Fri, 21 Jan 2011 00:01:41 -0800
7013812: C1: deopt blob too far from patching stub
Summary: Use long jumps to get from patching stubs to deopt blob
Reviewed-by: kvn, never
1 /*
2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "c1/c1_Defs.hpp"
27 #include "c1/c1_MacroAssembler.hpp"
28 #include "c1/c1_Runtime1.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "nativeInst_sparc.hpp"
31 #include "oops/compiledICHolderOop.hpp"
32 #include "oops/oop.inline.hpp"
33 #include "prims/jvmtiExport.hpp"
34 #include "register_sparc.hpp"
35 #include "runtime/sharedRuntime.hpp"
36 #include "runtime/signature.hpp"
37 #include "runtime/vframeArray.hpp"
38 #include "vmreg_sparc.inline.hpp"
40 // Implementation of StubAssembler
42 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) {
43 // for sparc changing the number of arguments doesn't change
44 // anything about the frame size so we'll always lie and claim that
45 // we are only passing 1 argument.
46 set_num_rt_args(1);
48 assert_not_delayed();
49 // bang stack before going to runtime
50 set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
51 st(G0, SP, G3_scratch);
53 // debugging support
54 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
56 set_last_Java_frame(SP, noreg);
57 if (VerifyThread) mov(G2_thread, O0); // about to be smashed; pass early
58 save_thread(L7_thread_cache);
59 // do the call
60 call(entry_point, relocInfo::runtime_call_type);
61 if (!VerifyThread) {
62 delayed()->mov(G2_thread, O0); // pass thread as first argument
63 } else {
64 delayed()->nop(); // (thread already passed)
65 }
66 int call_offset = offset(); // offset of return address
67 restore_thread(L7_thread_cache);
68 reset_last_Java_frame();
70 // check for pending exceptions
71 { Label L;
72 Address exception_addr(G2_thread, Thread::pending_exception_offset());
73 ld_ptr(exception_addr, Gtemp);
74 br_null(Gtemp, false, pt, L);
75 delayed()->nop();
76 Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
77 st_ptr(G0, vm_result_addr);
78 Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
79 st_ptr(G0, vm_result_addr_2);
81 if (frame_size() == no_frame_size) {
82 // we use O7 linkage so that forward_exception_entry has the issuing PC
83 call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
84 delayed()->restore();
85 } else if (_stub_id == Runtime1::forward_exception_id) {
86 should_not_reach_here();
87 } else {
88 AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));
89 jump_to(exc, G4);
90 delayed()->nop();
91 }
92 bind(L);
93 }
95 // get oop result if there is one and reset the value in the thread
96 if (oop_result1->is_valid()) { // get oop result if there is one and reset it in the thread
97 get_vm_result (oop_result1);
98 } else {
99 // be a little paranoid and clear the result
100 Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
101 st_ptr(G0, vm_result_addr);
102 }
104 if (oop_result2->is_valid()) {
105 get_vm_result_2(oop_result2);
106 } else {
107 // be a little paranoid and clear the result
108 Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
109 st_ptr(G0, vm_result_addr_2);
110 }
112 return call_offset;
113 }
116 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
117 // O0 is reserved for the thread
118 mov(arg1, O1);
119 return call_RT(oop_result1, oop_result2, entry, 1);
120 }
123 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
124 // O0 is reserved for the thread
125 mov(arg1, O1);
126 mov(arg2, O2); assert(arg2 != O1, "smashed argument");
127 return call_RT(oop_result1, oop_result2, entry, 2);
128 }
131 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
132 // O0 is reserved for the thread
133 mov(arg1, O1);
134 mov(arg2, O2); assert(arg2 != O1, "smashed argument");
135 mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
136 return call_RT(oop_result1, oop_result2, entry, 3);
137 }
140 // Implementation of Runtime1
142 #define __ sasm->
144 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
145 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
146 static int reg_save_size_in_words;
147 static int frame_size_in_bytes = -1;
149 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
150 assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
151 " mismatch in calculation");
152 sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
153 int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
154 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
156 int i;
157 for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
158 Register r = as_Register(i);
159 if (r == G1 || r == G3 || r == G4 || r == G5) {
160 int sp_offset = cpu_reg_save_offsets[i];
161 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
162 r->as_VMReg());
163 }
164 }
166 if (save_fpu_registers) {
167 for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
168 FloatRegister r = as_FloatRegister(i);
169 int sp_offset = fpu_reg_save_offsets[i];
170 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
171 r->as_VMReg());
172 }
173 }
174 return oop_map;
175 }
177 static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
178 assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
179 " mismatch in calculation");
180 __ save_frame_c1(frame_size_in_bytes);
181 sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
183 // Record volatile registers as callee-save values in an OopMap so their save locations will be
184 // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
185 // deoptimization; see compiledVFrame::create_stack_value). The caller's I, L and O registers
186 // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
187 // (as the stub's I's) when the runtime routine called by the stub creates its frame.
