.
473
474 // An i2c adapter is frameless because the *caller* frame, which is
475 // interpreted, routinely repairs its own sp (from
476 // interpreter_frame_last_sp), even if a callee has modified the
477 // stack pointer. It also recalculates and aligns sp.
478
479 // A c2i adapter is frameless because the *callee* frame, which is
480 // interpreted, routinely repairs its caller's sp (from sender_sp,
481 // which is set up via the senderSP register).
482
483 // In other words, if *either* the caller or callee is interpreted, we can
484 // get the stack pointer repaired after a call.
485
486 // This is why c2i and i2c adapters cannot be indefinitely composed.
487 // In particular, if a c2i adapter were to somehow call an i2c adapter,
488 // both caller and callee would be compiled methods, and neither would
489 // clean up the stack pointer changes performed by the two adapters.
490 // If this happens, control eventually transfers back to the compiled
491 // caller, but with an uncorrected stack, causing delayed havoc.
492
493 if (VerifyAdapterCalls &&
494 (Interpreter::code() != NULL || StubRoutines::code1() != NULL)) {
495 // So, let's test for cascading c2i/i2c adapters right now.
496 // assert(Interpreter::contains($return_addr) ||
497 // StubRoutines::contains($return_addr),
498 // "i2c adapter must return to an interpreter frame");
499 __ block_comment("verify_i2c { ");
500 Label L_ok;
501 if (Interpreter::code() != NULL)
502 range_check(masm, lr, rscratch1,
503 Interpreter::code()->code_start(), Interpreter::code()->code_end(),
504 L_ok);
505 if (StubRoutines::code1() != NULL)
506 range_check(masm, lr, rscratch1,
507 StubRoutines::code1()->code_begin(), StubRoutines::code1()->code_end(),
508 L_ok);
509 if (StubRoutines::code2() != NULL)
510 range_check(masm, lr, rscratch1,
511 StubRoutines::code2()->code_begin(), StubRoutines::code2()->code_end(),
512 L_ok);
513 const char* msg = "i2c adapter must return to an interpreter frame";
514 __ block_comment(msg);
515 __ stop(msg);
516 __ bind(L_ok);
517 __ block_comment("} verify_i2ce ");
518 }
519
520 const int stack_space = round_to(comp_args_on_stack * VMRegImpl::stack_slot_size, StackAlignmentInBytes);
521 const int ld_high = total_args_passed *Interpreter::stackElementSize;
522 // Point to interpreter value (vs. tag)
523 const int next_off = -Interpreter::stackElementSize; // offset from ld ptr
524 const Register loadCounter = lr;
525
526 // Align sp to StackAlignmentInBytes so compiled frame starts always aligned
527 // This is required by APCS, so all native code depends on it. The compiled
528 // Java code is not required to follow this standard however doing so
529 // simplifies the code because allows to have fixed size for compiled frames
530 __ mov(rscratch2, sp);
531 __ align_stack();
532 if(total_args_passed) {
533 // put below reserved stack space, imm12 should be enough
534 __ str(loadCounter, Address(sp, -(stack_space + wordSize)));
535
536 if(__ is_valid_for_imm12(ld_high)) {
537 __ add(loadCounter, rscratch2, ld_high);
538 } else {
539 // add operates encoded imm12, we need plain
540 __ mov(rscratch1, ld_high);
541 __ add(loadCounter, rscratch2, rscratch1);
542 }
543 }
544
545 if(comp_args_on_stack) {
546 if(__ is_valid_for_imm12(stack_space)) {
547 __ sub(sp, sp, stack_space);
548 } else {
549 // add operates encoded imm12, we need plain
550 __ mov(rscratch1, stack_space);
551 __ sub(sp, sp, rscratch1);
552 }
553 }
554
555 // +------+ -> r4
556 // | 0 | \
557 // | 1 | \
558 // | 2 | - > Load in argument order going down.
559 // | x | /
560 // | N | /
561 // +------+ -> inital sp
562 // | pad | maybe 1 word to align the stack to 8 bytes
563 // | M | \
564 // | x | \
565 // | 2 | -> Load in argument order going up.
566 // | 1 | /
567 // | 0 | /
568 // +------+ ->
569
570
571 int sp_offset = 0;
572
573 // Now generate the shuffle code.
574 for (int i = 0; i < total_args_passed; i++) {
575
576 if (sig_bt[i] == T_VOID) {
577 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
578 continue;
579 }
580
581 // Pick up 0, 1 or 2 words from SP+offset.
582
583 //
584 //
585 //
586 VMReg r_1 = regs[i].first();
587 VMReg r_2 = regs[i].second();
588 if (!r_1->is_valid()) {
589 assert(!r_2->is_valid(), "");
590 continue;
591 }
592
593 if (r_2->is_valid()) {
594 assert(i + 1 < total_args_passed && sig_bt[i + 1] == T_VOID, "going to overrwrite reg_2 value");
595 }
596
597 if (r_1->is_stack()) {
598 // Convert stack slot to an SP offset
599 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size - sp_offset;
600
601 if (!r_2->is_valid()) {
602 __ ldr(rscratch2, Address(loadCounter, next_off, Address::pre));
603 __ str(rscratch2, Address(sp, st_off));
604 } else {
605 int tmp_off = st_off;
606 if(abs(st_off) > (255 << 2)) {
607 //st_off doesn't fit imm8 required by strd
608
609 if(__ is_valid_for_imm12(st_off)) {
610 __ add(sp, sp, st_off);
611 } else {
612 // add operates encoded imm12, NOT plain
613 __ mov(rscratch1, st_off);
614 __ add(sp, sp, rscratch1);
615 }
616 tmp_off = 0;
617 sp_offset += st_off;
618 }
619
620
621 // Interpreter local[n] == MSW, local[n+1] == LSW however locals
622 // are accessed as negative so LSW is at LOW address
623
624 // this can be a misaligned move
625 __ ldrd(rscratch1, rscratch2, Address(loadCounter, 2 * next_off, Address::pre));
626 __ strd(rscratch1, rscratch2, Address(sp, tmp_off));
627 }
628 } else if (r_1->is_Register()) { // Register argument
629 Register r = r_1->as_Register();
630 assert(r != loadCounter, "loadCounter is reloaded");
631 if (r_2->is_valid()) {
632 assert(r_2->as_Register() != loadCounter, "loadCounter is reloaded");
633 // this can be a misaligned move
634 // ldrd can handle inconsecutive registers
635 __ ldrd(r, r_2->as_Register(), Address(loadCounter, 2 * next_off, Address::pre));
636 } else {
637 __ ldr(r, Address(loadCounter, next_off, Address::pre));
638 }
639 } else {
640 assert(r_1->is_FloatRegister(), "");
641 if (!r_2->is_valid()) {
642 // Can't do pre or post addressing for vldr, vstr
643 __ add(loadCounter, loadCounter, next_off);
644 __ vldr_f32(r_1->as_FloatRegister(), Address(loadCounter));
645 } else {
646 // TODO assert(r_2->is_FloatRegister() && r_2->as_FloatRegister() == r_1->as_FloatRegister() + 1, "");
647 // Can't do pre or post addressing for vldr, vstr
648 __ add(loadCounter, loadCounter, 2 * next_off);
649 __ vldr_f64(r_1->as_FloatRegister(), Address(loadCounter));
650 }
651 }
652 }
653
654 // restore sp
655 if(sp_offset) {
656 if(__ is_valid_for_imm12(sp_offset)) {
657 __ sub(sp, sp, sp_offset);
658 } else {
659 // add operates encoded imm12, we need plain
660 __ mov(rscratch1, sp_offset);
661 __ sub(sp, sp, rscratch1);
662 }
663 }
664
665 if(total_args_passed) {
666 // restore loadCounter
667 __ ldr(loadCounter, Address(sp, -wordSize));
668 }
669
670 // 6243940 We might end up in handle_wrong_method if
671 // the callee is deoptimized as we race thru here. If that
672 // happens we don't want to take a safepoint because the
673 // caller frame will look interpreted and arguments are now
674 // "compiled" so it is much better to make this transition
675 // invisible to the stack walking code. Unfortunately if
676 // we try and find the callee by normal means a safepoint
677 // is possible. So we stash the desired callee in the thread
678 // and the vm will find there should this case occur.
679
680 __ str(rmethod, Address(rthread, JavaThread::callee_target_offset()));
681
682 // Will jump to the compiled code just as if compiled code was doing it.
683 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::from_compiled_offset())));
684 __ b(rscratch1);
685 }
686
687 // ---------------------------------------------------------------
688 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
689 int total_args_passed,
690 int comp_args_on_stack,
691 const BasicType *sig_bt,
692 const VMRegPair *regs,
693 AdapterFingerPrint* fingerprint) {
694 address i2c_entry = __ pc();
695 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
696
697 address c2i_unverified_entry = __ pc();
698 Label skip_fixup;
699
700 Label ok;
701
702 Register holder = rscratch2;
703 Register receiver = j_rarg0;
704 Register tmp = r10; // A call-clobbered register not used for arg passing
705
706 // -------------------------------------------------------------------------
707 // Generate a C2I adapter. On entry we know rmethod holds the Method* during calls
708 // to the interpreter. The args start out packed in the compiled layout. They
709 // need to be unpacked into the interpreter layout. This will almost always
710 // require some stack space. We grow the current (compiled) stack, then repack
711 // the args. We finally end in a jump to the generic interpreter entry point.
712 // On exit from the interpreter, the interpreter will restore our SP (lest the
713 // compiled code, which relys solely on SP and not FP, get sick).
714
715 {
716 __ block_comment("c2i_unverified_entry {");
717 __ load_klass(rscratch1, receiver);
718 __ ldr(tmp, Address(holder, CompiledICHolder::holder_klass_offset()));
719 __ cmp(rscratch1, tmp);
720 __ ldr(rmethod, Address(holder, CompiledICHolder::holder_method_offset()));
721 __ b(ok, Assembler::EQ);
722 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
723
724 __ bind(ok);
725 // Method might have been compiled since the call site was patched to
726 // interpreted; if that is the case treat it as a miss so we can get
727 // the call site corrected.
728 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
729 __ cbz(rscratch1, skip_fixup);
730 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
731 __ block_comment("} c2i_unverified_entry");
732 }
733
734 address c2i_entry = __ pc();
735
736 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
737
738 __ flush();
739 return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry);
740 }
741
742 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
743 VMRegPair *regs,
744 VMRegPair *regs2,
745 int total_args_passed) {
746 assert(regs2 == NULL, "not needed on AArch32");
747
748 // We return the amount of VMRegImpl stack slots we need to reserve for all
749 // the arguments NOT counting out_preserve_stack_slots.
