1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "gc/shared/barrierSetAssembler.hpp"
29 #include "gc/shared/collectedHeap.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "gc/shared/tlab_globals.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/markWord.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/os.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40
41 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
42 FloatRegister f0, FloatRegister f1,
43 Register result)
44 {
45 Label done;
46 if (is_float) {
47 fcmps(f0, f1);
48 } else {
49 fcmpd(f0, f1);
50 }
51 if (unordered_result < 0) {
52 // we want -1 for unordered or less than, 0 for equal and 1 for
53 // greater than.
54 cset(result, NE); // Not equal or unordered
55 cneg(result, result, LT); // Less than or unordered
56 } else {
57 // we want -1 for less than, 0 for equal and 1 for unordered or
58 // greater than.
59 cset(result, NE); // Not equal or unordered
60 cneg(result, result, LO); // Less than
61 }
62 }
63
64 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
65 const int aligned_mask = BytesPerWord -1;
66 const int hdr_offset = oopDesc::mark_offset_in_bytes();
67 assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
68 int null_check_offset = -1;
69
70 verify_oop(obj);
71
72 // save object being locked into the BasicObjectLock
73 str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
74
75 null_check_offset = offset();
76
77 if (DiagnoseSyncOnValueBasedClasses != 0) {
78 load_klass(hdr, obj);
79 ldrb(hdr, Address(hdr, Klass::misc_flags_offset()));
80 tst(hdr, KlassFlags::_misc_is_value_based_class);
81 br(Assembler::NE, slow_case);
82 }
83
84 if (LockingMode == LM_LIGHTWEIGHT) {
85 lightweight_lock(disp_hdr, obj, hdr, temp, rscratch2, slow_case);
86 } else if (LockingMode == LM_LEGACY) {
87 Label done;
88 // Load object header
89 ldr(hdr, Address(obj, hdr_offset));
90 // and mark it as unlocked
91 orr(hdr, hdr, markWord::unlocked_value);
92
93 if (EnableValhalla) {
94 // Mask always_locked bit such that we go to the slow path if object is an inline type
95 andr(hdr, hdr, ~markWord::inline_type_bit_in_place);
96 }
97
98 // save unlocked object header into the displaced header location on the stack
99 str(hdr, Address(disp_hdr, 0));
100 // test if object header is still the same (i.e. unlocked), and if so, store the
101 // displaced header address in the object header - if it is not the same, get the
102 // object header instead
103 lea(rscratch2, Address(obj, hdr_offset));
104 cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
105 // if the object header was the same, we're done
106 // if the object header was not the same, it is now in the hdr register
107 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
108 //
109 // 1) (hdr & aligned_mask) == 0
110 // 2) sp <= hdr
111 // 3) hdr <= sp + page_size
112 //
113 // these 3 tests can be done by evaluating the following expression:
114 //
115 // (hdr - sp) & (aligned_mask - page_size)
116 //
117 // assuming both the stack pointer and page_size have their least
118 // significant 2 bits cleared and page_size is a power of 2
119 mov(rscratch1, sp);
120 sub(hdr, hdr, rscratch1);
121 ands(hdr, hdr, aligned_mask - (int)os::vm_page_size());
122 // for recursive locking, the result is zero => save it in the displaced header
123 // location (null in the displaced hdr location indicates recursive locking)
124 str(hdr, Address(disp_hdr, 0));
125 // otherwise we don't care about the result and handle locking via runtime call
126 cbnz(hdr, slow_case);
127 // done
128 bind(done);
129 inc_held_monitor_count(rscratch1);
130 }
131 return null_check_offset;
132 }
133
134
135 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
136 const int aligned_mask = BytesPerWord -1;
137 const int hdr_offset = oopDesc::mark_offset_in_bytes();
138 assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
139 Label done;
140
141 if (LockingMode != LM_LIGHTWEIGHT) {
142 // load displaced header
143 ldr(hdr, Address(disp_hdr, 0));
144 // if the loaded hdr is null we had recursive locking
145 // if we had recursive locking, we are done
146 cbz(hdr, done);
147 }
148
149 // load object
150 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
151 verify_oop(obj);
152
153 if (LockingMode == LM_LIGHTWEIGHT) {
154 lightweight_unlock(obj, hdr, temp, rscratch2, slow_case);
155 } else if (LockingMode == LM_LEGACY) {
156 // test if object header is pointing to the displaced header, and if so, restore
157 // the displaced header in the object - if the object header is not pointing to
158 // the displaced header, get the object header instead
159 // if the object header was not pointing to the displaced header,
160 // we do unlocking via runtime call
161 if (hdr_offset) {
162 lea(rscratch1, Address(obj, hdr_offset));
163 cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
164 } else {
165 cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
166 }
167 // done
168 bind(done);
169 dec_held_monitor_count(rscratch1);
170 }
171 }
172
173
174 // Defines obj, preserves var_size_in_bytes
175 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
176 if (UseTLAB) {
177 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
178 } else {
179 b(slow_case);
180 }
181 }
182
183 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
184 assert_different_registers(obj, klass, len);
185
186 if (UseCompactObjectHeaders || EnableValhalla) {
187 ldr(t1, Address(klass, Klass::prototype_header_offset()));
188 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
189 } else {
190 mov(t1, checked_cast<int32_t>(markWord::prototype().value()));
191 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
192 }
193
194 if (!UseCompactObjectHeaders) {
195 if (UseCompressedClassPointers) { // Take care not to kill klass
196 encode_klass_not_null(t1, klass);
197 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
198 } else {
199 str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
200 }
201 }
202
203 if (len->is_valid()) {
204 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
205 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
206 if (!is_aligned(base_offset, BytesPerWord)) {
207 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
208 // Clear gap/first 4 bytes following the length field.