188 // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
190 int i;
191 for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
192 Register r = as_Register(i);
193 if (r == G1 || r == G3 || r == G4 || r == G5) {
194 int sp_offset = cpu_reg_save_offsets[i];
195 __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
196 }
197 }
199 if (save_fpu_registers) {
200 for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
201 FloatRegister r = as_FloatRegister(i);
202 int sp_offset = fpu_reg_save_offsets[i];
203 __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
204 }
205 }
207 return generate_oop_map(sasm, save_fpu_registers);
208 }
210 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
211 for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
212 Register r = as_Register(i);
213 if (r == G1 || r == G3 || r == G4 || r == G5) {
214 __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
215 }
216 }
218 if (restore_fpu_registers) {
219 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
220 FloatRegister r = as_FloatRegister(i);
221 __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
222 }
223 }
224 }
227 void Runtime1::initialize_pd() {
228 // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
229 //
230 // A stub routine will have a frame that is at least large enough to hold
231 // a register window save area (obviously) and the volatile g registers
232 // and floating registers. A user of save_live_registers can have a frame
233 // that has more scratch area in it (although typically they will use L-regs).
234 // in that case the frame will look like this (stack growing down)
235 //
236 // FP -> | |
237 // | scratch mem |
238 // | " " |
239 // --------------
240 // | float regs |
241 // | " " |
242 // ---------------
243 // | G regs |
244 // | " " |
245 // ---------------
246 // | abi reg. |
247 // | window save |
248 // | area |
249 // SP -> ---------------
250 //
251 int i;
252 int sp_offset = round_to(frame::register_save_words, 2); // start doubleword aligned
254 // only G int registers are saved explicitly; others are found in register windows
255 for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
256 Register r = as_Register(i);
257 if (r == G1 || r == G3 || r == G4 || r == G5) {
258 cpu_reg_save_offsets[i] = sp_offset;
259 sp_offset++;
260 }
261 }
263 // all float registers are saved explicitly
264 assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
265 for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
266 fpu_reg_save_offsets[i] = sp_offset;
267 sp_offset++;
268 }
269 reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
270 // this should match assembler::total_frame_size_in_bytes, which
271 // isn't callable from this context. It's checked by an assert when
272 // it's used though.
273 frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
274 }
277 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
278 // make a frame and preserve the caller's caller-save registers
279 OopMap* oop_map = save_live_registers(sasm);
280 int call_offset;
281 if (!has_argument) {
282 call_offset = __ call_RT(noreg, noreg, target);
283 } else {
284 call_offset = __ call_RT(noreg, noreg, target, G4);
285 }
286 OopMapSet* oop_maps = new OopMapSet();
287 oop_maps->add_gc_map(call_offset, oop_map);
289 __ should_not_reach_here();
290 return oop_maps;
291 }
294 OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
295 Register arg1, Register arg2, Register arg3) {
296 // make a frame and preserve the caller's caller-save registers
297 OopMap* oop_map = save_live_registers(sasm);
299 int call_offset;
300 if (arg1 == noreg) {
301 call_offset = __ call_RT(result, noreg, target);
302 } else if (arg2 == noreg) {
303 call_offset = __ call_RT(result, noreg, target, arg1);
304 } else if (arg3 == noreg) {
305 call_offset = __ call_RT(result, noreg, target, arg1, arg2);
306 } else {
307 call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
308 }
309 OopMapSet* oop_maps = NULL;
311 oop_maps = new OopMapSet();
312 oop_maps->add_gc_map(call_offset, oop_map);
313 restore_live_registers(sasm);
315 __ ret();
316 __ delayed()->restore();
318 return oop_maps;
319 }
322 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
323 // make a frame and preserve the caller's caller-save registers
324 OopMap* oop_map = save_live_registers(sasm);
326 // call the runtime patching routine, returns non-zero if nmethod got deopted.