750
751 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
752 c_rarg0, c_rarg1, c_rarg2, c_rarg3
753 };
754 const int FP_ArgReg_N = 16;
755 static const FloatRegister FP_ArgReg[] = {
756 f0, f1, f2, f3,
757 f4, f5, f6, f7,
758 f8, f9, f10, f11,
759 f12, f13, f14, f15,
760 };
761 unsigned long fp_free_mask = (1 << FP_ArgReg_N) - 1;
762
763 uint int_args = 0;
764 uint fp_args = 0;
765 uint stk_args = 0;
766
767 for (int i = 0; i < total_args_passed; i++) {
768 switch (sig_bt[i]) {
769 case T_BOOLEAN:
770 case T_CHAR:
771 case T_BYTE:
772 case T_SHORT:
773 case T_INT:
774 case T_OBJECT:
775 case T_ARRAY:
776 case T_ADDRESS:
777 case T_METADATA:
778 if (int_args < Argument::n_int_register_parameters_c) {
779 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
780 } else {
781 regs[i].set1(VMRegImpl::stack2reg(stk_args));
782 stk_args += 1;
783 }
784 break;
785 case T_LONG:
786 assert(sig_bt[i + 1] == T_VOID, "expecting half");
787 if (int_args + 1 < Argument::n_int_register_parameters_c) {
788 if ((int_args % 2) != 0) {
789 ++int_args;
790 }
791 regs[i].set2(INT_ArgReg[int_args]->as_VMReg());
792 int_args += 2;
793 } else {
794 if (stk_args % 2 != 0) {
795 ++stk_args;
796 }
797 regs[i].set2(VMRegImpl::stack2reg(stk_args));
798 stk_args += 2;
799 int_args = Argument::n_int_register_parameters_c;
800 }
801 break;
802 case T_FLOAT:
803 if (fp_free_mask & ((1 << FP_ArgReg_N)-1)) {
804 unsigned index = __builtin_ctz(fp_free_mask);
805 regs[i].set1(FP_ArgReg[index]->as_VMReg());
806 fp_free_mask &= ~(1 << index);
807 fp_args += 2 * ((~index) & 1);
808 } else {
809 regs[i].set1(VMRegImpl::stack2reg(stk_args));
810 stk_args += 1;
811 }
812 break;
813 case T_DOUBLE:
814 assert(sig_bt[i + 1] == T_VOID, "expecting half");
815 if (fp_args + 1 < FP_ArgReg_N) {
816 fp_free_mask &= ~(3 << fp_args);
817 regs[i].set2(FP_ArgReg[fp_args]->as_VMReg());
818 fp_args += 2;
819 } else {
820 regs[i].set2(VMRegImpl::stack2reg(stk_args));
821 stk_args += 2;
822 }
823 break;
824 case T_VOID: // Halves of longs and doubles
825 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
826 regs[i].set_bad();
827 break;
828 default:
829 ShouldNotReachHere();
830 break;
831 }
832 }
833
834 return round_to(stk_args, StackAlignmentInBytes/wordSize);
835 }
836
837 // On 64 bit we will store integer like items to the stack as
838 // 64 bits items (sparc abi) even though java would only store
839 // 32bits for a parameter. On 32bit it will simply be 32 bits
840 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
841
842 static void move_int(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
843 if (src.first()->is_stack()) {
844 if (dst.first()->is_stack()) {
845 // stack to stack
846 __ ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
847 __ str(rscratch1, Address(sp, reg2offset_out(dst.first())));
848 } else {
849 // stack to reg
850 __ ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
851 }
852 } else if (dst.first()->is_stack()) {
853 // reg to stack
854 __ str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
855 } else {
856 if (dst.first() != src.first()) {
857 __ mov(dst.first()->as_Register(), src.first()->as_Register());
858 }
859 }
860 }
861
862 // An oop arg. Must pass a handle not the oop itself
863 static void object_move(MacroAssembler* masm,
864 OopMap* map,
865 int oop_handle_offset,
866 int framesize_in_slots,
867 VMRegPair src,
868 VMRegPair dst,
869 bool is_receiver,
870 int* receiver_offset) {
871
872 // must pass a handle. First figure out the location we use as a handle
873
874 Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
875
876 // See if oop is NULL if it is we need no handle
877
878 if (src.first()->is_stack()) {
879
880 // Oop is already on the stack as an argument
881 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
882 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
883 if (is_receiver) {
884 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
885 }
886
887 __ ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
888 __ lea(rHandle, Address(rfp, reg2offset_in(src.first())));
889 // conditionally move a NULL
890 __ cmp(rscratch1, 0);
891 __ mov(rHandle, 0, Assembler::EQ);
892 } else {
893
894 // Oop is in an a register we must store it to the space we reserve
895 // on the stack for oop_handles and pass a handle if oop is non-NULL
896
897 const Register rOop = src.first()->as_Register();
898 int oop_slot;
899 if (rOop == j_rarg0)
900 oop_slot = 0;
901 else if (rOop == j_rarg1)
902 oop_slot = 1;
903 else if (rOop == j_rarg2)
904 oop_slot = 2;
905 else {
906 assert(rOop == j_rarg3, "wrong register");
907 oop_slot = 3;
908 }
909
910 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
911 int offset = oop_slot*VMRegImpl::stack_slot_size;
912
913 map->set_oop(VMRegImpl::stack2reg(oop_slot));
914 // Store oop in handle area, may be NULL
915 __ str(rOop, Address(sp, offset));
916 if (is_receiver) {
917 *receiver_offset = offset;
918 }
919
920 __ cmp(rOop, 0);
921 __ lea(rHandle, Address(sp, offset));
922 // conditionally move a NULL
923 __ mov(rHandle, 0, Assembler::EQ);
924 }
925
926 // If arg is on the stack then place it otherwise it is already in correct reg.
927 if (dst.first()->is_stack()) {
928 __ str(rHandle, Address(sp, reg2offset_out(dst.first())));
929 }
930 }
931
932 // A float arg may have to do float reg int reg conversion
933 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
934 if (src.first()->is_stack()) {
935 if (dst.first()->is_stack()) {
936 // stack to stack
937 // Have no vfp scratch registers, so copy via gpr
938 __ ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
939 __ str(rscratch1, Address(sp, reg2offset_out(dst.first())));
940 } else {
941 // stack to reg
942 __ vldr_f32(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
943 }
944 } else if (dst.first()->is_stack()) {
945 // reg to stack
946 __ vstr_f32(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
947 } else {
948 if (dst.first() != src.first()) {
949 __ vmov_f32(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
950 }
951 }
952 }
953
954 // A long move
955 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
956 if (src.first()->is_stack()) {
957 if (dst.first()->is_stack()) {
958 // stack to stack
959 __ ldrd(rscratch1, rscratch2, Address(rfp, reg2offset_in(src.first())));
960 __ strd(rscratch1, rscratch2, Address(sp, reg2offset_out(dst.first())));
961 } else {
962 // stack to reg
963 __ ldrd(dst.first()->as_Register(), dst.second()->as_Register(),
964 Address(rfp, reg2offset_in(src.first())));
965 }
966 } else if (dst.first()->is_stack()) {
967 // reg to stack
968 __ strd(src.first()->as_Register(), src.second()->as_Register(),
969 Address(sp, reg2offset_out(dst.first())));
970 } else {
971 // reg to reg
972 if (dst.first() != src.first()) {
973 if (dst.first() != src.second()) {
974 __ mov(dst.first()->as_Register(), src.first()->as_Register());
975 __ mov(dst.second()->as_Register(), src.second()->as_Register());
976 } else {
977 __ mov(dst.second()->as_Register(), src.second()->as_Register());
978 __ mov(dst.first()->as_Register(), src.first()->as_Register());
979 }
980 }
981 }
982 }
983
984 // A double move
985 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
986 if (src.first()->is_stack()) {
987 if (dst.first()->is_stack()) {
988 // stack to stack
989 // Have no vfp scratch registers, so copy via gpr
990 __ ldrd(rscratch1, rscratch2, Address(rfp, reg2offset_in(src.first())));
991 __ strd(rscratch1, rscratch2, Address(sp, reg2offset_out(dst.first())));
992 } else {
993 // stack to reg
994 __ vldr_f64(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
995 }
996 } else if (dst.first()->is_stack()) {
997 // reg to stack
998 __ vstr_f64(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
999 } else {
1000 if (dst.first() != src.first()) {
1001 __ vmov_f64(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
1002 }
1003 }
1004 }
1005
1006
1007 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1008 // We always ignore the frame_slots arg and just use the space just below frame pointer
1009 // which by this time is free to use
1010 switch (ret_type) {
1011 case T_FLOAT:
1012 __ vstr_f32(f0, Address(rfp, -2 * wordSize));
1013 break;
1014 case T_DOUBLE:
1015 __ vstr_f64(d0, Address(rfp, -3 * wordSize));
1016 break;
1017 case T_LONG:
1018 __ strd(r0, r1, Address(rfp, -3 * wordSize));
1019 break;
1020 case T_VOID:
1021 break;
1022 default:
1023 __ str(r0, Address(rfp, -2 * wordSize));
1024 break;
1025 }
1026 }
1027
1028 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1029 // We always ignore the frame_slots arg and just use the space just below frame pointer
1030 // which by this time is free to use
1031 switch (ret_type) {
1032 case T_FLOAT:
1033 __ vldr_f32(d0, Address(rfp, -2 * wordSize));
1034 break;
1035 case T_DOUBLE:
1036 __ vldr_f64(d0, Address(rfp, -3 * wordSize));
1037 break;
1038 case T_LONG:
1039 __ ldrd(r0, r1, Address(rfp, -3 * wordSize));
1040 break;
1041 case T_VOID:
1042 break;
1043 default:
1044 __ ldr(r0, Address(rfp, -2 * wordSize));
1045 break;
1046 }
1047 }
1048
1049 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1050 RegSet x;
1051 for ( int i = first_arg ; i < arg_count ; i++ ) {
1052 if (args[i].first()->is_Register()) {
1053 x = x + args[i].first()->as_Register();
1054 } else if (args[i].first()->is_FloatRegister()) {
1055 FloatRegister fr = args[i].first()->as_FloatRegister();
1056
1057 if (args[i].second()->is_FloatRegister()) {
1058 assert(args[i].is_single_phys_reg(), "doubles should be 2 consequents float regs");
1059 __ decrement(sp, 2 * wordSize);
1060 __ vstr_f64(fr, Address(sp));
1061 } else {
1062 __ decrement(sp, wordSize);
1063 __ vstr_f32(fr, Address(sp));
1064 }
1065 }
1066 }
1067 __ push(x, sp);
1068 }
1069
1070 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1071 RegSet x;
1072 for ( int i = first_arg ; i < arg_count ; i++ ) {
1073 if (args[i].first()->is_Register()) {
1074 x = x + args[i].first()->as_Register();
1075 } else {
1076 ;
1077 }
1078 }
1079 __ pop(x, sp);
1080 for ( int i = first_arg ; i < arg_count ; i++ ) {
1081 if (args[i].first()->is_Register()) {
1082 ;
1083 } else if (args[i].first()->is_FloatRegister()) {
1084 FloatRegister fr = args[i].first()->as_FloatRegister();
1085
1086 if (args[i].second()->is_FloatRegister()) {
1087 assert(args[i].is_single_phys_reg(), "doubles should be 2 consequents float regs");
1088 __ vstr_f64(fr, Address(sp));
1089 __ increment(sp, 2 * wordSize);
1090 } else {
1091 __ vstr_f32(fr, Address(sp));
1092 __ increment(sp, wordSize);
1093 }
1094 }
1095 }
1096 }
1097
1098
1099 // Check GC_locker::needs_gc and enter the runtime if it's true. This
1100 // keeps a new JNI critical region from starting until a GC has been
1101 // forced. Save down any oops in registers and describe them in an
1102 // OopMap.
