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 basic_lock, Register temp, Label& slow_case) {
65 assert_different_registers(hdr, obj, basic_lock, temp, rscratch2);
66 int null_check_offset = -1;
67
68 verify_oop(obj);
69
70 // save object being locked into the BasicObjectLock
71 str(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
72
73 null_check_offset = offset();
74
75 lightweight_lock(basic_lock, obj, hdr, temp, rscratch2, slow_case);
76
77 return null_check_offset;
78 }
79
80
81 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register basic_lock, Register temp, Label& slow_case) {
82 assert_different_registers(hdr, obj, basic_lock, temp, rscratch2);
83
84 // load object
85 ldr(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
86 verify_oop(obj);
87
88 lightweight_unlock(obj, hdr, temp, rscratch2, slow_case);
89 }
90
91
92 // Defines obj, preserves var_size_in_bytes
93 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
94 if (UseTLAB) {
95 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
96 } else {
97 b(slow_case);
98 }
99 }
100
101 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
102 assert_different_registers(obj, klass, len);
103
104 if (UseCompactObjectHeaders || EnableValhalla) {
105 // COH: Markword contains class pointer which is only known at runtime.
106 // Valhalla: Could have value class which has a different prototype header to a normal object.
107 // In both cases, we need to fetch dynamically.
108 ldr(t1, Address(klass, Klass::prototype_header_offset()));
109 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
110 } else {
111 // Otherwise: Can use the statically computed prototype header which is the same for every object.
112 mov(t1, checked_cast<int32_t>(markWord::prototype().value()));
113 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
114 }
115
116 if (!UseCompactObjectHeaders) {
117 // COH: Markword already contains class pointer. Nothing else to do.
118 // Otherwise: Fetch klass pointer following the markword
119 if (UseCompressedClassPointers) { // Take care not to kill klass
120 encode_klass_not_null(t1, klass);
121 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
122 } else {
123 str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
124 }
125 }
126
127 if (len->is_valid()) {
128 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
129 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
130 if (!is_aligned(base_offset, BytesPerWord)) {
131 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
132 // Clear gap/first 4 bytes following the length field.
133 strw(zr, Address(obj, base_offset));
134 }
135 } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
136 store_klass_gap(obj, zr);
137 }
138 }
139
140 // preserves obj, destroys len_in_bytes
141 //
142 // Scratch registers: t1 = r10, t2 = r11
143 //
144 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
145 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
146 assert(t1 == r10 && t2 == r11, "must be");
147
148 Label done;
149
150 // len_in_bytes is positive and ptr sized
151 subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
152 br(Assembler::EQ, done);
153
154 // zero_words() takes ptr in r10 and count in words in r11
155 mov(rscratch1, len_in_bytes);
156 lea(t1, Address(obj, hdr_size_in_bytes));
157 lsr(t2, rscratch1, LogBytesPerWord);
158 address tpc = zero_words(t1, t2);
159
160 bind(done);
161 if (tpc == nullptr) {
162 Compilation::current()->bailout("no space for trampoline stub");
163 }
164 }
165
166
167 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
168 assert_different_registers(obj, t1, t2); // XXX really?
169 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
170
171 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
172
173 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
174 }
175
176 // Scratch registers: t1 = r10, t2 = r11
177 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) {
178 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
179 "con_size_in_bytes is not multiple of alignment");
180 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
181
182 initialize_header(obj, klass, noreg, t1, t2);
183
184 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
185 // clear rest of allocated space
186 const Register index = t2;
187 if (var_size_in_bytes != noreg) {
188 mov(index, var_size_in_bytes);
189 initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
190 if (Compilation::current()->bailed_out()) {
191 return;
192 }
193 } else if (con_size_in_bytes > hdr_size_in_bytes) {
194 con_size_in_bytes -= hdr_size_in_bytes;
195 lea(t1, Address(obj, hdr_size_in_bytes));
196 address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
197 if (tpc == nullptr) {
198 Compilation::current()->bailout("no space for trampoline stub");
199 return;
200 }
201 }
202 }
203
204 membar(StoreStore);
205
206 if (CURRENT_ENV->dtrace_alloc_probes()) {
207 assert(obj == r0, "must be");
208 far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
209 }
210
211 verify_oop(obj);
212 }
213 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) {
214 assert_different_registers(obj, len, t1, t2, klass);
215
216 // determine alignment mask
217 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
218
219 // check for negative or excessive length
220 mov(rscratch1, (int32_t)max_array_allocation_length);
221 cmp(len, rscratch1);
222 br(Assembler::HS, slow_case);
223
224 const Register arr_size = t2; // okay to be the same
225 // align object end
226 mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
227 add(arr_size, arr_size, len, ext::uxtw, f);
228 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
229
230 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
231
232 initialize_header(obj, klass, len, t1, t2);
233
234 // Align-up to word boundary, because we clear the 4 bytes potentially
235 // following the length field in initialize_header().