327 int call_offset = __ call_RT(noreg, noreg, target);
328 OopMapSet* oop_maps = new OopMapSet();
329 oop_maps->add_gc_map(call_offset, oop_map);
331 // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
332 // deoptimization handler entry that will cause re-execution of the current bytecode
333 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
334 assert(deopt_blob != NULL, "deoptimization blob must have been created");
336 Label no_deopt;
337 __ tst(O0);
338 __ brx(Assembler::equal, false, Assembler::pt, no_deopt);
339 __ delayed()->nop();
341 // return to the deoptimization handler entry for unpacking and rexecute
342 // if we simply returned the we'd deopt as if any call we patched had just
343 // returned.
345 restore_live_registers(sasm);
347 AddressLiteral dest(deopt_blob->unpack_with_reexecution());
348 __ jump_to(dest, O0);
349 __ delayed()->restore();
351 __ bind(no_deopt);
352 restore_live_registers(sasm);
353 __ ret();
354 __ delayed()->restore();
356 return oop_maps;
357 }
359 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
361 OopMapSet* oop_maps = NULL;
362 // for better readability
363 const bool must_gc_arguments = true;
364 const bool dont_gc_arguments = false;
366 // stub code & info for the different stubs
367 switch (id) {
368 case forward_exception_id:
369 {
370 // we're handling an exception in the context of a compiled
371 // frame. The registers have been saved in the standard
372 // places. Perform an exception lookup in the caller and
373 // dispatch to the handler if found. Otherwise unwind and
374 // dispatch to the callers exception handler.
376 oop_maps = new OopMapSet();
377 OopMap* oop_map = generate_oop_map(sasm, true);
379 // transfer the pending exception to the exception_oop
380 __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
381 __ ld_ptr(Oexception, 0, G0);
382 __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
383 __ add(I7, frame::pc_return_offset, Oissuing_pc);
385 generate_handle_exception(sasm, oop_maps, oop_map);
386 __ should_not_reach_here();
387 }
388 break;
390 case new_instance_id:
391 case fast_new_instance_id:
392 case fast_new_instance_init_check_id:
393 {
394 Register G5_klass = G5; // Incoming
395 Register O0_obj = O0; // Outgoing
397 if (id == new_instance_id) {
398 __ set_info("new_instance", dont_gc_arguments);
399 } else if (id == fast_new_instance_id) {
400 __ set_info("fast new_instance", dont_gc_arguments);
401 } else {
402 assert(id == fast_new_instance_init_check_id, "bad StubID");
403 __ set_info("fast new_instance init check", dont_gc_arguments);
404 }
406 if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
407 UseTLAB && FastTLABRefill) {
408 Label slow_path;
409 Register G1_obj_size = G1;
410 Register G3_t1 = G3;
411 Register G4_t2 = G4;
412 assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
414 // Push a frame since we may do dtrace notification for the
415 // allocation which requires calling out and we don't want
416 // to stomp the real return address.
417 __ save_frame(0);
419 if (id == fast_new_instance_init_check_id) {
420 // make sure the klass is initialized
421 __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);
422 __ cmp(G3_t1, instanceKlass::fully_initialized);
423 __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
424 __ delayed()->nop();
425 }
426 #ifdef ASSERT
427 // assert object can be fast path allocated
428 {
429 Label ok, not_ok;
430 __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);
431 __ cmp(G1_obj_size, 0); // make sure it's an instance (LH > 0)
432 __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);
433 __ delayed()->nop();
434 __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
435 __ br(Assembler::zero, false, Assembler::pn, ok);
436 __ delayed()->nop();
437 __ bind(not_ok);
438 __ stop("assert(can be fast path allocated)");
439 __ should_not_reach_here();
440 __ bind(ok);
441 }
442 #endif // ASSERT
443 // if we got here then the TLAB allocation failed, so try
444 // refilling the TLAB or allocating directly from eden.
