1 /*
2 * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2022, Huawei Technologies Co., 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/assembler.hpp"
29 #include "c1/c1_CodeStubs.hpp"
30 #include "c1/c1_Defs.hpp"
31 #include "c1/c1_MacroAssembler.hpp"
32 #include "c1/c1_Runtime1.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "compiler/oopMap.hpp"
35 #include "gc/shared/cardTable.hpp"
36 #include "gc/shared/cardTableBarrierSet.hpp"
37 #include "interpreter/interpreter.hpp"
38 #include "memory/universe.hpp"
39 #include "nativeInst_riscv.hpp"
40 #include "oops/compiledICHolder.hpp"
41 #include "oops/oop.inline.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "register_riscv.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/signature.hpp"
46 #include "runtime/stubRoutines.hpp"
47 #include "runtime/vframe.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/powerOfTwo.hpp"
50 #include "vmreg_riscv.inline.hpp"
51
52
53 // Implementation of StubAssembler
54
55 int StubAssembler::call_RT(Register oop_result, Register metadata_result, address entry, int args_size) {
56 // setup registers
57 assert(!(oop_result->is_valid() || metadata_result->is_valid()) || oop_result != metadata_result,
58 "registers must be different");
59 assert(oop_result != xthread && metadata_result != xthread, "registers must be different");
60 assert(args_size >= 0, "illegal args_size");
61 bool align_stack = false;
62
63 mv(c_rarg0, xthread);
64 set_num_rt_args(0); // Nothing on stack
65
66 Label retaddr;
67 set_last_Java_frame(sp, fp, retaddr, t0);
68
69 // do the call
70 int32_t off = 0;
71 la_patchable(t0, RuntimeAddress(entry), off);
72 jalr(x1, t0, off);
73 bind(retaddr);
74 int call_offset = offset();
75 // verify callee-saved register
76 #ifdef ASSERT
77 push_reg(x10, sp);
78 { Label L;
79 get_thread(x10);
80 beq(xthread, x10, L);
81 stop("StubAssembler::call_RT: xthread not callee saved?");
82 bind(L);
83 }
84 pop_reg(x10, sp);
85 #endif
86 reset_last_Java_frame(true);
87
88 // check for pending exceptions
89 { Label L;
90 // check for pending exceptions (java_thread is set upon return)
91 ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
92 beqz(t0, L);
93 // exception pending => remove activation and forward to exception handler
94 // make sure that the vm_results are cleared
95 if (oop_result->is_valid()) {
96 sd(zr, Address(xthread, JavaThread::vm_result_offset()));
97 }
98 if (metadata_result->is_valid()) {
99 sd(zr, Address(xthread, JavaThread::vm_result_2_offset()));
100 }
101 if (frame_size() == no_frame_size) {
102 leave();
103 far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
104 } else if (_stub_id == Runtime1::forward_exception_id) {
105 should_not_reach_here();
106 } else {
107 far_jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
108 }
109 bind(L);
110 }
111 // get oop results if there are any and reset the values in the thread
112 if (oop_result->is_valid()) {
113 get_vm_result(oop_result, xthread);
114 }
115 if (metadata_result->is_valid()) {
116 get_vm_result_2(metadata_result, xthread);
117 }
118 return call_offset;
119 }
120
121 int StubAssembler::call_RT(Register oop_result, Register metadata_result, address entry, Register arg1) {
122 mv(c_rarg1, arg1);
123 return call_RT(oop_result, metadata_result, entry, 1);
124 }
125
126 int StubAssembler::call_RT(Register oop_result, Register metadata_result, address entry, Register arg1, Register arg2) {
127 const int arg_num = 2;
128 if (c_rarg1 == arg2) {
129 if (c_rarg2 == arg1) {
130 xorr(arg1, arg1, arg2);
131 xorr(arg2, arg1, arg2);
132 xorr(arg1, arg1, arg2);
133 } else {
134 mv(c_rarg2, arg2);
135 mv(c_rarg1, arg1);
136 }
137 } else {
138 mv(c_rarg1, arg1);
139 mv(c_rarg2, arg2);
140 }
141 return call_RT(oop_result, metadata_result, entry, arg_num);
142 }
143
144 int StubAssembler::call_RT(Register oop_result, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
145 const int arg_num = 3;
146 // if there is any conflict use the stack
147 if (arg1 == c_rarg2 || arg1 == c_rarg3 ||
148 arg2 == c_rarg1 || arg2 == c_rarg3 ||
149 arg3 == c_rarg1 || arg3 == c_rarg2) {
150 const int arg1_sp_offset = 0;
151 const int arg2_sp_offset = 1;
152 const int arg3_sp_offset = 2;
153 addi(sp, sp, -(arg_num + 1) * wordSize);
154 sd(arg1, Address(sp, arg1_sp_offset * wordSize));
155 sd(arg2, Address(sp, arg2_sp_offset * wordSize));
156 sd(arg3, Address(sp, arg3_sp_offset * wordSize));
157
158 ld(c_rarg1, Address(sp, arg1_sp_offset * wordSize));
159 ld(c_rarg2, Address(sp, arg2_sp_offset * wordSize));
160 ld(c_rarg3, Address(sp, arg3_sp_offset * wordSize));
161 addi(sp, sp, (arg_num + 1) * wordSize);
162 } else {
163 mv(c_rarg1, arg1);
164 mv(c_rarg2, arg2);
165 mv(c_rarg3, arg3);
166 }
167 return call_RT(oop_result, metadata_result, entry, arg_num);
168 }
169
170 enum return_state_t {
171 does_not_return, requires_return
172 };
173
174 // Implementation of StubFrame
175
176 class StubFrame: public StackObj {
177 private:
178 StubAssembler* _sasm;
179 bool _return_state;
180
181 public:
182 StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, return_state_t return_state=requires_return);
183 void load_argument(int offset_in_words, Register reg);
184
185 ~StubFrame();
186 };;
187
188 void StubAssembler::prologue(const char* name, bool must_gc_arguments) {
189 set_info(name, must_gc_arguments);
190 enter();
191 }
192