236 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
237 // clear rest of allocated space
238 if (zero_array) {
239 initialize_body(obj, arr_size, base_offset, t1, t2);
240 }
241 if (Compilation::current()->bailed_out()) {
242 return;
243 }
244
245 membar(StoreStore);
246
247 if (CURRENT_ENV->dtrace_alloc_probes()) {
248 assert(obj == r0, "must be");
249 far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
250 }
251
252 verify_oop(obj);
253 }
254
255 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) {
256 MacroAssembler::build_frame(frame_size_in_bytes);
257
258 if (needs_stack_repair) {
259 save_stack_increment(sp_inc, frame_size_in_bytes);
260 }
261 if (reset_orig_pc) {
262 // Zero orig_pc to detect deoptimization during buffering in the entry points
263 str(zr, Address(sp, sp_offset_for_orig_pc));
264 }
265 }
266
267 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) {
268 // Make sure there is enough stack space for this method's activation.
269 // Note that we do this before creating a frame.
270 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
271 generate_stack_overflow_check(bang_size_in_bytes);
272
273 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
274
275 // Insert nmethod entry barrier into frame.
276 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
277 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
278
279 if (verified_inline_entry_label != nullptr) {
280 // Jump here from the scalarized entry points that already created the frame.
281 bind(*verified_inline_entry_label);
282 }
283 }
284
285 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
286 // If we have to make this method not-entrant we'll overwrite its
287 // first instruction with a jump. For this action to be legal we
288 // must ensure that this first instruction is a B, BL, NOP, BKPT,
289 // SVC, HVC, or SMC. Make it a NOP.
290 nop();
291 if (C1Breakpoint) brk(1);
292 }
293
294 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) {
295 assert(InlineTypePassFieldsAsArgs, "sanity");
296 // Make sure there is enough stack space for this method's activation.
297 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
298 generate_stack_overflow_check(bang_size_in_bytes);
299
300 GrowableArray<SigEntry>* sig = ces->sig();
301 GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
302 VMRegPair* regs = ces->regs();
303 VMRegPair* regs_cc = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
304 int args_on_stack = ces->args_on_stack();
305 int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
306
307 assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
308 BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
309 int args_passed = sig->length();
310 int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
311
312 // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
313 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
314
315 // The runtime call might safepoint, make sure nmethod entry barrier is executed
316 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
317 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
318 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
319
320 // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
321 mov(r19, (intptr_t) ces->method());
322 if (is_inline_ro_entry) {
323 far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_buffer_inline_args_no_receiver_id)));
324 } else {
325 far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_buffer_inline_args_id)));
326 }
327 int rt_call_offset = offset();
328
329 // The runtime call returns the new array in r20 instead of the usual r0
330 // because r0 is also j_rarg7 which may be holding a live argument here.
331 Register val_array = r20;
332
333 // Remove the temp frame
334 MacroAssembler::remove_frame(frame_size_in_bytes);
335
336 // Check if we need to extend the stack for packing
337 int sp_inc = 0;
338 if (args_on_stack > args_on_stack_cc) {
339 sp_inc = extend_stack_for_inline_args(args_on_stack);
340 }
341
342 shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
343 args_passed_cc, args_on_stack_cc, regs_cc, // from
344 args_passed, args_on_stack, regs, // to
345 sp_inc, val_array);
346
347 // Create the real frame. Below jump will then skip over the stack banging and frame
348 // setup code in the verified_inline_entry (which has a different real_frame_size).
349 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
350
351 b(verified_inline_entry_label);
352 return rt_call_offset;
353 }
354
355
356 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
357 // rfp, + 0: link
358 // + 1: return address
359 // + 2: argument with offset 0
360 // + 3: argument with offset 1
361 // + 4: ...
362
363 ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
364 }
365
366 #ifndef PRODUCT
367
368 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
369 if (!VerifyOops) return;
370 verify_oop_addr(Address(sp, stack_offset));
371 }
372
373 void C1_MacroAssembler::verify_not_null_oop(Register r) {
374 if (!VerifyOops) return;
375 Label not_null;
376 cbnz(r, not_null);
377 stop("non-null oop required");
378 bind(not_null);
379 verify_oop(r);
380 }
381
382 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
383 #ifdef ASSERT
384 static int nn;
385 if (inv_r0) mov(r0, 0xDEAD);
386 if (inv_r19) mov(r19, 0xDEAD);
387 if (inv_r2) mov(r2, nn++);
388 if (inv_r3) mov(r3, 0xDEAD);
389 if (inv_r4) mov(r4, 0xDEAD);
390 if (inv_r5) mov(r5, 0xDEAD);
391 #endif
392 }
393 #endif // ifndef PRODUCT