445 Label retry_tlab, try_eden;
446 __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
448 __ bind(retry_tlab);
450 // get the instance size
451 __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
453 __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
455 __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
456 __ verify_oop(O0_obj);
457 __ mov(O0, I0);
458 __ ret();
459 __ delayed()->restore();
461 __ bind(try_eden);
462 // get the instance size
463 __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
464 __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
465 __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
467 __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
468 __ verify_oop(O0_obj);
469 __ mov(O0, I0);
470 __ ret();
471 __ delayed()->restore();
473 __ bind(slow_path);
475 // pop this frame so generate_stub_call can push it's own
476 __ restore();
477 }
479 oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
480 // I0->O0: new instance
481 }
483 break;
485 case counter_overflow_id:
486 // G4 contains bci, G5 contains method
487 oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
488 break;
490 case new_type_array_id:
491 case new_object_array_id:
492 {
493 Register G5_klass = G5; // Incoming
494 Register G4_length = G4; // Incoming
495 Register O0_obj = O0; // Outgoing
497 Address klass_lh(G5_klass, ((klassOopDesc::header_size() * HeapWordSize)
498 + Klass::layout_helper_offset_in_bytes()));
499 assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
500 assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
501 // Use this offset to pick out an individual byte of the layout_helper:
502 const int klass_lh_header_size_offset = ((BytesPerInt - 1) // 3 - 2 selects byte {0,1,0,0}
503 - Klass::_lh_header_size_shift / BitsPerByte);
505 if (id == new_type_array_id) {
506 __ set_info("new_type_array", dont_gc_arguments);
507 } else {
508 __ set_info("new_object_array", dont_gc_arguments);
509 }
511 #ifdef ASSERT
512 // assert object type is really an array of the proper kind
513 {
514 Label ok;
515 Register G3_t1 = G3;
516 __ ld(klass_lh, G3_t1);
517 __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
518 int tag = ((id == new_type_array_id)
519 ? Klass::_lh_array_tag_type_value
520 : Klass::_lh_array_tag_obj_value);
521 __ cmp(G3_t1, tag);
522 __ brx(Assembler::equal, false, Assembler::pt, ok);
523 __ delayed()->nop();
524 __ stop("assert(is an array klass)");
525 __ should_not_reach_here();
526 __ bind(ok);
527 }
528 #endif // ASSERT
530 if (UseTLAB && FastTLABRefill) {
531 Label slow_path;
532 Register G1_arr_size = G1;
533 Register G3_t1 = G3;
534 Register O1_t2 = O1;
535 assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
537 // check that array length is small enough for fast path
538 __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
539 __ cmp(G4_length, G3_t1);
540 __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
541 __ delayed()->nop();
543 // if we got here then the TLAB allocation failed, so try
544 // refilling the TLAB or allocating directly from eden.
545 Label retry_tlab, try_eden;
546 __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
548 __ bind(retry_tlab);
550 // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
551 __ ld(klass_lh, G3_t1);
552 __ sll(G4_length, G3_t1, G1_arr_size);
553 __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
554 __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
555 __ add(G1_arr_size, G3_t1, G1_arr_size);
556 __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size); // align up
557 __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
559 __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path); // preserves G1_arr_size
561 __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
562 __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
563 __ sub(G1_arr_size, G3_t1, O1_t2); // body length
564 __ add(O0_obj, G3_t1, G3_t1); // body start
565 __ initialize_body(G3_t1, O1_t2);
566 __ verify_oop(O0_obj);
567 __ retl();
568 __ delayed()->nop();
570 __ bind(try_eden);
571 // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
572 __ ld(klass_lh, G3_t1);
573 __ sll(G4_length, G3_t1, G1_arr_size);
574 __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
575 __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
576 __ add(G1_arr_size, G3_t1, G1_arr_size);
577 __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
578 __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
580 __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path); // preserves G1_arr_size
581 __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