1103 static void check_needs_gc_for_critical_native(MacroAssembler* masm,
1104 int stack_slots,
1105 int total_c_args,
1106 int total_in_args,
1107 int arg_save_area,
1108 OopMapSet* oop_maps,
1109 VMRegPair* in_regs,
1110 BasicType* in_sig_bt) { Unimplemented(); }
1111
1112 // Unpack an array argument into a pointer to the body and the length
1113 // if the array is non-null, otherwise pass 0 for both.
1114 static void unpack_array_argument(MacroAssembler* masm, VMRegPair reg, BasicType in_elem_type, VMRegPair body_arg, VMRegPair length_arg) { Unimplemented(); }
1115
1116
1117 class ComputeMoveOrder: public StackObj {
1118 class MoveOperation: public ResourceObj {
1119 friend class ComputeMoveOrder;
1120 private:
1121 VMRegPair _src;
1122 VMRegPair _dst;
1123 int _src_index;
1124 int _dst_index;
1125 bool _processed;
1126 MoveOperation* _next;
1127 MoveOperation* _prev;
1128
1129 static int get_id(VMRegPair r) { Unimplemented(); return 0; }
1130
1131 public:
1132 MoveOperation(int src_index, VMRegPair src, int dst_index, VMRegPair dst):
1133 _src(src)
1134 , _src_index(src_index)
1135 , _dst(dst)
1136 , _dst_index(dst_index)
1137 , _next(NULL)
1138 , _prev(NULL)
1139 , _processed(false) { Unimplemented(); }
1140
1141 VMRegPair src() const { Unimplemented(); return _src; }
1142 int src_id() const { Unimplemented(); return 0; }
1143 int src_index() const { Unimplemented(); return 0; }
1144 VMRegPair dst() const { Unimplemented(); return _src; }
1145 void set_dst(int i, VMRegPair dst) { Unimplemented(); }
1146 int dst_index() const { Unimplemented(); return 0; }
1147 int dst_id() const { Unimplemented(); return 0; }
1148 MoveOperation* next() const { Unimplemented(); return 0; }
1149 MoveOperation* prev() const { Unimplemented(); return 0; }
1150 void set_processed() { Unimplemented(); }
1151 bool is_processed() const { Unimplemented(); return 0; }
1152
1153 // insert
1154 void break_cycle(VMRegPair temp_register) { Unimplemented(); }
1155
1156 void link(GrowableArray<MoveOperation*>& killer) { Unimplemented(); }
1157 };
1158
1159 private:
1160 GrowableArray<MoveOperation*> edges;
1161
1162 public:
1163 ComputeMoveOrder(int total_in_args, VMRegPair* in_regs, int total_c_args, VMRegPair* out_regs,
1164 BasicType* in_sig_bt, GrowableArray<int>& arg_order, VMRegPair tmp_vmreg) { Unimplemented(); }
1165
1166 // Collected all the move operations
1167 void add_edge(int src_index, VMRegPair src, int dst_index, VMRegPair dst) { Unimplemented(); }
1168
1169 // Walk the edges breaking cycles between moves. The result list
1170 // can be walked in order to produce the proper set of loads
1171 GrowableArray<MoveOperation*>* get_store_order(VMRegPair temp_register) { Unimplemented(); return 0; }
1172 };
1173
1174
1175 static void rt_call(MacroAssembler* masm, address dest) {
1176 CodeBlob *cb = CodeCache::find_blob(dest);
1177 if (cb) {
1178 __ far_call(RuntimeAddress(dest), NULL, rscratch2);
1179 } else {
1180 __ lea(rscratch2, RuntimeAddress(dest));
1181 __ bl(rscratch2);
1182 __ maybe_isb();
1183 }
1184 }
1185
1186 static void verify_oop_args(MacroAssembler* masm,
1187 methodHandle method,
1188 const BasicType* sig_bt,
1189 const VMRegPair* regs) {
1190 Register temp_reg = rscratch2; // not part of any compiled calling seq
1191 if (VerifyOops) {
1192 for (int i = 0; i < method->size_of_parameters(); i++) {
1193 if (sig_bt[i] == T_OBJECT ||
1194 sig_bt[i] == T_ARRAY) {
1195 VMReg r = regs[i].first();
1196 assert(r->is_valid(), "bad oop arg");
1197 if (r->is_stack()) {
1198 __ ldr(temp_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1199 __ verify_oop(temp_reg);
1200 } else {
1201 __ verify_oop(r->as_Register());
1202 }
1203 }
1204 }
1205 }
1206 }
1207
1208 static void gen_special_dispatch(MacroAssembler* masm,
1209 methodHandle method,
1210 const BasicType* sig_bt,
1211 const VMRegPair* regs) {
1212 verify_oop_args(masm, method, sig_bt, regs);
1213 vmIntrinsics::ID iid = method->intrinsic_id();
1214
1215 // Now write the args into the outgoing interpreter space
1216 bool has_receiver = false;
1217 Register receiver_reg = noreg;
1218 int member_arg_pos = -1;
1219 Register member_reg = noreg;
1220 int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1221 if (ref_kind != 0) {
1222 member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument
1223 member_reg = r4;
1224 has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1225 } else if (iid == vmIntrinsics::_invokeBasic) {
1226 has_receiver = true;
1227 } else {
1228 fatal(err_msg_res("unexpected intrinsic id %d", iid));
1229 }
1230
1231 if (member_reg != noreg) {
1232 // Load the member_arg into register, if necessary.
1233 SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1234 VMReg r = regs[member_arg_pos].first();
1235 if (r->is_stack()) {
1236 __ ldr(member_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1237 } else {
1238 // no data motion is needed
1239 member_reg = r->as_Register();
1240 }
1241 }
1242
1243 if (has_receiver) {
1244 // Make sure the receiver is loaded into a register.
1245 assert(method->size_of_parameters() > 0, "oob");
1246 assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1247 VMReg r = regs[0].first();
1248 assert(r->is_valid(), "bad receiver arg");
1249 if (r->is_stack()) {
1250 // Porting note: This assumes that compiled calling conventions always
1251 // pass the receiver oop in a register. If this is not true on some
1252 // platform, pick a temp and load the receiver from stack.
1253 fatal("receiver always in a register");
1254 } else {
1255 // no data motion is needed
1256 receiver_reg = r->as_Register();
1257 }
1258 }
1259
1260 // Figure out which address we are really jumping to:
1261 MethodHandles::generate_method_handle_dispatch(masm, iid,
1262 receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1263 }
1264
1265 // ---------------------------------------------------------------------------
1266 // Generate a native wrapper for a given method. The method takes arguments
1267 // in the Java compiled code convention, marshals them to the native
1268 // convention (handlizes oops, etc), transitions to native, makes the call,
1269 // returns to java state (possibly blocking), unhandlizes any result and
1270 // returns.
1271 //
1272 // Critical native functions are a shorthand for the use of
1273 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1274 // functions. The wrapper is expected to unpack the arguments before
1275 // passing them to the callee and perform checks before and after the
1276 // native call to ensure that they GC_locker
1277 // lock_critical/unlock_critical semantics are followed. Some other
1278 // parts of JNI setup are skipped like the tear down of the JNI handle
1279 // block and the check for pending exceptions it's impossible for them
1280 // to be thrown.
1281 //
1282 // They are roughly structured like this:
1283 // if (GC_locker::needs_gc())
1284 // SharedRuntime::block_for_jni_critical();
1285 // tranistion to thread_in_native
1286 // unpack arrray arguments and call native entry point
1287 // check for safepoint in progress
1288 // check if any thread suspend flags are set
1289 // call into JVM and possible unlock the JNI critical
1290 // if a GC was suppressed while in the critical native.
1291 // transition back to thread_in_Java
1292 // return to caller
1293 //
1294 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1295 methodHandle method,
1296 int compile_id,
1297 BasicType* in_sig_bt,
1298 VMRegPair* in_regs,
1299 BasicType ret_type) {
1300 if (method->is_method_handle_intrinsic()) {
1301 vmIntrinsics::ID iid = method->intrinsic_id();
1302 intptr_t start = (intptr_t)__ pc();
1303 int vep_offset = ((intptr_t)__ pc()) - start;
1304
1305 // First instruction must be a nop as it may need to be patched on deoptimisation
1306 __ nop();
1307 gen_special_dispatch(masm,
1308 method,
1309 in_sig_bt,
1310 in_regs);
1311 int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period
1312 __ flush();
1313 int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually
1314 return nmethod::new_native_nmethod(method,
1315 compile_id,
1316 masm->code(),
1317 vep_offset,
1318 frame_complete,
1319 stack_slots / VMRegImpl::slots_per_word,
1320 in_ByteSize(-1),
1321 in_ByteSize(-1),
1322 (OopMapSet*)NULL);
1323 }
1324
1325 bool is_critical_native = true;
1326 address native_func = method->critical_native_function();
1327 if (native_func == NULL) {
1328 native_func = method->native_function();
1329 is_critical_native = false;
1330 }
1331 assert(native_func != NULL, "must have function");
1332
1333 // An OopMap for lock (and class if static)
1334 OopMapSet *oop_maps = new OopMapSet();
1335 intptr_t start = (intptr_t)__ pc();
1336
1337 // We have received a description of where all the java arg are located
1338 // on entry to the wrapper. We need to convert these args to where
1339 // the jni function will expect them. To figure out where they go
1340 // we convert the java signature to a C signature by inserting
1341 // the hidden arguments as arg[0] and possibly arg[1] (static method)
1342
1343 const int total_in_args = method->size_of_parameters();
1344 int total_c_args = total_in_args;
1345 if (!is_critical_native) {
1346 total_c_args += 1;
1347 if (method->is_static()) {
1348 total_c_args++;
1349 }
1350 } else {
1351 for (int i = 0; i < total_in_args; i++) {
1352 if (in_sig_bt[i] == T_ARRAY) {
1353 total_c_args++;
1354 }
1355 }
1356 }
1357
1358 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1359 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1360 BasicType* in_elem_bt = NULL;
1361
1362 int argc = 0;
1363 if (!is_critical_native) {
1364 out_sig_bt[argc++] = T_ADDRESS;
1365 if (method->is_static()) {
1366 out_sig_bt[argc++] = T_OBJECT;
1367 }
1368
1369 for (int i = 0; i < total_in_args ; i++ ) {
1370 out_sig_bt[argc++] = in_sig_bt[i];
1371 }
1372 } else {
1373 Thread* THREAD = Thread::current();
1374 in_elem_bt = NEW_RESOURCE_ARRAY(BasicType, total_in_args);
1375 SignatureStream ss(method->signature());
1376 for (int i = 0; i < total_in_args ; i++ ) {
1377 if (in_sig_bt[i] == T_ARRAY) {
1378 // Arrays are passed as int, elem* pair
1379 out_sig_bt[argc++] = T_INT;
1380 out_sig_bt[argc++] = T_ADDRESS;
1381 Symbol* atype = ss.as_symbol(CHECK_NULL);
1382 const char* at = atype->as_C_string();
1383 if (strlen(at) == 2) {
1384 assert(at[0] == '[', "must be");
1385 switch (at[1]) {
1386 case 'B': in_elem_bt[i] = T_BYTE; break;
1387 case 'C': in_elem_bt[i] = T_CHAR; break;
1388 case 'D': in_elem_bt[i] = T_DOUBLE; break;
1389 case 'F': in_elem_bt[i] = T_FLOAT; break;
1390 case 'I': in_elem_bt[i] = T_INT; break;
1391 case 'J': in_elem_bt[i] = T_LONG; break;
1392 case 'S': in_elem_bt[i] = T_SHORT; break;
1393 case 'Z': in_elem_bt[i] = T_BOOLEAN; break;
1394 default: ShouldNotReachHere();
1395 }
1396 }
1397 } else {
1398 out_sig_bt[argc++] = in_sig_bt[i];
1399 in_elem_bt[i] = T_VOID;
1400 }
1401 if (in_sig_bt[i] != T_VOID) {
1402 assert(in_sig_bt[i] == ss.type(), "must match");
1403 ss.next();
1404 }
1405 }
1406 }
1407
1408 // Now figure out where the args must be stored and how much stack space
1409 // they require.