193 void StubAssembler::epilogue() {
194 leave();
195 ret();
196 }
197
198 #define __ _sasm->
199
200 StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, return_state_t return_state) {
201 _sasm = sasm;
202 _return_state = return_state;
203 __ prologue(name, must_gc_arguments);
204 }
205
206 // load parameters that were stored with LIR_Assembler::store_parameter
207 // Note: offsets for store_parameter and load_argument must match
208 void StubFrame::load_argument(int offset_in_words, Register reg) {
209 __ load_parameter(offset_in_words, reg);
210 }
211
212
213 StubFrame::~StubFrame() {
214 if (_return_state == requires_return) {
215 __ epilogue();
216 } else {
217 __ should_not_reach_here();
218 }
219 _sasm = NULL;
220 }
221
222 #undef __
223
224
225 // Implementation of Runtime1
226
227 #define __ sasm->
228
229 const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2;
230
231 // Stack layout for saving/restoring all the registers needed during a runtime
232 // call (this includes deoptimization)
233 // Note: note that users of this frame may well have arguments to some runtime
234 // while these values are on the stack. These positions neglect those arguments
235 // but the code in save_live_registers will take the argument count into
236 // account.
237 //
238
239 enum reg_save_layout {
240 reg_save_frame_size = 32 /* float */ + 30 /* integer excluding x3, x4 */
241 };
242
243 // Save off registers which might be killed by calls into the runtime.
244 // Tries to smart of about FPU registers. In particular we separate
245 // saving and describing the FPU registers for deoptimization since we
246 // have to save the FPU registers twice if we describe them. The
247 // deopt blob is the only thing which needs to describe FPU registers.
248 // In all other cases it should be sufficient to simply save their
249 // current value.
250
251 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
252 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
253
254 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
255 int frame_size_in_bytes = reg_save_frame_size * BytesPerWord;
256 sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
257 int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
258 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
259 assert_cond(oop_map != NULL);
260
261 // caller save registers only, see FrameMap::initialize
262 // in c1_FrameMap_riscv.cpp for detail.
263 const static Register caller_save_cpu_regs[FrameMap::max_nof_caller_save_cpu_regs] = {
264 x7, x10, x11, x12, x13, x14, x15, x16, x17, x28, x29, x30, x31
265 };
266
267 for (int i = 0; i < FrameMap::max_nof_caller_save_cpu_regs; i++) {
268 Register r = caller_save_cpu_regs[i];
269 int sp_offset = cpu_reg_save_offsets[r->encoding()];
270 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
271 r->as_VMReg());
272 }
273
274 // fpu_regs
275 if (save_fpu_registers) {
276 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
277 FloatRegister r = as_FloatRegister(i);
278 int sp_offset = fpu_reg_save_offsets[i];
279 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
280 r->as_VMReg());
281 }
282 }
283 return oop_map;
284 }
285
286 static OopMap* save_live_registers(StubAssembler* sasm,
287 bool save_fpu_registers = true) {
288 __ block_comment("save_live_registers");
289
290 // if the number of pushed regs is odd, one slot will be reserved for alignment
291 __ push_reg(RegSet::range(x5, x31), sp); // integer registers except ra(x1) & sp(x2) & gp(x3) & tp(x4)
292
293 if (save_fpu_registers) {
294 // float registers
295 __ addi(sp, sp, -(FrameMap::nof_fpu_regs * wordSize));
296 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
297 __ fsd(as_FloatRegister(i), Address(sp, i * wordSize));
298 }
299 } else {
300 // we define reg_save_layout = 62 as the fixed frame size,
301 // we should also sub 32 * wordSize to sp when save_fpu_registers == false
302 __ addi(sp, sp, -32 * wordSize);
303 }
304
305 return generate_oop_map(sasm, save_fpu_registers);
306 }
307
308 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
309 if (restore_fpu_registers) {
310 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
311 __ fld(as_FloatRegister(i), Address(sp, i * wordSize));
312 }
313 __ addi(sp, sp, FrameMap::nof_fpu_regs * wordSize);
314 } else {
315 // we define reg_save_layout = 64 as the fixed frame size,
316 // we should also add 32 * wordSize to sp when save_fpu_registers == false
317 __ addi(sp, sp, 32 * wordSize);
318 }
319
320 // if the number of popped regs is odd, the reserved slot for alignment will be removed
321 __ pop_reg(RegSet::range(x5, x31), sp); // integer registers except ra(x1) & sp(x2) & gp(x3) & tp(x4)
322 }
323
324 static void restore_live_registers_except_r10(StubAssembler* sasm, bool restore_fpu_registers = true) {
325 if (restore_fpu_registers) {
326 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
327 __ fld(as_FloatRegister(i), Address(sp, i * wordSize));
328 }
329 __ addi(sp, sp, FrameMap::nof_fpu_regs * wordSize);
330 } else {
331 // we define reg_save_layout = 64 as the fixed frame size,
332 // we should also add 32 * wordSize to sp when save_fpu_registers == false
333 __ addi(sp, sp, 32 * wordSize);
334 }
335
336 // pop integer registers except ra(x1) & sp(x2) & gp(x3) & tp(x4) & x10
337 // there is one reserved slot for alignment on the stack in save_live_registers().