209 strw(zr, Address(obj, base_offset));
210 }
211 } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
212 store_klass_gap(obj, zr);
213 }
214 }
215
216 // preserves obj, destroys len_in_bytes
217 //
218 // Scratch registers: t1 = r10, t2 = r11
219 //
220 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
221 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
222 assert(t1 == r10 && t2 == r11, "must be");
223
224 Label done;
225
226 // len_in_bytes is positive and ptr sized
227 subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
228 br(Assembler::EQ, done);
229
230 // zero_words() takes ptr in r10 and count in words in r11
231 mov(rscratch1, len_in_bytes);
232 lea(t1, Address(obj, hdr_size_in_bytes));
233 lsr(t2, rscratch1, LogBytesPerWord);
234 address tpc = zero_words(t1, t2);
235
236 bind(done);
237 if (tpc == nullptr) {
238 Compilation::current()->bailout("no space for trampoline stub");
239 }
240 }
241
242
243 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
244 assert_different_registers(obj, t1, t2); // XXX really?
245 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
246
247 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
248
249 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
250 }
251
252 // Scratch registers: t1 = r10, t2 = r11
253 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
254 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
255 "con_size_in_bytes is not multiple of alignment");
256 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
257
258 initialize_header(obj, klass, noreg, t1, t2);
259
260 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
261 // clear rest of allocated space
262 const Register index = t2;
263 if (var_size_in_bytes != noreg) {
264 mov(index, var_size_in_bytes);
265 initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
266 if (Compilation::current()->bailed_out()) {
267 return;
268 }
269 } else if (con_size_in_bytes > hdr_size_in_bytes) {
270 con_size_in_bytes -= hdr_size_in_bytes;
271 lea(t1, Address(obj, hdr_size_in_bytes));
272 address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
273 if (tpc == nullptr) {
274 Compilation::current()->bailout("no space for trampoline stub");
275 return;
276 }
277 }
278 }
279
280 membar(StoreStore);
281
282 if (CURRENT_ENV->dtrace_alloc_probes()) {
283 assert(obj == r0, "must be");
284 far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
285 }
286
287 verify_oop(obj);
288 }
289 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, int f, Register klass, Label& slow_case, bool zero_array) {
290 assert_different_registers(obj, len, t1, t2, klass);
291
292 // determine alignment mask
293 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
294
295 // check for negative or excessive length
296 mov(rscratch1, (int32_t)max_array_allocation_length);
297 cmp(len, rscratch1);
298 br(Assembler::HS, slow_case);
299
300 const Register arr_size = t2; // okay to be the same
301 // align object end
302 mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
303 add(arr_size, arr_size, len, ext::uxtw, f);
304 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
305
306 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
307
308 initialize_header(obj, klass, len, t1, t2);
309
310 // Align-up to word boundary, because we clear the 4 bytes potentially
311 // following the length field in initialize_header().