583 __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
584 __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
585 __ sub(G1_arr_size, G3_t1, O1_t2); // body length
586 __ add(O0_obj, G3_t1, G3_t1); // body start
587 __ initialize_body(G3_t1, O1_t2);
588 __ verify_oop(O0_obj);
589 __ retl();
590 __ delayed()->nop();
592 __ bind(slow_path);
593 }
595 if (id == new_type_array_id) {
596 oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
597 } else {
598 oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
599 }
600 // I0 -> O0: new array
601 }
602 break;
604 case new_multi_array_id:
605 { // O0: klass
606 // O1: rank
607 // O2: address of 1st dimension
608 __ set_info("new_multi_array", dont_gc_arguments);
609 oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
610 // I0 -> O0: new multi array
611 }
612 break;
614 case register_finalizer_id:
615 {
616 __ set_info("register_finalizer", dont_gc_arguments);
618 // load the klass and check the has finalizer flag
619 Label register_finalizer;
620 Register t = O1;
621 __ load_klass(O0, t);
622 __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t);
623 __ set(JVM_ACC_HAS_FINALIZER, G3);
624 __ andcc(G3, t, G0);
625 __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
626 __ delayed()->nop();
628 // do a leaf return
629 __ retl();
630 __ delayed()->nop();
632 __ bind(register_finalizer);
633 OopMap* oop_map = save_live_registers(sasm);
634 int call_offset = __ call_RT(noreg, noreg,
635 CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
636 oop_maps = new OopMapSet();
637 oop_maps->add_gc_map(call_offset, oop_map);
639 // Now restore all the live registers
640 restore_live_registers(sasm);
642 __ ret();
643 __ delayed()->restore();
644 }
645 break;
647 case throw_range_check_failed_id:
648 { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
649 // G4: index
650 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
651 }
652 break;
654 case throw_index_exception_id:
655 { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
656 // G4: index
657 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
658 }
659 break;
661 case throw_div0_exception_id:
662 { __ set_info("throw_div0_exception", dont_gc_arguments);
663 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
664 }
665 break;
667 case throw_null_pointer_exception_id:
668 { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
669 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
670 }
671 break;
673 case handle_exception_id:
674 {
675 __ set_info("handle_exception", dont_gc_arguments);
676 // make a frame and preserve the caller's caller-save registers
678 oop_maps = new OopMapSet();
679 OopMap* oop_map = save_live_registers(sasm);
680 __ mov(Oexception->after_save(), Oexception);
681 __ mov(Oissuing_pc->after_save(), Oissuing_pc);
682 generate_handle_exception(sasm, oop_maps, oop_map);
683 }
684 break;
686 case unwind_exception_id:
687 {
688 // O0: exception
689 // I7: address of call to this method
691 __ set_info("unwind_exception", dont_gc_arguments);
692 __ mov(Oexception, Oexception->after_save());
693 __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
695 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
696 G2_thread, Oissuing_pc->after_save());
697 __ verify_not_null_oop(Oexception->after_save());
699 // Restore SP from L7 if the exception PC is a MethodHandle call site.
700 __ mov(O0, G5); // Save the target address.
701 __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
702 __ tst(L0); // Condition codes are preserved over the restore.
703 __ restore();
705 __ jmp(G5, 0);
706 __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP); // Restore SP if required.
707 }
708 break;
710 case throw_array_store_exception_id:
711 {
712 __ set_info("throw_array_store_exception", dont_gc_arguments);
713 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false);
714 }
715 break;
717 case throw_class_cast_exception_id:
718 {
719 // G4: object
720 __ set_info("throw_class_cast_exception", dont_gc_arguments);
721 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
722 }
723 break;
725 case throw_incompatible_class_change_error_id:
726 {
727 __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
728 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
729 }
730 break;
732 case slow_subtype_check_id:
733 { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
734 // Arguments :
735 //
736 // ret : G3
737 // sub : G3, argument, destroyed
738 // super: G1, argument, not changed
739 // raddr: O7, blown by call
740 Label miss;
742 __ save_frame(0); // Blow no registers!