338 __ pop_reg(RegSet::range(x5, x9), sp); // pop x5 ~ x9 with the reserved slot for alignment
339 __ pop_reg(RegSet::range(x11, x31), sp); // pop x11 ~ x31; x10 will be automatically skipped here
340 }
341
342 void Runtime1::initialize_pd() {
343 int i = 0;
344 int sp_offset = 0;
345 const int step = 2; // SP offsets are in halfwords
346
347 // all float registers are saved explicitly
348 for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
349 fpu_reg_save_offsets[i] = sp_offset;
350 sp_offset += step;
351 }
352
353 // a slot reserved for stack 16-byte alignment, see MacroAssembler::push_reg
354 sp_offset += step;
355 // we save x5 ~ x31, except x0 ~ x4: loop starts from x5
356 for (i = 5; i < FrameMap::nof_cpu_regs; i++) {
357 cpu_reg_save_offsets[i] = sp_offset;
358 sp_offset += step;
359 }
360 }
361
362 // target: the entry point of the method that creates and posts the exception oop
363 // has_argument: true if the exception needs arguments (passed in t0 and t1)
364
365 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
366 // make a frame and preserve the caller's caller-save registers
367 OopMap* oop_map = save_live_registers(sasm);
368 assert_cond(oop_map != NULL);
369 int call_offset = 0;
370 if (!has_argument) {
371 call_offset = __ call_RT(noreg, noreg, target);
372 } else {
373 __ mv(c_rarg1, t0);
374 __ mv(c_rarg2, t1);
375 call_offset = __ call_RT(noreg, noreg, target);
376 }
377 OopMapSet* oop_maps = new OopMapSet();
378 assert_cond(oop_maps != NULL);
379 oop_maps->add_gc_map(call_offset, oop_map);
380
381 return oop_maps;
382 }
383
384 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
385 __ block_comment("generate_handle_exception");
386
387 // incoming parameters
388 const Register exception_oop = x10;
389 const Register exception_pc = x13;
390
391 OopMapSet* oop_maps = new OopMapSet();
392 assert_cond(oop_maps != NULL);
393 OopMap* oop_map = NULL;
394
395 switch (id) {
396 case forward_exception_id:
397 // We're handling an exception in the context of a compiled frame.
398 // The registers have been saved in the standard places. Perform
399 // an exception lookup in the caller and dispatch to the handler
400 // if found. Otherwise unwind and dispatch to the callers
401 // exception handler.
402 oop_map = generate_oop_map(sasm, 1 /* thread */);
403
404 // load and clear pending exception oop into x10
405 __ ld(exception_oop, Address(xthread, Thread::pending_exception_offset()));
406 __ sd(zr, Address(xthread, Thread::pending_exception_offset()));
407
408 // load issuing PC (the return address for this stub) into x13
409 __ ld(exception_pc, Address(fp, frame::return_addr_offset * BytesPerWord));
410
411 // make sure that the vm_results are cleared (may be unnecessary)
412 __ sd(zr, Address(xthread, JavaThread::vm_result_offset()));
413 __ sd(zr, Address(xthread, JavaThread::vm_result_2_offset()));
414 break;
415 case handle_exception_nofpu_id:
416 case handle_exception_id:
417 // At this point all registers MAY be live.
418 oop_map = save_live_registers(sasm, id != handle_exception_nofpu_id);
419 break;
420 case handle_exception_from_callee_id: {
421 // At this point all registers except exception oop (x10) and
422 // exception pc (ra) are dead.