1410 int out_arg_slots;
1411 out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args);
1412
1413 // Compute framesize for the wrapper. We need to handlize all oops in
1414 // incoming registers
1415
1416 // Calculate the total number of stack slots we will need.
1417
1418 // First count the abi requirement plus all of the outgoing args
1419 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1420
1421 // Now the space for the inbound oop handle area
1422 int total_save_slots = -1;
1423 if (is_critical_native) {
1424 // Critical natives may have to call out so they need a save area
1425 // for register arguments.
1426 int double_slots = 0;
1427 int single_slots = 0;
1428 for ( int i = 0; i < total_in_args; i++) {
1429 if (in_regs[i].first()->is_Register()) {
1430 const Register reg = in_regs[i].first()->as_Register();
1431 switch (in_sig_bt[i]) {
1432 case T_ARRAY: // critical array (uses 2 slots on LP64)
1433 case T_BOOLEAN:
1434 case T_BYTE:
1435 case T_SHORT:
1436 case T_CHAR:
1437 case T_INT: single_slots++; break;
1438 case T_LONG: double_slots++; break;
1439 default: ShouldNotReachHere();
1440 }
1441 } else if (in_regs[i].first()->is_FloatRegister()) {
1442 ShouldNotReachHere();
1443 }
1444 }
1445 total_save_slots = double_slots * 2 + single_slots;
1446 // align the save area
1447 if (double_slots != 0) {
1448 stack_slots = round_to(stack_slots, 2);
1449 }
1450 } else {
1451 total_save_slots = 4 * VMRegImpl::slots_per_word; // 4 arguments passed in registers
1452 }
1453 assert(total_save_slots != -1, "initialize total_save_slots!");
1454
1455 int oop_handle_offset = stack_slots;
1456 stack_slots += total_save_slots;
1457
1458 // Now any space we need for handlizing a klass if static method
1459
1460 int klass_slot_offset = 0;
1461 int klass_offset = -1;
1462 int lock_slot_offset = 0;
1463 bool is_static = false;
1464
1465 if (method->is_static()) {
1466 klass_slot_offset = stack_slots;
1467 stack_slots += VMRegImpl::slots_per_word;
1468 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1469 is_static = true;
1470 }
1471
1472 // Plus a lock if needed
1473
1474 if (method->is_synchronized()) {
1475 lock_slot_offset = stack_slots;
1476 stack_slots += VMRegImpl::slots_per_word;
1477 }
1478
1479 // Now a place (+2) to save return values or temp during shuffling
1480 // + 2 for return address (which we own) and saved rfp
1481 stack_slots += 4;
1482
1483 // Ok The space we have allocated will look like:
1484 //
1485 //
1486 // FP-> | saved lr |
1487 // |---------------------|
1488 // | saved fp |
1489 // |---------------------|
1490 // | 2 slots for moves |
1491 // |---------------------|
1492 // | lock box (if sync) |
1493 // |---------------------| <- lock_slot_offset
1494 // | klass (if static) |
1495 // |---------------------| <- klass_slot_offset
1496 // | oopHandle area |
1497 // |---------------------| <- oop_handle_offset (8 java arg registers)
1498 // | outbound memory |
1499 // | based arguments |
1500 // | |
1501 // |---------------------|
1502 // | |
1503 // SP-> | out_preserved_slots |
1504 //
1505 //
1506
1507
1508 // Now compute actual number of stack words we need rounding to make
1509 // stack properly aligned.
1510 stack_slots = round_to(stack_slots, StackAlignmentInSlots);
1511
1512 int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1513
1514 // First thing make an ic check to see if we should even be here
1515
1516 // We are free to use all registers as temps without saving them and
1517 // restoring them except rfp. rfp is the only callee save register
1518 // as far as the interpreter and the compiler(s) are concerned.
1519
1520
1521 const Register ic_reg = rscratch2;
1522 const Register receiver = j_rarg0;
1523
1524 Label hit;
1525 Label exception_pending;
1526
1527 assert_different_registers(ic_reg, receiver, rscratch1);
1528 __ verify_oop(receiver);
1529 __ cmp_klass(receiver, ic_reg, rscratch1);
1530 __ b(hit, Assembler::EQ);
1531
1532 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1533
1534 // Verified entry point must be aligned
1535 __ align(8);
1536
1537 __ bind(hit);
1538
1539 #ifdef ASSERT
1540 __ mov(ic_reg, 0xdead); // trash ic_reg(rscratch2), as used as real scratch further
1541 #endif
1542
1543 int vep_offset = ((intptr_t)__ pc()) - start;
1544
1545 // Generate stack overflow check
1546
1547 // If we have to make this method not-entrant we'll overwrite its
1548 // first instruction with a jump. For this action to be legal we
1549 // must ensure that this first instruction is a B, BL, NOP, BKPT,
1550 // SVC, HVC, or SMC. Make it a NOP.
1551 __ nop();
1552
1553 if (UseStackBanging) {
1554 __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
1555 } else {
1556 Unimplemented();
1557 }
1558
1559 // Generate a new frame for the wrapper.
1560 __ enter();
1561 // -2 because return address is already present and so is saved rfp
1562 __ sub(sp, sp, stack_size - 2*wordSize);
1563
1564 // Frame is now completed as far as size and linkage.
1565 int frame_complete = ((intptr_t)__ pc()) - start;
1566
1567 if (is_critical_native) {
1568 check_needs_gc_for_critical_native(masm, stack_slots, total_c_args, total_in_args,
1569 oop_handle_offset, oop_maps, in_regs, in_sig_bt);
1570 }
1571
1572 //
1573 // We immediately shuffle the arguments so that any vm call we have to
1574 // make from here on out (sync slow path, jvmti, etc.) we will have
1575 // captured the oops from our caller and have a valid oopMap for
1576 // them.
1577
1578 // -----------------
1579 // The Grand Shuffle
1580
1581 // The Java calling convention is either equal (linux) or denser (win64) than the
1582 // c calling convention. However the because of the jni_env argument the c calling
1583 // convention always has at least one more (and two for static) arguments than Java.
1584 // Therefore if we move the args from java -> c backwards then we will never have
1585 // a register->register conflict and we don't have to build a dependency graph
1586 // and figure out how to break any cycles.
1587 //
1588
1589 // Record sp-based slot for receiver on stack for non-static methods
1590 int receiver_offset = -1;
1591
1592 // This is a trick. We double the stack slots so we can claim
1593 // the oops in the caller's frame. Since we are sure to have
1594 // more args than the caller doubling is enough to make
1595 // sure we can capture all the incoming oop args from the
1596 // caller.
1597 //
1598 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1599
1600 // Mark location of rfp (someday)
1601 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rfp));
1602
1603
1604 #ifdef ASSERT
1605 bool reg_destroyed[RegisterImpl::number_of_registers];
1606 bool freg_destroyed[FloatRegisterImpl::number_of_registers];
1607 for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) {
1608 reg_destroyed[r] = false;
1609 }
1610 for ( int f = 0 ; f < FloatRegisterImpl::number_of_registers ; f++ ) {
1611 freg_destroyed[f] = false;
1612 }
1613
1614 #endif // ASSERT
1615
1616 // This may iterate in two different directions depending on the
1617 // kind of native it is. The reason is that for regular JNI natives
1618 // the incoming and outgoing registers are offset upwards and for
1619 // critical natives they are offset down.