744 __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
746 __ mov(1, G3);
747 __ ret(); // Result in G5 is 'true'
748 __ delayed()->restore(); // free copy or add can go here
750 __ bind(miss);
751 __ mov(0, G3);
752 __ ret(); // Result in G5 is 'false'
753 __ delayed()->restore(); // free copy or add can go here
754 }
756 case monitorenter_nofpu_id:
757 case monitorenter_id:
758 { // G4: object
759 // G5: lock address
760 __ set_info("monitorenter", dont_gc_arguments);
762 int save_fpu_registers = (id == monitorenter_id);
763 // make a frame and preserve the caller's caller-save registers
764 OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
766 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
768 oop_maps = new OopMapSet();
769 oop_maps->add_gc_map(call_offset, oop_map);
770 restore_live_registers(sasm, save_fpu_registers);
772 __ ret();
773 __ delayed()->restore();
774 }
775 break;
777 case monitorexit_nofpu_id:
778 case monitorexit_id:
779 { // G4: lock address
780 // note: really a leaf routine but must setup last java sp
781 // => use call_RT for now (speed can be improved by
782 // doing last java sp setup manually)
783 __ set_info("monitorexit", dont_gc_arguments);
785 int save_fpu_registers = (id == monitorexit_id);
786 // make a frame and preserve the caller's caller-save registers
787 OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
789 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
791 oop_maps = new OopMapSet();
792 oop_maps->add_gc_map(call_offset, oop_map);
793 restore_live_registers(sasm, save_fpu_registers);
795 __ ret();
796 __ delayed()->restore();
798 }
799 break;
801 case access_field_patching_id:
802 { __ set_info("access_field_patching", dont_gc_arguments);
803 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
804 }
805 break;
807 case load_klass_patching_id:
808 { __ set_info("load_klass_patching", dont_gc_arguments);
809 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
810 }
811 break;
813 case jvmti_exception_throw_id:
814 { // Oexception : exception
815 __ set_info("jvmti_exception_throw", dont_gc_arguments);
816 oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0);
817 }
818 break;
820 case dtrace_object_alloc_id:
821 { // O0: object
822 __ set_info("dtrace_object_alloc", dont_gc_arguments);
823 // we can't gc here so skip the oopmap but make sure that all
824 // the live registers get saved.
825 save_live_registers(sasm);
827 __ save_thread(L7_thread_cache);
828 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
829 relocInfo::runtime_call_type);
830 __ delayed()->mov(I0, O0);
831 __ restore_thread(L7_thread_cache);
833 restore_live_registers(sasm);
834 __ ret();
835 __ delayed()->restore();
836 }
837 break;
839 #ifndef SERIALGC
840 case g1_pre_barrier_slow_id:
841 { // G4: previous value of memory
842 BarrierSet* bs = Universe::heap()->barrier_set();
843 if (bs->kind() != BarrierSet::G1SATBCTLogging) {
844 __ save_frame(0);
845 __ set((int)id, O1);
846 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
847 __ should_not_reach_here();
848 break;
849 }
851 __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
853 Register pre_val = G4;
854 Register tmp = G1_scratch;
855 Register tmp2 = G3_scratch;
857 Label refill, restart;
858 bool with_frame = false; // I don't know if we can do with-frame.
859 int satb_q_index_byte_offset =
860 in_bytes(JavaThread::satb_mark_queue_offset() +
861 PtrQueue::byte_offset_of_index());
862 int satb_q_buf_byte_offset =
863 in_bytes(JavaThread::satb_mark_queue_offset() +
864 PtrQueue::byte_offset_of_buf());
865 __ bind(restart);
866 __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
868 __ br_on_reg_cond(Assembler::rc_z, /*annul*/false,
869 Assembler::pn, tmp, refill);
871 // If the branch is taken, no harm in executing this in the delay slot.
872 __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
873 __ sub(tmp, oopSize, tmp);
875 __ st_ptr(pre_val, tmp2, tmp); // [_buf + index] := <address_of_card>
876 // Use return-from-leaf
877 __ retl();
878 __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
880 __ bind(refill);
881 __ save_frame(0);
883 __ mov(pre_val, L0);
884 __ mov(tmp, L1);
885 __ mov(tmp2, L2);
887 __ call_VM_leaf(L7_thread_cache,
888 CAST_FROM_FN_PTR(address,
889 SATBMarkQueueSet::handle_zero_index_for_thread),
890 G2_thread);
892 __ mov(L0, pre_val);
893 __ mov(L1, tmp);
894 __ mov(L2, tmp2);
896 __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
897 __ delayed()->restore();
898 }
899 break;
901 case g1_post_barrier_slow_id:
902 {
903 BarrierSet* bs = Universe::heap()->barrier_set();
904 if (bs->kind() != BarrierSet::G1SATBCTLogging) {
905 __ save_frame(0);
906 __ set((int)id, O1);
907 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
908 __ should_not_reach_here();
909 break;
910 }
912 __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
914 Register addr = G4;
915 Register cardtable = G5;
916 Register tmp = G1_scratch;
917 Register tmp2 = G3_scratch;
918 jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
920 Label not_already_dirty, restart, refill;
922 #ifdef _LP64
923 __ srlx(addr, CardTableModRefBS::card_shift, addr);
924 #else
925 __ srl(addr, CardTableModRefBS::card_shift, addr);
926 #endif
928 AddressLiteral rs(byte_map_base);
929 __ set(rs, cardtable); // cardtable := <card table base>
930 __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
932 __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
933 tmp, not_already_dirty);
934 // Get cardtable + tmp into a reg by itself -- useful in the take-the-branch
935 // case, harmless if not.