423 const int frame_size = 2 /* fp, return address */;
424 oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
425 sasm->set_frame_size(frame_size);
426 break;
427 }
428 default: ShouldNotReachHere();
429 }
430
431 // verify that only x10 and x13 are valid at this time
432 __ invalidate_registers(false, true, true, false, true, true);
433 // verify that x10 contains a valid exception
434 __ verify_not_null_oop(exception_oop);
435
436 #ifdef ASSERT
437 // check that fields in JavaThread for exception oop and issuing pc are
438 // empty before writing to them
439 Label oop_empty;
440 __ ld(t0, Address(xthread, JavaThread::exception_oop_offset()));
441 __ beqz(t0, oop_empty);
442 __ stop("exception oop already set");
443 __ bind(oop_empty);
444
445 Label pc_empty;
446 __ ld(t0, Address(xthread, JavaThread::exception_pc_offset()));
447 __ beqz(t0, pc_empty);
448 __ stop("exception pc already set");
449 __ bind(pc_empty);
450 #endif
451
452 // save exception oop and issuing pc into JavaThread
453 // (exception handler will load it from here)
454 __ sd(exception_oop, Address(xthread, JavaThread::exception_oop_offset()));
455 __ sd(exception_pc, Address(xthread, JavaThread::exception_pc_offset()));
456
457 // patch throwing pc into return address (has bci & oop map)
458 __ sd(exception_pc, Address(fp, frame::return_addr_offset * BytesPerWord));
459
460 // compute the exception handler.
461 // the exception oop and the throwing pc are read from the fields in JavaThread
462 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
463 guarantee(oop_map != NULL, "NULL oop_map!");
464 oop_maps->add_gc_map(call_offset, oop_map);
465
466 // x10: handler address
467 // will be the deopt blob if nmethod was deoptimized while we looked up
468 // handler regardless of whether handler existed in the nmethod.
469
470 // only x10 is valid at this time, all other registers have been destroyed by the runtime call
471 __ invalidate_registers(false, true, true, true, true, true);
472
473 // patch the return address, this stub will directly return to the exception handler
474 __ sd(x10, Address(fp, frame::return_addr_offset * BytesPerWord));
475
476 switch (id) {
477 case forward_exception_id:
478 case handle_exception_nofpu_id:
479 case handle_exception_id:
480 // Restore the registers that were saved at the beginning.
481 restore_live_registers(sasm, id != handle_exception_nofpu_id);
482 break;
483 case handle_exception_from_callee_id:
484 break;
485 default: ShouldNotReachHere();
486 }
487
488 return oop_maps;
489 }
490
491
492 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
493 // incoming parameters
494 const Register exception_oop = x10;
495 // other registers used in this stub
496 const Register handler_addr = x11;
497
498 // verify that only x10, is valid at this time
499 __ invalidate_registers(false, true, true, true, true, true);
500
501 #ifdef ASSERT
502 // check that fields in JavaThread for exception oop and issuing pc are empty
503 Label oop_empty;
504 __ ld(t0, Address(xthread, JavaThread::exception_oop_offset()));
505 __ beqz(t0, oop_empty);
506 __ stop("exception oop must be empty");
507 __ bind(oop_empty);
508
509 Label pc_empty;
510 __ ld(t0, Address(xthread, JavaThread::exception_pc_offset()));
511 __ beqz(t0, pc_empty);
512 __ stop("exception pc must be empty");
513 __ bind(pc_empty);
514 #endif
515
516 // Save our return address because
517 // exception_handler_for_return_address will destroy it. We also
518 // save exception_oop
519 __ addi(sp, sp, -2 * wordSize);
520 __ sd(exception_oop, Address(sp, wordSize));
521 __ sd(ra, Address(sp));
522
523 // search the exception handler address of the caller (using the return address)
524 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), xthread, ra);
525 // x10: exception handler address of the caller
526
527 // Only x10 is valid at this time; all other registers have been
528 // destroyed by the call.
529 __ invalidate_registers(false, true, true, true, false, true);
530
531 // move result of call into correct register
532 __ mv(handler_addr, x10);
533
534 // get throwing pc (= return address).
535 // ra has been destroyed by the call
536 __ ld(ra, Address(sp));
537 __ ld(exception_oop, Address(sp, wordSize));
538 __ addi(sp, sp, 2 * wordSize);
539 __ mv(x13, ra);
540
541 __ verify_not_null_oop(exception_oop);
542
543 // continue at exception handler (return address removed)
544 // note: do *not* remove arguments when unwinding the
545 // activation since the caller assumes having
546 // all arguments on the stack when entering the
547 // runtime to determine the exception handler
548 // (GC happens at call site with arguments!)
549 // x10: exception oop
550 // x13: throwing pc
551 // x11: exception handler
552 __ jr(handler_addr);
553 }
554
555 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
556 // use the maximum number of runtime-arguments here because it is difficult to
557 // distinguish each RT-Call.
558 // Note: This number affects also the RT-Call in generate_handle_exception because
559 // the oop-map is shared for all calls.