1620 GrowableArray<int> arg_order(2 * total_in_args);
1621 VMRegPair tmp_vmreg;
1622 tmp_vmreg.set1(rscratch2->as_VMReg());
1623
1624 if (!is_critical_native) {
1625 for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1626 arg_order.push(i);
1627 arg_order.push(c_arg);
1628 }
1629 } else {
1630 // Compute a valid move order, using tmp_vmreg to break any cycles
1631 ComputeMoveOrder cmo(total_in_args, in_regs, total_c_args, out_regs, in_sig_bt, arg_order, tmp_vmreg);
1632 }
1633
1634 int temploc = -1;
1635 for (int ai = 0; ai < arg_order.length(); ai += 2) {
1636 int i = arg_order.at(ai);
1637 int c_arg = arg_order.at(ai + 1);
1638 __ block_comment(err_msg("move %d -> %d", i, c_arg));
1639 if (c_arg == -1) {
1640 assert(is_critical_native, "should only be required for critical natives");
1641 // This arg needs to be moved to a temporary
1642 __ mov(tmp_vmreg.first()->as_Register(), in_regs[i].first()->as_Register());
1643 in_regs[i] = tmp_vmreg;
1644 temploc = i;
1645 continue;
1646 } else if (i == -1) {
1647 assert(is_critical_native, "should only be required for critical natives");
1648 // Read from the temporary location
1649 assert(temploc != -1, "must be valid");
1650 i = temploc;
1651 temploc = -1;
1652 }
1653 #ifdef ASSERT
1654 if (in_regs[i].first()->is_Register()) {
1655 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1656 } else if (in_regs[i].first()->is_FloatRegister()) {
1657 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!");
1658 }
1659 if (out_regs[c_arg].first()->is_Register()) {
1660 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1661 } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1662 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1663 }
1664 #endif // ASSERT
1665 switch (in_sig_bt[i]) {
1666 case T_ARRAY:
1667 if (is_critical_native) {
1668 unpack_array_argument(masm, in_regs[i], in_elem_bt[i], out_regs[c_arg + 1], out_regs[c_arg]);
1669 c_arg++;
1670 #ifdef ASSERT
1671 if (out_regs[c_arg].first()->is_Register()) {
1672 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1673 } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1674 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1675 }
1676 #endif
1677 break;
1678 }
1679 case T_OBJECT:
1680 assert(!is_critical_native, "no oop arguments");
1681 object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1682 ((i == 0) && (!is_static)),
1683 &receiver_offset);
1684 break;
1685 case T_VOID:
1686 break;
1687
1688 case T_FLOAT:
1689 float_move(masm, in_regs[i], out_regs[c_arg]);
1690 break;
1691
1692 case T_DOUBLE:
1693 assert( i + 1 < total_in_args &&
1694 in_sig_bt[i + 1] == T_VOID &&
1695 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1696 double_move(masm, in_regs[i], out_regs[c_arg]);
1697 break;
1698
1699 case T_LONG :
1700 long_move(masm, in_regs[i], out_regs[c_arg]);
1701 break;
1702
1703 case T_BOOLEAN :
1704 case T_BYTE :
1705 case T_CHAR :
1706 case T_SHORT :
1707 case T_INT :
1708 move_int(masm, in_regs[i], out_regs[c_arg]);
1709 break;
1710
1711 case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1712 case T_NARROWOOP :
1713 case T_METADATA :
1714 case T_NARROWKLASS :
1715 default:
1716 ShouldNotReachHere();
1717 }
1718 }
1719
1720 // point c_arg at the first arg that is already loaded in case we
1721 // need to spill before we call out
1722 int c_arg = total_c_args - total_in_args;
1723
1724 // We use r4 as the oop handle for the receiver/klass
1725 // It is callee save so it survives the call to native
1726
1727 const Register oop_handle_reg = r4;
1728
1729 // Pre-load a static method's oop. Used both by locking code and
1730 // the normal JNI call code.
1731 if (method->is_static() && !is_critical_native) {
1732
1733 // load oop into a register
1734 __ movoop(oop_handle_reg,
1735 JNIHandles::make_local(method->method_holder()->java_mirror()),
1736 /*immediate*/true);
1737
1738 // Now handlize the static class mirror it's known not-null.
1739 __ str(oop_handle_reg, Address(sp, klass_offset));
1740 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1741
1742 // Now get the handle
1743 __ lea(oop_handle_reg, Address(sp, klass_offset));
1744 // store the klass handle as second argument
1745 __ mov(c_rarg1, oop_handle_reg);
1746 // and protect the arg if we must spill
1747 c_arg--;
1748 }
1749
1750 // Change state to native (we save the return address in the thread, since it might not
1751 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1752 // points into the right code segment. It does not have to be the correct return pc.
1753 // We use the same pc/oopMap repeatedly when we call out
1754
1755 intptr_t the_pc = (intptr_t) __ pc();
1756 oop_maps->add_gc_map(the_pc - start, map);
1757
1758 __ set_last_Java_frame(sp, noreg, (address)the_pc, rscratch1);
1759
1760
1761 // We have all of the arguments setup at this point. We must not touch any register
1762 // argument registers at this point (what if we save/restore them there are no oop?
1763
1764 #ifdef DTRACE_ENABLED
1765 {
1766 SkipIfEqual skip(masm, &DTraceMethodProbes, false);
1767 // protect the args we've loaded
1768 save_args(masm, total_c_args, c_arg, out_regs);
1769 __ mov_metadata(c_rarg1, method());
1770 __ call_VM_leaf(
1771 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1772 rthread, c_rarg1);
1773 restore_args(masm, total_c_args, c_arg, out_regs);
1774 }
1775 #endif
1776
1777 // RedefineClasses() tracing support for obsolete method entry
1778 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1779 // protect the args we've loaded
1780 save_args(masm, total_c_args, c_arg, out_regs);
1781 __ mov_metadata(c_rarg1, method());
1782 __ call_VM_leaf(
1783 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1784 rthread, c_rarg1);
1785 restore_args(masm, total_c_args, c_arg, out_regs);
1786 }
1787
1788 // Lock a synchronized method
1789
1790 // Register definitions used by locking and unlocking
1791
1792 Label slow_path_lock;
1793 Label lock_done;
1794
1795 if (method->is_synchronized()) {
1796 assert(!is_critical_native, "unhandled");
1797 // TODO Fast path disabled as requires at least 4 registers, which already contain arguments prepared for call
1798
1799 // Get the handle (the 2nd argument)
1800 __ mov(oop_handle_reg, c_rarg1);
1801 __ b(slow_path_lock);
1802
1803 // Slow path will re-enter here
1804 __ bind(lock_done);
1805 }
1806
1807
1808 // Finally just about ready to make the JNI call
1809
1810
1811 // get JNIEnv* which is first argument to native
1812 if (!is_critical_native) {
1813 __ lea(c_rarg0, Address(rthread, in_bytes(JavaThread::jni_environment_offset())));
1814 }
1815
1816 // Now set thread in native
1817 __ mov(rscratch1, _thread_in_native);
1818 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1819 __ dmb(Assembler::ISH);
1820 __ str(rscratch1, rscratch2);
1821
1822 // Do the call
1823 rt_call(masm, native_func);
1824
1825 // Unpack native results.
1826 switch (ret_type) {
1827 case T_BOOLEAN: __ uxtb(r0, r0); break;
1828 case T_CHAR : __ uxth(r0, r0); break;
1829 case T_BYTE : __ sxtb(r0, r0); break;
1830 case T_SHORT : __ sxth(r0, r0); break;
1831 case T_INT : break;
1832 case T_DOUBLE :
1833 case T_FLOAT :
1834 // Result is in d0 we'll save as needed
1835 break;
1836 case T_ARRAY: // Really a handle
1837 case T_OBJECT: // Really a handle
1838 break; // can't de-handlize until after safepoint check
1839 case T_VOID: break;
1840 case T_LONG: break;
1841 default : ShouldNotReachHere();
1842 }
1843
1844 // Switch thread to "native transition" state before reading the synchronization state.
1845 // This additional state is necessary because reading and testing the synchronization
1846 // state is not atomic w.r.t. GC, as this scenario demonstrates:
1847 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1848 // VM thread changes sync state to synchronizing and suspends threads for GC.
1849 // Thread A is resumed to finish this native method, but doesn't block here since it
1850 // didn't see any synchronization is progress, and escapes.
1851 __ mov(rscratch1, _thread_in_native_trans);
1852 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1853 __ dmb(Assembler::ISH);
1854 __ str(rscratch1, rscratch2);
1855
1856 if(os::is_MP()) {
1857 if (UseMembar) {
1858 // Force this write out before the read below
1859 __ dmb(Assembler::SY);
1860 } else {
1861 // Write serialization page so VM thread can do a pseudo remote membar.
1862 // We use the current thread pointer to calculate a thread specific
1863 // offset to write to within the page. This minimizes bus traffic
1864 // due to cache line collision.
1865 __ serialize_memory(rthread, rscratch1);
1866 }
1867 }
1868
1869 Label after_transition;
1870
1871 // check for safepoint operation in progress and/or pending suspend requests
1872 {
1873 Label Continue;
1874
1875 __ mov(rscratch1, ExternalAddress((address)SafepointSynchronize::address_of_state()));
1876 __ ldr(rscratch1, Address(rscratch1));
1877 __ cmp(rscratch1, SafepointSynchronize::_not_synchronized);
1878
1879 Label L;
1880 __ b(L, Assembler::NE);
1881 __ ldr(rscratch1, Address(rthread, JavaThread::suspend_flags_offset()));
1882 __ cbz(rscratch1, Continue);
1883 __ bind(L);
1884
1885 // Don't use call_VM as it will see a possible pending exception and forward it
1886 // and never return here preventing us from clearing _last_native_pc down below.
1887 //
1888 save_native_result(masm, ret_type, stack_slots);
1889 __ mov(c_rarg0, rthread);
1890 #ifndef PRODUCT
1891 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
1892 #endif
1893 if (!is_critical_native) {
1894 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1895 } else {
1896 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans_and_transition)));
1897 }
1898 __ bl(rscratch1);
1899 __ maybe_isb();
1900 // Restore any method result value
1901 restore_native_result(masm, ret_type, stack_slots);
1902
1903 if (is_critical_native) {
1904 // The call above performed the transition to thread_in_Java so
1905 // skip the transition logic below.
1906 __ b(after_transition);
1907 }
1908
1909 __ bind(Continue);
1910 }
1911
1912 // change thread state
1913 __ mov(rscratch1, _thread_in_Java);
1914 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1915 __ dmb(Assembler::ISH);
1916 __ str(rscratch1, rscratch2);
1917 __ bind(after_transition);
1918
1919 Label reguard;
1920 Label reguard_done;
1921 __ ldrb(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
1922 __ cmp(rscratch1, JavaThread::stack_guard_yellow_disabled);
1923 __ b(reguard, Assembler::EQ);
1924 __ bind(reguard_done);
1925
1926 // native result if any is live
1927
1928 // Unlock
1929 Label unlock_done;
1930 Label slow_path_unlock;
1931 if (method->is_synchronized()) {
1932 // TODO fast path disabled as requires at least 4 registers, but r0,r1 contains result
1933 __ b(slow_path_unlock);
1934
1935 // slow path re-enters here
1936 __ bind(unlock_done);
1937 }
1938
1939 #ifdef DTRACE_ENABLED
1940 {
1941 SkipIfEqual skip(masm, &DTraceMethodProbes, false);
1942 save_native_result(masm, ret_type, stack_slots);
1943 __ mov_metadata(c_rarg1, method());
1944 __ call_VM_leaf(
1945 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1946 rthread, c_rarg1);
1947 restore_native_result(masm, ret_type, stack_slots);
1948 }
1949 #endif
1950
1951 __ reset_last_Java_frame(false, true);
1952
1953 // Unpack oop result
1954 if (ret_type == T_OBJECT || ret_type == T_ARRAY) {
1955 Label L;
1956 __ cbz(r0, L);
1957 __ ldr(r0, Address(r0, 0));
1958 __ bind(L);
1959 __ verify_oop(r0);
1960 }
1961
1962 if (!is_critical_native) {
1963 // reset handle block
1964 __ mov(rscratch1, 0);
1965 __ ldr(r2, Address(rthread, JavaThread::active_handles_offset()));
1966 __ str(rscratch1, Address(r2, JNIHandleBlock::top_offset_in_bytes()));
1967 }
1968
1969 __ leave();
1970
1971 if (!is_critical_native) {
1972 // Any exception pending?