312 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
313 // clear rest of allocated space
314 if (zero_array) {
315 initialize_body(obj, arr_size, base_offset, t1, t2);
316 }
317 if (Compilation::current()->bailed_out()) {
318 return;
319 }
320
321 membar(StoreStore);
322
323 if (CURRENT_ENV->dtrace_alloc_probes()) {
324 assert(obj == r0, "must be");
325 far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
326 }
327
328 verify_oop(obj);
329 }
330
331 void C1_MacroAssembler::build_frame_helper(int frame_size_in_bytes, int sp_offset_for_orig_pc, int sp_inc, bool reset_orig_pc, bool needs_stack_repair) {
332 MacroAssembler::build_frame(frame_size_in_bytes);
333
334 if (needs_stack_repair) {
335 save_stack_increment(sp_inc, frame_size_in_bytes);
336 }
337 if (reset_orig_pc) {
338 // Zero orig_pc to detect deoptimization during buffering in the entry points
339 str(zr, Address(sp, sp_offset_for_orig_pc));
340 }
341 }
342
343 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, bool needs_stack_repair, bool has_scalarized_args, Label* verified_inline_entry_label) {
344 // Make sure there is enough stack space for this method's activation.
345 // Note that we do this before creating a frame.
346 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
347 generate_stack_overflow_check(bang_size_in_bytes);
348
349 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
350
351 // Insert nmethod entry barrier into frame.
352 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
353 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
354
355 if (verified_inline_entry_label != nullptr) {
356 // Jump here from the scalarized entry points that already created the frame.
357 bind(*verified_inline_entry_label);
358 }
359 }
360
361 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
362 // If we have to make this method not-entrant we'll overwrite its
363 // first instruction with a jump. For this action to be legal we
364 // must ensure that this first instruction is a B, BL, NOP, BKPT,
365 // SVC, HVC, or SMC. Make it a NOP.
366 nop();
367 if (C1Breakpoint) brk(1);
368 }
369
370 int C1_MacroAssembler::scalarized_entry(const CompiledEntrySignature* ces, int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, Label& verified_inline_entry_label, bool is_inline_ro_entry) {
371 assert(InlineTypePassFieldsAsArgs, "sanity");
372 // Make sure there is enough stack space for this method's activation.
373 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
374 generate_stack_overflow_check(bang_size_in_bytes);
375
376 GrowableArray<SigEntry>* sig = ces->sig();
377 GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
378 VMRegPair* regs = ces->regs();
379 VMRegPair* regs_cc = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
380 int args_on_stack = ces->args_on_stack();
381 int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
382
383 assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
384 BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
385 int args_passed = sig->length();
386 int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
387
388 // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
389 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
390
391 // The runtime call might safepoint, make sure nmethod entry barrier is executed
392 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
393 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
394 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
395
396 // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
397 mov(r19, (intptr_t) ces->method());
398 if (is_inline_ro_entry) {
399 far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_no_receiver_id)));
400 } else {
401 far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_id)));
402 }
403 int rt_call_offset = offset();
404
405 // The runtime call returns the new array in r20 instead of the usual r0
406 // because r0 is also j_rarg7 which may be holding a live argument here.
407 Register val_array = r20;
408
409 // Remove the temp frame
410 MacroAssembler::remove_frame(frame_size_in_bytes);
411
412 // Check if we need to extend the stack for packing
413 int sp_inc = 0;
414 if (args_on_stack > args_on_stack_cc) {
415 sp_inc = extend_stack_for_inline_args(args_on_stack);
416 }
417
418 shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
419 args_passed_cc, args_on_stack_cc, regs_cc, // from
420 args_passed, args_on_stack, regs, // to
421 sp_inc, val_array);
422
423 // Create the real frame. Below jump will then skip over the stack banging and frame
424 // setup code in the verified_inline_entry (which has a different real_frame_size).
425 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
426
427 b(verified_inline_entry_label);
428 return rt_call_offset;
429 }
430
431
432 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
433 // rfp, + 0: link
434 // + 1: return address
435 // + 2: argument with offset 0
436 // + 3: argument with offset 1
437 // + 4: ...
438
439 ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
440 }
441
442 #ifndef PRODUCT
443
444 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
445 if (!VerifyOops) return;
446 verify_oop_addr(Address(sp, stack_offset));
447 }
448
449 void C1_MacroAssembler::verify_not_null_oop(Register r) {
450 if (!VerifyOops) return;
451 Label not_null;
452 cbnz(r, not_null);
453 stop("non-null oop required");
454 bind(not_null);
455 verify_oop(r);
456 }
457
458 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
459 #ifdef ASSERT
460 static int nn;
461 if (inv_r0) mov(r0, 0xDEAD);
462 if (inv_r19) mov(r19, 0xDEAD);
463 if (inv_r2) mov(r2, nn++);
464 if (inv_r3) mov(r3, 0xDEAD);
465 if (inv_r4) mov(r4, 0xDEAD);
466 if (inv_r5) mov(r5, 0xDEAD);
467 #endif
468 }
469 #endif // ifndef PRODUCT