560 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
561 assert(deopt_blob != NULL, "deoptimization blob must have been created");
562
563 OopMap* oop_map = save_live_registers(sasm);
564 assert_cond(oop_map != NULL);
565
566 __ mv(c_rarg0, xthread);
567 Label retaddr;
568 __ set_last_Java_frame(sp, fp, retaddr, t0);
569 // do the call
570 int32_t off = 0;
571 __ la_patchable(t0, RuntimeAddress(target), off);
572 __ jalr(x1, t0, off);
573 __ bind(retaddr);
574 OopMapSet* oop_maps = new OopMapSet();
575 assert_cond(oop_maps != NULL);
576 oop_maps->add_gc_map(__ offset(), oop_map);
577 // verify callee-saved register
578 #ifdef ASSERT
579 { Label L;
580 __ get_thread(t0);
581 __ beq(xthread, t0, L);
582 __ stop("StubAssembler::call_RT: xthread not callee saved?");
583 __ bind(L);
584 }
585 #endif
586 __ reset_last_Java_frame(true);
587
588 #ifdef ASSERT
589 // Check that fields in JavaThread for exception oop and issuing pc are empty
590 Label oop_empty;
591 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
592 __ beqz(t0, oop_empty);
593 __ stop("exception oop must be empty");
594 __ bind(oop_empty);
595
596 Label pc_empty;
597 __ ld(t0, Address(xthread, JavaThread::exception_pc_offset()));
598 __ beqz(t0, pc_empty);
599 __ stop("exception pc must be empty");
600 __ bind(pc_empty);
601 #endif
602
603 // Runtime will return true if the nmethod has been deoptimized, this is the
604 // expected scenario and anything else is an error. Note that we maintain a
605 // check on the result purely as a defensive measure.
606 Label no_deopt;
607 __ beqz(x10, no_deopt); // Have we deoptimized?
608
609 // Perform a re-execute. The proper return address is already on the stack,
610 // we just need to restore registers, pop all of our frames but the return
611 // address and jump to the deopt blob.
612
613 restore_live_registers(sasm);
614 __ leave();
615 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
616
617 __ bind(no_deopt);
618 __ stop("deopt not performed");
619
620 return oop_maps;
621 }
622
623 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
624 // for better readability
625 const bool dont_gc_arguments = false;
626
627 // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
628 bool save_fpu_registers = true;
629
630 // stub code & info for the different stubs
631 OopMapSet* oop_maps = NULL;
632 switch (id) {
633 {
634 case forward_exception_id:
635 {
636 oop_maps = generate_handle_exception(id, sasm);
637 __ leave();
638 __ ret();
639 }
640 break;
641
642 case throw_div0_exception_id:
643 {
644 StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments, does_not_return);
645 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
646 }
647 break;
648
649 case throw_null_pointer_exception_id:
650 { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments, does_not_return);
651 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
652 }
653 break;
654
655 case new_instance_id:
656 case fast_new_instance_id:
657 case fast_new_instance_init_check_id:
658 {
659 Register klass = x13; // Incoming
660 Register obj = x10; // Result
661
662 if (id == new_instance_id) {
663 __ set_info("new_instance", dont_gc_arguments);
664 } else if (id == fast_new_instance_id) {
665 __ set_info("fast new_instance", dont_gc_arguments);
666 } else {
667 assert(id == fast_new_instance_init_check_id, "bad StubID");
668 __ set_info("fast new_instance init check", dont_gc_arguments);
669 }
670
671 // If TLAB is disabled, see if there is support for inlining contiguous
672 // allocations.
673 // Otherwise, just go to the slow path.
674 if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
675 !UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
676 Label slow_path;
677 Register obj_size = x12;
678 Register tmp1 = x9;
679 Register tmp2 = x14;
680 assert_different_registers(klass, obj, obj_size, tmp1, tmp2);
681
682 const int sp_offset = 2;
683 const int x9_offset = 1;
684 const int zr_offset = 0;
685 __ addi(sp, sp, -(sp_offset * wordSize));
686 __ sd(x9, Address(sp, x9_offset * wordSize));
687 __ sd(zr, Address(sp, zr_offset * wordSize));
688
689 if (id == fast_new_instance_init_check_id) {
690 // make sure the klass is initialized
691 __ lbu(t0, Address(klass, InstanceKlass::init_state_offset()));
692 __ mv(t1, InstanceKlass::fully_initialized);
693 __ bne(t0, t1, slow_path);
694 }
695
696 #ifdef ASSERT
697 // assert object can be fast path allocated
698 {
699 Label ok, not_ok;
700 __ lw(obj_size, Address(klass, Klass::layout_helper_offset()));
701 // make sure it's an instance. For instances, layout helper is a positive number.