936 __ delayed()->add(addr, cardtable, tmp2);
938 // We didn't take the branch, so we're already dirty: return.
939 // Use return-from-leaf
940 __ retl();
941 __ delayed()->nop();
943 // Not dirty.
944 __ bind(not_already_dirty);
945 // First, dirty it.
946 __ stb(G0, tmp2, 0); // [cardPtr] := 0 (i.e., dirty).
948 Register tmp3 = cardtable;
949 Register tmp4 = tmp;
951 // these registers are now dead
952 addr = cardtable = tmp = noreg;
954 int dirty_card_q_index_byte_offset =
955 in_bytes(JavaThread::dirty_card_queue_offset() +
956 PtrQueue::byte_offset_of_index());
957 int dirty_card_q_buf_byte_offset =
958 in_bytes(JavaThread::dirty_card_queue_offset() +
959 PtrQueue::byte_offset_of_buf());
960 __ bind(restart);
961 __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
963 __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
964 tmp3, refill);
965 // If the branch is taken, no harm in executing this in the delay slot.
966 __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
967 __ sub(tmp3, oopSize, tmp3);
969 __ st_ptr(tmp2, tmp4, tmp3); // [_buf + index] := <address_of_card>
970 // Use return-from-leaf
971 __ retl();
972 __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
974 __ bind(refill);
975 __ save_frame(0);
977 __ mov(tmp2, L0);
978 __ mov(tmp3, L1);
979 __ mov(tmp4, L2);
981 __ call_VM_leaf(L7_thread_cache,
982 CAST_FROM_FN_PTR(address,
983 DirtyCardQueueSet::handle_zero_index_for_thread),
984 G2_thread);
986 __ mov(L0, tmp2);
987 __ mov(L1, tmp3);
988 __ mov(L2, tmp4);
990 __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
991 __ delayed()->restore();
992 }
993 break;
994 #endif // !SERIALGC
996 default:
997 { __ set_info("unimplemented entry", dont_gc_arguments);
998 __ save_frame(0);
999 __ set((int)id, O1);
1000 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
1001 __ should_not_reach_here();
1002 }
1003 break;
1004 }
1005 return oop_maps;
1006 }
1009 void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) {
1010 Label no_deopt;
1012 __ verify_not_null_oop(Oexception);
1014 // save the exception and issuing pc in the thread
1015 __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1016 __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1018 // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map)
1019 __ mov(I7, L0);
1020 __ mov(Oissuing_pc, I7);
1021 __ sub(I7, frame::pc_return_offset, I7);
1022 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1024 // Note: if nmethod has been deoptimized then regardless of
1025 // whether it had a handler or not we will deoptimize
1026 // by entering the deopt blob with a pending exception.
1028 #ifdef ASSERT
1029 Label done;
1030 __ tst(O0);
1031 __ br(Assembler::notZero, false, Assembler::pn, done);
1032 __ delayed()->nop();
1033 __ stop("should have found address");
1034 __ bind(done);
1035 #endif
1037 // restore the registers that were saved at the beginning and jump to the exception handler.
1038 restore_live_registers(sasm);
1040 __ jmp(O0, 0);
1041 __ delayed()->restore();
1043 oop_maps->add_gc_map(call_offset, oop_map);
1044 }
1047 #undef __
1049 #define __ masm->
1051 const char *Runtime1::pd_name_for_address(address entry) {
1052 return "<unknown function>";
1053 }