1973 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1974 __ cbnz(rscratch1, exception_pending);
1975 }
1976
1977 // We're done
1978 __ b(lr);
1979
1980 // Unexpected paths are out of line and go here
1981
1982 if (!is_critical_native) {
1983 // forward the exception
1984 __ bind(exception_pending);
1985
1986 // and forward the exception
1987 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1988 }
1989
1990 // Slow path locking & unlocking
1991 if (method->is_synchronized()) {
1992
1993 // BEGIN Slow path lock
1994 __ bind(slow_path_lock);
1995
1996 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1997 // args are (oop obj, BasicLock* lock, JavaThread* thread)
1998
1999 // protect the args we've loaded
2000 save_args(masm, total_c_args, c_arg, out_regs);
2001
2002 __ ldr(c_rarg0, Address(oop_handle_reg));
2003 __ lea(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
2004 __ mov(c_rarg2, rthread);
2005
2006 // Not a leaf but we have last_Java_frame setup as we want
2007 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
2008 restore_args(masm, total_c_args, c_arg, out_regs);
2009
2010 #ifdef ASSERT
2011 { Label L;
2012 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2013 __ cbz(rscratch1, L);
2014 __ stop("no pending exception allowed on exit from monitorenter");
2015 __ bind(L);
2016 }
2017 #endif
2018 __ b(lock_done);
2019
2020 // END Slow path lock
2021
2022 // BEGIN Slow path unlock
2023 __ bind(slow_path_unlock);
2024
2025 // If we haven't already saved the native result we must save it now as xmm registers
2026 // are still exposed.
2027
2028 save_native_result(masm, ret_type, stack_slots);
2029
2030 __ ldr(c_rarg0, Address(oop_handle_reg));
2031 __ lea(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
2032
2033 // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
2034 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2035 __ mov(rscratch2, 0);
2036 __ str(rscratch2, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2037
2038 rt_call(masm, CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C));
2039
2040 #ifdef ASSERT
2041 {
2042 Label L;
2043 __ ldr(rscratch2, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2044 __ cbz(rscratch2, L);
2045 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
2046 __ bind(L);
2047 }
2048 #endif // ASSERT
2049
2050 __ str(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2051
2052 restore_native_result(masm, ret_type, stack_slots);
2053
2054 __ b(unlock_done);
2055
2056 // END Slow path unlock
2057
2058 } // synchronized
2059
2060 // SLOW PATH Reguard the stack if needed
2061
2062 __ bind(reguard);
2063 save_native_result(masm, ret_type, stack_slots);
2064 rt_call(masm, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
2065 restore_native_result(masm, ret_type, stack_slots);
2066 // and continue
2067 __ b(reguard_done);
2068
2069
2070
2071 __ flush();
2072
2073 nmethod *nm = nmethod::new_native_nmethod(method,
2074 compile_id,
2075 masm->code(),
2076 vep_offset,
2077 frame_complete,
2078 stack_slots / VMRegImpl::slots_per_word,
2079 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2080 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2081 oop_maps);
2082
2083 if (is_critical_native) {
2084 nm->set_lazy_critical_native(true);
2085 }
2086
2087 return nm;
2088 }
2089
2090 // this function returns the adjust size (in number of words) to a c2i adapter
2091 // activation for use during deoptimization
2092 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) {
2093 assert(callee_locals >= callee_parameters,
2094 "test and remove; got more parms than locals");
2095 if (callee_locals < callee_parameters)
2096 return 0; // No adjustment for negative locals
2097 int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2098 // diff is counted in stack words
2099 return round_to(diff, 2);
2100 }
2101
2102
2103 //------------------------------generate_deopt_blob----------------------------
2104 void SharedRuntime::generate_deopt_blob() {
2105
2106 // Allocate space for the code
2107 ResourceMark rm;
2108 // Setup code generation tools
2109 CodeBuffer buffer("deopt_blob", 2048, 1024);
2110 MacroAssembler* masm = new MacroAssembler(&buffer);
2111 int frame_size_in_words;
2112 OopMap* map = NULL;
2113 OopMapSet *oop_maps = new OopMapSet();
2114
2115 // -------------
2116 // This code enters when returning to a de-optimized nmethod. A return
2117 // address has been pushed on the the stack, and return values are in
2118 // registers.
2119 // If we are doing a normal deopt then we were called from the patched
2120 // nmethod from the point we returned to the nmethod. So the return
2121 // address on the stack is wrong by NativeCall::instruction_size
2122 // We will adjust the value so it looks like we have the original return
2123 // address on the stack (like when we eagerly deoptimized).
2124 // In the case of an exception pending when deoptimizing, we enter
2125 // with a return address on the stack that points after the call we patched
2126 // into the exception handler. We have the following register state from,
2127 // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
2128 // r0: exception oop
2129 // r7: exception handler
2130 // r3: throwing pc
2131 // So in this case we simply jam r3 into the useless return address and
2132 // the stack looks just like we want.
2133 //
2134 // At this point we need to de-opt. We save the argument return
2135 // registers. We call the first C routine, fetch_unroll_info(). This
2136 // routine captures the return values and returns a structure which
2137 // describes the current frame size and the sizes of all replacement frames.
2138 // The current frame is compiled code and may contain many inlined
2139 // functions, each with their own JVM state. We pop the current frame, then
2140 // push all the new frames. Then we call the C routine unpack_frames() to
2141 // populate these frames. Finally unpack_frames() returns us the new target
2142 // address. Notice that callee-save registers are BLOWN here; they have
2143 // already been captured in the vframeArray at the time the return PC was
2144 // patched.
2145 address start = __ pc();
2146 Label cont;
2147
2148 // Prolog for non exception case!
2149
2150 // Save everything in sight.
2151 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2152
2153 // Normal deoptimization. Save exec mode for unpack_frames.
2154 __ mov(r7, Deoptimization::Unpack_deopt); // callee-saved
2155 __ b(cont);
2156
2157 int reexecute_offset = __ pc() - start;
2158
2159 // Reexecute case
2160 // return address is the pc describes what bci to do re-execute at
2161
2162 // No need to update map as each call to save_live_registers will produce identical oopmap
2163 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2164
2165 __ mov(r7, Deoptimization::Unpack_reexecute); // callee-saved
2166 __ b(cont);
2167
2168 int exception_offset = __ pc() - start;
2169
2170 // Prolog for exception case
2171
2172 // all registers are dead at this entry point, except for r0, and
2173 // r3 which contain the exception oop and exception pc
2174 // respectively. Set them in TLS and fall thru to the
2175 // unpack_with_exception_in_tls entry point.
2176
2177 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
2178 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
2179
2180 int exception_in_tls_offset = __ pc() - start;
2181
2182 // new implementation because exception oop is now passed in JavaThread
2183
2184 // Prolog for exception case
2185 // All registers must be preserved because they might be used by LinearScan
2186 // Exceptiop oop and throwing PC are passed in JavaThread
2187 // tos: stack at point of call to method that threw the exception (i.e. only
2188 // args are on the stack, no return address)
2189
2190 // The return address pushed by save_live_registers will be patched
2191 // later with the throwing pc. The correct value is not available
2192 // now because loading it from memory would destroy registers.
2193
2194 // NB: The SP at this point must be the SP of the method that is
2195 // being deoptimized. Deoptimization assumes that the frame created
2196 // here by save_live_registers is immediately below the method's SP.
2197 // This is a somewhat fragile mechanism.
2198
2199 // Save everything in sight.
2200 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2201
2202 // Now it is safe to overwrite any register
2203
2204 // Deopt during an exception. Save exec mode for unpack_frames.
2205 __ mov(r7, Deoptimization::Unpack_exception); // callee-saved
2206
2207 // load throwing pc from JavaThread and patch it as the return address
2208 // of the current frame. Then clear the field in JavaThread
2209
2210 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2211 __ str(r3, Address(rfp));
2212 __ mov(rscratch1, 0);
2213 __ str(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
2214
2215 #ifdef ASSERT
2216 // verify that there is really an exception oop in JavaThread
2217 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2218 __ verify_oop(r0);
2219
2220 // verify that there is no pending exception
2221 Label no_pending_exception;
2222 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2223 __ cbz(rscratch1, no_pending_exception);
2224 __ stop("must not have pending exception here");
2225 __ bind(no_pending_exception);
2226 #endif
2227
2228 __ bind(cont);
2229
2230 // Call C code. Need thread and this frame, but NOT official VM entry
2231 // crud. We cannot block on this call, no GC can happen.
2232 //
2233 // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2234
2235 // fetch_unroll_info needs to call last_java_frame().
2236
2237 Label retaddr;
2238 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2239 #ifdef ASSERT0
2240 { Label L;
2241 __ ldr(rscratch1, Address(rthread,
2242 JavaThread::last_Java_fp_offset()));
2243 __ cbz(rscratch1, L);
2244 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2245 __ bind(L);
2246 }
2247 #endif // ASSERT
2248 __ mov(c_rarg0, rthread);
2249 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2250 __ bl(rscratch1);
2251 __ bind(retaddr);
2252
2253 // Need to have an oopmap that tells fetch_unroll_info where to
2254 // find any register it might need.
2255 oop_maps->add_gc_map(__ pc() - start, map);
2256
2257 __ reset_last_Java_frame(false, true);
2258
2259 // Load UnrollBlock* into rdi
2260 __ mov(r5, r0);
2261
2262 Label noException;
2263 __ cmp(r7, Deoptimization::Unpack_exception); // Was exception pending?
2264 __ b(noException, Assembler::NE);
2265 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2266 // QQQ this is useless it was NULL above
2267 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2268 __ mov(rscratch1, 0);
2269 __ str(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
2270 __ str(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
2271
2272 __ verify_oop(r0);
2273
2274 // Overwrite the result registers with the exception results.
2275 __ str(r0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::r0_off)));
2276 // I think this is useless
2277 // __ str(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2278
2279 __ bind(noException);
2280
2281 // Only register save data is on the stack.
2282 // Now restore the result registers. Everything else is either dead
2283 // or captured in the vframeArray.
2284 RegisterSaver::restore_result_registers(masm);
2285
2286 // All of the register save area has been popped of the stack. Only the
2287 // return address remains.
2288
2289 // Pop all the frames we must move/replace.
2290 //
2291 // Frame picture (youngest to oldest)
2292 // 1: self-frame (no frame link)
2293 // 2: deopting frame (no frame link)
2294 // 3: caller of deopting frame (could be compiled/interpreted).
2295 //
2296 // Note: by leaving the return address of self-frame on the stack
2297 // and using the size of frame 2 to adjust the stack
2298 // when we are done the return to frame 3 will still be on the stack.