702 // For arrays, layout helper is a negative number
703 __ blez(obj_size, not_ok);
704 __ andi(t0, obj_size, Klass::_lh_instance_slow_path_bit);
705 __ beqz(t0, ok);
706 __ bind(not_ok);
707 __ stop("assert(can be fast path allocated)");
708 __ should_not_reach_here();
709 __ bind(ok);
710 }
711 #endif // ASSERT
712
713 // get the instance size
714 __ lwu(obj_size, Address(klass, Klass::layout_helper_offset()));
715
716 __ eden_allocate(obj, obj_size, 0, tmp1, slow_path);
717
718 __ initialize_object(obj, klass, obj_size, 0, tmp1, tmp2, /* is_tlab_allocated */ false);
719 __ verify_oop(obj);
720 __ ld(x9, Address(sp, x9_offset * wordSize));
721 __ ld(zr, Address(sp, zr_offset * wordSize));
722 __ addi(sp, sp, sp_offset * wordSize);
723 __ ret();
724
725 __ bind(slow_path);
726 __ ld(x9, Address(sp, x9_offset * wordSize));
727 __ ld(zr, Address(sp, zr_offset * wordSize));
728 __ addi(sp, sp, sp_offset * wordSize);
729 }
730
731 __ enter();
732 OopMap* map = save_live_registers(sasm);
733 assert_cond(map != NULL);
734 int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
735 oop_maps = new OopMapSet();
736 assert_cond(oop_maps != NULL);
737 oop_maps->add_gc_map(call_offset, map);
738 restore_live_registers_except_r10(sasm);
739 __ verify_oop(obj);
740 __ leave();
741 __ ret();
742
743 // x10: new instance
744 }
745
746 break;
747
748 case counter_overflow_id:
749 {
750 Register bci = x10;
751 Register method = x11;
752 __ enter();
753 OopMap* map = save_live_registers(sasm);
754 assert_cond(map != NULL);
755
756 const int bci_off = 0;
757 const int method_off = 1;
758 // Retrieve bci
759 __ lw(bci, Address(fp, bci_off * BytesPerWord));
760 // And a pointer to the Method*
761 __ ld(method, Address(fp, method_off * BytesPerWord));
762 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
763 oop_maps = new OopMapSet();
764 assert_cond(oop_maps != NULL);
765 oop_maps->add_gc_map(call_offset, map);
766 restore_live_registers(sasm);
767 __ leave();
768 __ ret();
769 }
770 break;
771
772 case new_type_array_id:
773 case new_object_array_id:
774 {
775 Register length = x9; // Incoming
776 Register klass = x13; // Incoming
777 Register obj = x10; // Result
778
779 if (id == new_type_array_id) {
780 __ set_info("new_type_array", dont_gc_arguments);
781 } else {
782 __ set_info("new_object_array", dont_gc_arguments);
783 }
784
785 #ifdef ASSERT
786 // assert object type is really an array of the proper kind
787 {
788 Label ok;
789 Register tmp = obj;
790 __ lwu(tmp, Address(klass, Klass::layout_helper_offset()));
791 __ sraiw(tmp, tmp, Klass::_lh_array_tag_shift);
792 int tag = ((id == new_type_array_id) ? Klass::_lh_array_tag_type_value : Klass::_lh_array_tag_obj_value);
793 __ mv(t0, tag);
794 __ beq(t0, tmp, ok);
795 __ stop("assert(is an array klass)");
796 __ should_not_reach_here();
797 __ bind(ok);
798 }
799 #endif // ASSERT
800
801 // If TLAB is disabled, see if there is support for inlining contiguous
802 // allocations.
803 // Otherwise, just go to the slow path.
804 if (!UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
805 Register arr_size = x14;
806 Register tmp1 = x12;
807 Register tmp2 = x15;
808 Label slow_path;
809 assert_different_registers(length, klass, obj, arr_size, tmp1, tmp2);
810
811 // check that array length is small enough for fast path.
812 __ mv(t0, C1_MacroAssembler::max_array_allocation_length);
813 __ bgtu(length, t0, slow_path);
814
815 // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
816 __ lwu(tmp1, Address(klass, Klass::layout_helper_offset()));
817 __ andi(t0, tmp1, 0x1f);
818 __ sll(arr_size, length, t0);
819 int lh_header_size_width = exact_log2(Klass::_lh_header_size_mask + 1);
820 int lh_header_size_msb = Klass::_lh_header_size_shift + lh_header_size_width;
821 __ slli(tmp1, tmp1, XLEN - lh_header_size_msb);
822 __ srli(tmp1, tmp1, XLEN - lh_header_size_width);
823 __ add(arr_size, arr_size, tmp1);
824 __ addi(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
825 __ andi(arr_size, arr_size, ~(uint)MinObjAlignmentInBytesMask);
826
827 __ eden_allocate(obj, arr_size, 0, tmp1, slow_path); // preserves arr_size
828
829 __ initialize_header(obj, klass, length, tmp1, tmp2);
830 __ lbu(tmp1, Address(klass,
831 in_bytes(Klass::layout_helper_offset()) +
832 (Klass::_lh_header_size_shift / BitsPerByte)));
833 assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
834 assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
835 __ andi(tmp1, tmp1, Klass::_lh_header_size_mask);
836 __ sub(arr_size, arr_size, tmp1); // body length
837 __ add(tmp1, tmp1, obj); // body start
838 __ initialize_body(tmp1, arr_size, 0, tmp2);
839 __ membar(MacroAssembler::StoreStore);
840 __ verify_oop(obj);
841
842 __ ret();
843
844 __ bind(slow_path);
845 }
846
847 __ enter();
848 OopMap* map = save_live_registers(sasm);
849 assert_cond(map != NULL);
850 int call_offset = 0;
851 if (id == new_type_array_id) {
852 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
853 } else {
854 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
855 }
856
857 oop_maps = new OopMapSet();
858 assert_cond(oop_maps != NULL);
859 oop_maps->add_gc_map(call_offset, map);
860 restore_live_registers_except_r10(sasm);
861
862 __ verify_oop(obj);
863 __ leave();
864 __ ret();
865
866 // x10: new array
867 }
868 break;
869
870 case new_multi_array_id:
871 {
872 StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
873 // x10: klass
874 // x9: rank
875 // x12: address of 1st dimension
876 OopMap* map = save_live_registers(sasm);
877 assert_cond(map != NULL);
878 __ mv(c_rarg1, x10);
879 __ mv(c_rarg3, x12);
880 __ mv(c_rarg2, x9);
881 int call_offset = __ call_RT(x10, noreg, CAST_FROM_FN_PTR(address, new_multi_array), x11, x12, x13);
882
883 oop_maps = new OopMapSet();
884 assert_cond(oop_maps != NULL);
885 oop_maps->add_gc_map(call_offset, map);
886 restore_live_registers_except_r10(sasm);
887
888 // x10: new multi array
889 __ verify_oop(x10);
890 }
891 break;
892
893 case register_finalizer_id:
894 {
895 __ set_info("register_finalizer", dont_gc_arguments);
896
897 // This is called via call_runtime so the arguments
898 // will be place in C abi locations
899 __ verify_oop(c_rarg0);
900
901 // load the klass and check the has finalizer flag
902 Label register_finalizer;
903 Register t = x15;
904 __ load_klass(t, x10);
905 __ lwu(t, Address(t, Klass::access_flags_offset()));
906 __ andi(t0, t, JVM_ACC_HAS_FINALIZER);
907 __ bnez(t0, register_finalizer);
908 __ ret();
909
910 __ bind(register_finalizer);
911 __ enter();
912 OopMap* oop_map = save_live_registers(sasm);
913 assert_cond(oop_map != NULL);
914 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), x10);
915 oop_maps = new OopMapSet();
916 assert_cond(oop_maps != NULL);
917 oop_maps->add_gc_map(call_offset, oop_map);
918
919 // Now restore all the live registers
920 restore_live_registers(sasm);
921
922 __ leave();
923 __ ret();
924 }
925 break;
926
927 case throw_class_cast_exception_id:
928 {
929 StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments, does_not_return);
930 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
931 }
932 break;
933
934 case throw_incompatible_class_change_error_id:
935 {
936 StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments, does_not_return);
937 oop_maps = generate_exception_throw(sasm,
938 CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
939 }
940 break;
941
942 case slow_subtype_check_id:
943 {
944 // Typical calling sequence:
945 // push klass_RInfo (object klass or other subclass)
946 // push sup_k_RInfo (array element klass or other superclass)
947 // jump to slow_subtype_check
948 // Note that the subclass is pushed first, and is therefore deepest.
949 enum layout {
950 x10_off, x10_off_hi,
951 x12_off, x12_off_hi,
952 x14_off, x14_off_hi,
953 x15_off, x15_off_hi,
954 sup_k_off, sup_k_off_hi,
955 klass_off, klass_off_hi,
956 framesize,
957 result_off = sup_k_off
958 };
959
960 __ set_info("slow_subtype_check", dont_gc_arguments);
961 __ push_reg(RegSet::of(x10, x12, x14, x15), sp);
962
963 __ ld(x14, Address(sp, (klass_off) * VMRegImpl::stack_slot_size)); // sub klass
964 __ ld(x10, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size)); // super klass
965
966 Label miss;
967 __ check_klass_subtype_slow_path(x14, x10, x12, x15, NULL, &miss);
968
969 // fallthrough on success:
970 __ li(t0, 1);
971 __ sd(t0, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
972 __ pop_reg(RegSet::of(x10, x12, x14, x15), sp);
973 __ ret();
974
975 __ bind(miss);
976 __ sd(zr, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
977 __ pop_reg(RegSet::of(x10, x12, x14, x15), sp);
978 __ ret();
979 }
980 break;
981
982 case monitorenter_nofpu_id:
983 save_fpu_registers = false;
984 // fall through
985 case monitorenter_id:
986 {
987 StubFrame f(sasm, "monitorenter", dont_gc_arguments);
988 OopMap* map = save_live_registers(sasm, save_fpu_registers);
989 assert_cond(map != NULL);
990
991 // Called with store_parameter and not C abi
992 f.load_argument(1, x10); // x10: object
993 f.load_argument(0, x11); // x11: lock address
994
995 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), x10, x11);
996
997 oop_maps = new OopMapSet();
998 assert_cond(oop_maps != NULL);
999 oop_maps->add_gc_map(call_offset, map);
1000 restore_live_registers(sasm, save_fpu_registers);
1001 }
1002 break;
1003
1004 case monitorexit_nofpu_id:
1005 save_fpu_registers = false;
1006 // fall through
1007 case monitorexit_id:
1008 {
1009 StubFrame f(sasm, "monitorexit", dont_gc_arguments);
1010 OopMap* map = save_live_registers(sasm, save_fpu_registers);
1011 assert_cond(map != NULL);
1012
1013 // Called with store_parameter and not C abi
1014 f.load_argument(0, x10); // x10: lock address