2299
2300 // Pop deoptimized frame
2301 __ ldr(r2, Address(r5, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
2302 __ sub(r2, r2, 2 * wordSize);
2303 __ add(sp, sp, r2);
2304 __ ldrd(rfp, lr, __ post(sp, 2 * wordSize));
2305 // LR should now be the return address to the caller (3)
2306
2307 #ifdef ASSERT
2308 // Compilers generate code that bang the stack by as much as the
2309 // interpreter would need. So this stack banging should never
2310 // trigger a fault. Verify that it does not on non product builds.
2311 if (UseStackBanging) {
2312 __ ldr(rscratch2, Address(r5, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
2313 __ bang_stack_size(rscratch2, r2);
2314 }
2315 #endif
2316 // Load address of array of frame pcs into r2
2317 __ ldr(r2, Address(r5, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2318
2319 // Trash the old pc
2320 // __ addptr(sp, wordSize); FIXME ????
2321
2322 // Load address of array of frame sizes into r4
2323 __ ldr(r4, Address(r5, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
2324
2325 // Load counter into r3
2326 __ ldr(r3, Address(r5, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
2327
2328 // Now adjust the caller's stack to make up for the extra locals
2329 // but record the original sp so that we can save it in the skeletal interpreter
2330 // frame and the stack walking of interpreter_sender will get the unextended sp
2331 // value and not the "real" sp value.
2332
2333 const Register sender_sp = r6;
2334
2335 __ mov(sender_sp, sp);
2336 __ ldr(rscratch1, Address(r5,
2337 Deoptimization::UnrollBlock::
2338 caller_adjustment_offset_in_bytes()));
2339 __ sub(sp, sp, rscratch1);
2340
2341 // Push interpreter frames in a loop
2342 __ mov(rscratch1, (address)0xDEADDEAD); // Make a recognizable pattern
2343 // Initially used to place 0xDEADDEAD in rscratch2 as well - why?
2344 __ mov(rscratch2, 0);
2345 Label loop;
2346 __ bind(loop);
2347 __ ldr(rscratch1, Address(__ post(r4, wordSize))); // Load frame size
2348 __ sub(rscratch1, rscratch1, 2*wordSize); // We'll push pc and fp by hand
2349 __ ldr(lr, Address(__ post(r2, wordSize))); // Load pc
2350 __ enter(); // Save old & set new fp
2351 __ sub(sp, sp, rscratch1); // Prolog
2352 // This value is corrected by layout_activation_impl
2353 __ str(rscratch2, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
2354 __ str(sender_sp, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2355 __ mov(sender_sp, sp); // Pass sender_sp to next frame
2356 __ sub(r3, r3, 1); // Decrement counter
2357 __ cbnz(r3, loop);
2358
2359 // Re-push self-frame
2360 __ ldr(lr, Address(r2));
2361 __ enter();
2362
2363 // Allocate a full sized register save area. We subtract 2 because
2364 // enter() just pushed 2 words
2365 __ sub(sp, sp, (frame_size_in_words - 2) * wordSize);
2366
2367 // Restore frame locals after moving the frame
2368 __ vstr_f64(d0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::fpu_state_off)));
2369 __ strd(r0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::r0_off)));
2370
2371 // Call C code. Need thread but NOT official VM entry
2372 // crud. We cannot block on this call, no GC can happen. Call should
2373 // restore return values to their stack-slots with the new SP.
2374 //
2375 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2376
2377 // Use rfp because the frames look interpreted now
2378 // Don't need the precise return PC here, just precise enough to point into this code blob.
2379 address the_pc = __ pc();
2380 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2381
2382 __ mov(c_rarg0, rthread);
2383 __ mov(c_rarg1, r7); // second arg: exec_mode
2384 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2385 __ bl(rscratch1);
2386
2387 // Set an oopmap for the call site
2388 // Use the same PC we used for the last java frame
2389 oop_maps->add_gc_map(the_pc - start,
2390 new OopMap( frame_size_in_words, 0 ));
2391
2392 // Clear fp AND pc
2393 __ reset_last_Java_frame(true, true);
2394
2395 // Collect return values
2396 __ vldr_f64(d0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::fpu_state_off)));
2397 __ ldrd(r0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::r0_off)));
2398 // I think this is useless (throwing pc?)
2399 // __ ldr(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2400
2401 // Pop self-frame.
2402 __ leave(); // Epilog
2403
2404 // Jump to interpreter
2405 __ b(lr);
2406
2407 // Make sure all code is generated
2408 masm->flush();
2409
2410 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2411 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2412
2413 }
2414
2415 uint SharedRuntime::out_preserve_stack_slots() {
2416 return 0;
2417 }
2418
2419 #ifdef COMPILER2
2420 //------------------------------generate_uncommon_trap_blob--------------------
2421 /*void SharedRuntime::generate_uncommon_trap_blob() {
2422 // Allocate space for the code
2423 ResourceMark rm;
2424 // Setup code generation tools
2425 CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
2426 MacroAssembler* masm = new MacroAssembler(&buffer);
2427
2428 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2429
2430 address start = __ pc();
2431
2432 // Push self-frame. We get here with a return address in LR
2433 // and sp should be 16 byte aligned
2434 // push rfp and retaddr by hand
2435 __ strd(rfp, lr, Address(__ pre(sp, -2 * wordSize)));
2436 // we don't expect an arg reg save area
2437 #ifndef PRODUCT
2438 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2439 #endif
2440 // compiler left unloaded_class_index in j_rarg0 move to where the
2441 // runtime expects it.
2442 if (c_rarg1 != j_rarg0) {
2443 __ mov(c_rarg1, j_rarg0);
2444 }
2445
2446 // we need to set the past SP to the stack pointer of the stub frame
2447 // and the pc to the address where this runtime call will return
2448 // although actually any pc in this code blob will do).
2449 Label retaddr;
2450 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2451
2452 // Call C code. Need thread but NOT official VM entry
2453 // crud. We cannot block on this call, no GC can happen. Call should
2454 // capture callee-saved registers as well as return values.
2455 // Thread is in rdi already.
2456 //
2457 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
2458 //
2459 // n.b. 2 gp args, 0 fp args, integral return type
2460
2461 __ mov(c_rarg0, rthread);
2462 __ lea(rscratch1,
2463 RuntimeAddress(CAST_FROM_FN_PTR(address,
2464 Deoptimization::uncommon_trap)));
2465 __ bl(rscratch1);
2466 __ bind(retaddr);
2467
2468 // Set an oopmap for the call site
2469 OopMapSet* oop_maps = new OopMapSet();
2470 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
2471
2472 // location of rfp is known implicitly by the frame sender code
2473
2474 oop_maps->add_gc_map(__ pc() - start, map);
2475
2476 __ reset_last_Java_frame(false, true);
2477
2478 // move UnrollBlock* into r4
2479 __ mov(r4, r0);
2480
2481 // Pop all the frames we must move/replace.
2482 //
2483 // Frame picture (youngest to oldest)
2484 // 1: self-frame (no frame link)
2485 // 2: deopting frame (no frame link)
2486 // 3: caller of deopting frame (could be compiled/interpreted).
2487
2488 // Pop self-frame. We have no frame, and must rely only on r0 and sp.
2489 __ add(sp, sp, (SimpleRuntimeFrame::framesize) << LogBytesPerInt); // Epilog!
2490
2491 // Pop deoptimized frame (int)
2492 __ ldr(r2, Address(r4,
2493 Deoptimization::UnrollBlock::
2494 size_of_deoptimized_frame_offset_in_bytes()));
2495 __ sub(r2, r2, 2 * wordSize);
2496 __ add(sp, sp, r2);
2497 __ ldrd(rfp, lr, __ post(sp, 2 * wordSize));
2498 // LR should now be the return address to the caller (3) frame
2499
2500 #ifdef ASSERT
2501 // Compilers generate code that bang the stack by as much as the
2502 // interpreter would need. So this stack banging should never
2503 // trigger a fault. Verify that it does not on non product builds.
2504 if (UseStackBanging) {
2505 __ ldr(r1, Address(r4,
2506 Deoptimization::UnrollBlock::
2507 total_frame_sizes_offset_in_bytes()));
2508 __ bang_stack_size(r1, r2);
2509 }
2510 #endif
2511
2512 // Load address of array of frame pcs into r2 (address*)
2513 __ ldr(r2, Address(r4,
2514 Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2515
2516 // Load address of array of frame sizes into r5 (intptr_t*)
2517 __ ldr(r5, Address(r4,
2518 Deoptimization::UnrollBlock::
2519 frame_sizes_offset_in_bytes()));
2520
2521 // Counter
2522 __ ldr(r3, Address(r4,
2523 Deoptimization::UnrollBlock::
2524 number_of_frames_offset_in_bytes())); // (int)
2525
2526 // Now adjust the caller's stack to make up for the extra locals but
2527 // record the original sp so that we can save it in the skeletal
2528 // interpreter frame and the stack walking of interpreter_sender
2529 // will get the unextended sp value and not the "real" sp value.
2530
2531 const Register sender_sp = r8;
2532
2533 __ mov(sender_sp, sp);
2534 __ ldr(r1, Address(r4,
2535 Deoptimization::UnrollBlock::
2536 caller_adjustment_offset_in_bytes())); // (int)
2537 __ sub(sp, sp, r1);
2538
2539 __ mov(rscratch1, 0);
2540 // Push interpreter frames in a loop
2541 Label loop;
2542 __ bind(loop);
2543 __ ldr(r1, Address(r5, 0)); // Load frame size
2544 __ sub(r1, r1, 2 * wordSize); // We'll push pc and rfp by hand
2545 __ ldr(lr, Address(r2, 0)); // Save return address
2546 __ enter(); // and old rfp & set new rfp
2547 __ sub(sp, sp, r1); // Prolog
2548 __ str(sender_sp, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2549 // This value is corrected by layout_activation_impl
2550 __ str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); //zero it
2551 __ mov(sender_sp, sp); // Pass sender_sp to next frame
2552 __ add(r5, r5, wordSize); // Bump array pointer (sizes)
2553 __ add(r2, r2, wordSize); // Bump array pointer (pcs)
2554 __ subs(r3, r3, 1); // Decrement counter
2555 __ b(loop, Assembler::GT);
2556 __ ldr(lr, Address(r2, 0)); // save final return address
2557 // Re-push self-frame
2558 __ enter(); // & old rfp & set new rfp
2559
2560 // Use rfp because the frames look interpreted now
2561 // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
2562 // Don't need the precise return PC here, just precise enough to point into this code blob.
2563 address the_pc = __ pc();
2564 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2565
2566 // Call C code. Need thread but NOT official VM entry
2567 // crud. We cannot block on this call, no GC can happen. Call should
2568 // restore return values to their stack-slots with the new SP.
2569 // Thread is in rdi already.
2570 //
2571 // BasicType unpack_frames(JavaThread* thread, int exec_mode);
2572 //
2573 // n.b. 2 gp args, 0 fp args, integral return type
2574
2575 // sp should already be aligned
2576 __ mov(c_rarg0, rthread);
2577 __ mov(c_rarg1, (unsigned)Deoptimization::Unpack_uncommon_trap);
2578 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2579 __ bl(rscratch1);
2580
2581 // Set an oopmap for the call site
2582 // Use the same PC we used for the last java frame
2583 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2584
2585 // Clear fp AND pc
2586 __ reset_last_Java_frame(true, true);
2587
2588 // Pop self-frame.