1015
1016 // note: really a leaf routine but must setup last java sp
1017 // => use call_RT for now (speed can be improved by
1018 // doing last java sp setup manually)
1019 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), x10);
1020
1021 oop_maps = new OopMapSet();
1022 assert_cond(oop_maps != NULL);
1023 oop_maps->add_gc_map(call_offset, map);
1024 restore_live_registers(sasm, save_fpu_registers);
1025 }
1026 break;
1027
1028 case deoptimize_id:
1029 {
1030 StubFrame f(sasm, "deoptimize", dont_gc_arguments, does_not_return);
1031 OopMap* oop_map = save_live_registers(sasm);
1032 assert_cond(oop_map != NULL);
1033 f.load_argument(0, c_rarg1);
1034 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), c_rarg1);
1035
1036 oop_maps = new OopMapSet();
1037 assert_cond(oop_maps != NULL);
1038 oop_maps->add_gc_map(call_offset, oop_map);
1039 restore_live_registers(sasm);
1040 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1041 assert(deopt_blob != NULL, "deoptimization blob must have been created");
1042 __ leave();
1043 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1044 }
1045 break;
1046
1047 case throw_range_check_failed_id:
1048 {
1049 StubFrame f(sasm, "range_check_failed", dont_gc_arguments, does_not_return);
1050 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1051 }
1052 break;
1053
1054 case unwind_exception_id:
1055 {
1056 __ set_info("unwind_exception", dont_gc_arguments);
1057 // note: no stubframe since we are about to leave the current
1058 // activation and we are calling a leaf VM function only.
1059 generate_unwind_exception(sasm);
1060 }
1061 break;
1062
1063 case access_field_patching_id:
1064 {
1065 StubFrame f(sasm, "access_field_patching", dont_gc_arguments, does_not_return);
1066 // we should set up register map
1067 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1068 }
1069 break;
1070
1071 case load_klass_patching_id:
1072 {
1073 StubFrame f(sasm, "load_klass_patching", dont_gc_arguments, does_not_return);
1074 // we should set up register map
1075 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1076 }
1077 break;
1078
1079 case load_mirror_patching_id:
1080 {
1081 StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments, does_not_return);
1082 // we should set up register map
1083 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
1084 }
1085 break;
1086
1087 case load_appendix_patching_id:
1088 {
1089 StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments, does_not_return);
1090 // we should set up register map
1091 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
1092 }
1093 break;
1094
1095 case handle_exception_nofpu_id:
1096 case handle_exception_id:
1097 {
1098 StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1099 oop_maps = generate_handle_exception(id, sasm);
1100 }
1101 break;
1102
1103 case handle_exception_from_callee_id:
1104 {
1105 StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
1106 oop_maps = generate_handle_exception(id, sasm);
1107 }
1108 break;
1109
1110 case throw_index_exception_id:
1111 {
1112 StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments, does_not_return);
1113 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1114 }
1115 break;
1116
1117 case throw_array_store_exception_id:
1118 {
1119 StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments, does_not_return);
1120 // tos + 0: link
1121 // + 1: return address
1122 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
1123 }
1124 break;
1125
1126 case predicate_failed_trap_id:
1127 {
1128 StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments, does_not_return);
1129
1130 OopMap* map = save_live_registers(sasm);
1131 assert_cond(map != NULL);
1132
1133 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1134 oop_maps = new OopMapSet();
1135 assert_cond(oop_maps != NULL);
1136 oop_maps->add_gc_map(call_offset, map);
1137 restore_live_registers(sasm);
1138 __ leave();
1139 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1140 assert(deopt_blob != NULL, "deoptimization blob must have been created");
1141
1142 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1143 }
1144 break;
1145
1146 case dtrace_object_alloc_id:
1147 { // c_rarg0: object
1148 StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
1149 save_live_registers(sasm);
1150
1151 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), c_rarg0);
1152
1153 restore_live_registers(sasm);
1154 }
1155 break;
1156
1157 default:
1158 {
1159 StubFrame f(sasm, "unimplemented entry", dont_gc_arguments, does_not_return);
1160 __ li(x10, (int) id);
1161 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), x10);
1162 __ should_not_reach_here();
1163 }
1164 break;
1165 }
1166 }
1167 return oop_maps;
1168 }
1169
1170 #undef __
1171
1172 const char *Runtime1::pd_name_for_address(address entry) { Unimplemented(); return 0; }