2589 __ leave(); // Epilog
2590
2591 // Jump to interpreter
2592 __ b(lr);
2593
2594 // Make sure all code is generated
2595 masm->flush();
2596
2597 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps,
2598 SimpleRuntimeFrame::framesize >> 1);
2599
2600 } */
2601 #endif // COMPILER2
2602
2603
2604 //------------------------------generate_handler_blob------
2605 //
2606 // Generate a special Compile2Runtime blob that saves all registers,
2607 // and setup oopmap.
2608 //
2609 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2610 ResourceMark rm;
2611 OopMapSet *oop_maps = new OopMapSet();
2612 OopMap* map;
2613
2614 // Allocate space for the code. Setup code generation tools.
2615 CodeBuffer buffer("handler_blob", 2048, 1024);
2616 MacroAssembler* masm = new MacroAssembler(&buffer);
2617
2618 address start = __ pc();
2619 address call_pc = NULL;
2620 int frame_size_in_words;
2621 bool cause_return = (poll_type == POLL_AT_RETURN);
2622 bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
2623
2624 // Save registers, fpu state, and flags
2625 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2626
2627 // The following is basically a call_VM. However, we need the precise
2628 // address of the call in order to generate an oopmap. Hence, we do all the
2629 // work outselves.
2630
2631 Label retaddr;
2632 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2633
2634 // The return address must always be correct so that frame constructor never
2635 // sees an invalid pc.
2636
2637 if (!cause_return) {
2638 // overwrite the return address pushed by save_live_registers
2639 __ ldr(lr, Address(rthread, JavaThread::saved_exception_pc_offset()));
2640 __ str(lr, Address(rfp));
2641 }
2642
2643 // Do the call
2644 __ mov(c_rarg0, rthread);
2645 __ lea(rscratch1, RuntimeAddress(call_ptr));
2646 __ bl(rscratch1);
2647 __ bind(retaddr);
2648
2649 // Set an oopmap for the call site. This oopmap will map all
2650 // oop-registers and debug-info registers as callee-saved. This
2651 // will allow deoptimization at this safepoint to find all possible
2652 // debug-info recordings, as well as let GC find all oops.
2653
2654 oop_maps->add_gc_map( __ pc() - start, map);
2655
2656 Label noException;
2657
2658 __ reset_last_Java_frame(false, true);
2659
2660 __ maybe_isb();
2661 __ membar(Assembler::LoadLoad | Assembler::LoadStore);
2662
2663 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2664 __ cbz(rscratch1, noException);
2665
2666 // Exception pending
2667
2668 RegisterSaver::restore_live_registers(masm);
2669
2670 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2671
2672 // No exception case
2673 __ bind(noException);
2674
2675 // Normal exit, restore registers and exit.
2676 RegisterSaver::restore_live_registers(masm);
2677
2678 __ b(lr);
2679
2680 // Make sure all code is generated
2681 masm->flush();
2682
2683 // Fill-out other meta info
2684 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2685 }
2686
2687 //
2688 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2689 //
2690 // Generate a stub that calls into vm to find out the proper destination
2691 // of a java call. All the argument registers are live at this point
2692 // but since this is generic code we don't know what they are and the caller
2693 // must do any gc of the args.
2694 //
2695 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
2696 assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
2697
2698 // allocate space for the code
2699 ResourceMark rm;
2700
2701 //CodeBuffer buffer(name, 1000, 512);
2702 CodeBuffer buffer(name, 2048, 512 ); // changed as error later
2703 MacroAssembler* masm = new MacroAssembler(&buffer);
2704
2705 int frame_size_in_words;
2706
2707 OopMapSet *oop_maps = new OopMapSet();
2708 OopMap* map = NULL;
2709
2710 int start = __ offset();
2711
2712 map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2713
2714 int frame_complete = __ offset();
2715
2716 {
2717 Label retaddr;
2718 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2719
2720 __ mov(c_rarg0, rthread);
2721 __ lea(rscratch1, RuntimeAddress(destination));
2722
2723 __ bl(rscratch1);
2724 __ bind(retaddr);
2725 }
2726
2727 // Set an oopmap for the call site.
2728 // We need this not only for callee-saved registers, but also for volatile
2729 // registers that the compiler might be keeping live across a safepoint.
2730
2731 oop_maps->add_gc_map( __ offset() - start, map);
2732
2733 __ maybe_isb();
2734
2735 // r0 contains the address we are going to jump to assuming no exception got installed
2736
2737 // clear last_Java_sp
2738 // TODO x86 have different action: reset_last_Java_frame(thread, true(fp), false(pc));
2739 // TODO below is false(fp), true(pc)
2740 __ reset_last_Java_frame(false, true);
2741 // check for pending exceptions
2742 Label pending;
2743 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2744 __ cbnz(rscratch1, pending);
2745
2746 // get the returned Method*
2747 __ get_vm_result_2(rmethod, rthread);
2748 __ str(rmethod, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::rmethod_off)));
2749
2750 // r0 is where we want to jump, overwrite rscratch1 which is saved and scratch
2751 __ str(r0, Address(sp, RegisterSaver::offset_in_bytes(RegisterSaver::rscratch1_off)));
2752 RegisterSaver::restore_live_registers(masm);
2753
2754 // We are back the the original state on entry and ready to go.
2755
2756 __ b(rscratch1);
2757
2758 // Pending exception after the safepoint
2759
2760 __ bind(pending);
2761
2762 RegisterSaver::restore_live_registers(masm);
2763
2764 // exception pending => remove activation and forward to exception handler
2765 __ mov(rscratch1, 0);
2766 __ str(rscratch1, Address(rthread, JavaThread::vm_result_offset()));
2767
2768 __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
2769 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2770
2771 // -------------
2772 // make sure all code is generated
2773 masm->flush();
2774
2775 // return the blob
2776 // frame_size_words or bytes??
2777 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
2778 }
2779
2780
2781 #ifdef COMPILER2
2782 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame
2783 //
2784 //------------------------------generate_exception_blob---------------------------
2785 // creates exception blob at the end
2786 // Using exception blob, this code is jumped from a compiled method.
2787 // (see emit_exception_handler in x86_64.ad file)
2788 //
2789 // Given an exception pc at a call we call into the runtime for the
2790 // handler in this method. This handler might merely restore state
2791 // (i.e. callee save registers) unwind the frame and jump to the
2792 // exception handler for the nmethod if there is no Java level handler
2793 // for the nmethod.
2794 //
2795 // This code is entered with a jmp.
2796 //
2797 // Arguments:
2798 // r0: exception oop
2799 // r3: exception pc
2800 //
2801 // Results:
2802 // r0: exception oop
2803 // r3: exception pc in caller or ???
2804 // destination: exception handler of caller
2805 //
2806 // Note: the exception pc MUST be at a call (precise debug information)
2807 // Registers r0, r3, r2, r4, r5, r8-r11 are not callee saved.
2808 //
2809
2810 void OptoRuntime::generate_exception_blob() {
2811 assert(!OptoRuntime::is_callee_saved_register(R3_num), "");
2812 assert(!OptoRuntime::is_callee_saved_register(R0_num), "");
2813 assert(!OptoRuntime::is_callee_saved_register(R2_num), "");
2814
2815 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2816
2817 // Allocate space for the code
2818 ResourceMark rm;
2819 // Setup code generation tools
2820 CodeBuffer buffer("exception_blob", 2048, 1024);
2821 MacroAssembler* masm = new MacroAssembler(&buffer);
2822
2823 __ stop("FIXME generate_exception_blob");
2824 // TODO check various assumptions made here
2825 //
2826 // make sure we do so before running this
2827
2828 address start = __ pc();
2829
2830 // push rfp and retaddr by hand
2831 // Exception pc is 'return address' for stack walker
2832 __ strd(rfp, lr, Address(__ pre(sp, -2 * wordSize)));
2833 // there are no callee save registers and we don't expect an
2834 // arg reg save area
2835 #ifndef PRODUCT
2836 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2837 #endif
2838 // Store exception in Thread object. We cannot pass any arguments to the
2839 // handle_exception call, since we do not want to make any assumption
2840 // about the size of the frame where the exception happened in.
2841 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
2842 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
2843
2844 // This call does all the hard work. It checks if an exception handler
2845 // exists in the method.
2846 // If so, it returns the handler address.
2847 // If not, it prepares for stack-unwinding, restoring the callee-save
2848 // registers of the frame being removed.
2849 //
2850 // address OptoRuntime::handle_exception_C(JavaThread* thread)
2851 //
2852 // n.b. 1 gp arg, 0 fp args, integral return type
2853
2854 // the stack should always be aligned
2855 address the_pc = __ pc();
2856 __ set_last_Java_frame(sp, noreg, the_pc, rscratch1);
2857 __ mov(c_rarg0, rthread);
2858 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
2859 __ bl(rscratch1);
2860 __ maybe_isb();
2861
2862 // Set an oopmap for the call site. This oopmap will only be used if we
2863 // are unwinding the stack. Hence, all locations will be dead.
2864 // Callee-saved registers will be the same as the frame above (i.e.,
2865 // handle_exception_stub), since they were restored when we got the
2866 // exception.
2867
2868 OopMapSet* oop_maps = new OopMapSet();
2869
2870 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2871
2872 __ reset_last_Java_frame(false, true);
2873
2874 // Restore callee-saved registers
2875
2876 // rfp is an implicitly saved callee saved register (i.e. the calling
2877 // convention will save restore it in prolog/epilog) Other than that
2878 // there are no callee save registers now that adapter frames are gone.
2879 // and we dont' expect an arg reg save area
2880 __ ldrd(rfp, r3, Address(__ post(sp, 2 * wordSize)));
2881
2882 // r0: exception handler
2883
2884 // We have a handler in r0 (could be deopt blob).
2885 __ mov(r8, r0);
2886
2887 // Get the exception oop
2888 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2889 // Get the exception pc in case we are deoptimized
2890 __ ldr(r4, Address(rthread, JavaThread::exception_pc_offset()));
2891 __ mov(rscratch1, 0);
2892 #ifdef ASSERT
2893 __ str(rscratch1, Address(rthread, JavaThread::exception_handler_pc_offset()));
2894 __ str(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
2895 #endif
2896 // Clear the exception oop so GC no longer processes it as a root.
2897 __ str(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
2898
2899 // r0: exception oop
2900 // r8: exception handler
2901 // r4: exception pc
2902 // Jump to handler
2903
2904 __ b(r8);
2905
2906 // Make sure all code is generated
2907 masm->flush();
2908
2909 // Set exception blob
2910 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
2911 }
2912 #endif // COMPILER2