1 /*
2 * Copyright (c) 1999, 2025, 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 */
24
25 #include "c1/c1_MacroAssembler.hpp"
26 #include "c1/c1_Runtime1.hpp"
27 #include "code/compiledIC.hpp"
28 #include "compiler/compilerDefinitions.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/collectedHeap.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/frame.inline.hpp"
38 #include "runtime/globals.hpp"
39 #include "runtime/os.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "utilities/checkedCast.hpp"
43 #include "utilities/globalDefinitions.hpp"
44
45 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
46 const int aligned_mask = BytesPerWord -1;
47 const int hdr_offset = oopDesc::mark_offset_in_bytes();
48 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
49 assert_different_registers(hdr, obj, disp_hdr, tmp);
50 int null_check_offset = -1;
51
52 verify_oop(obj);
53
54 // save object being locked into the BasicObjectLock
55 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
56
57 null_check_offset = offset();
58
59 if (DiagnoseSyncOnValueBasedClasses != 0) {
60 load_klass(hdr, obj, rscratch1);
61 testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
62 jcc(Assembler::notZero, slow_case);
63 }
64
65 if (LockingMode == LM_LIGHTWEIGHT) {
66 #ifdef _LP64
67 const Register thread = r15_thread;
68 lightweight_lock(disp_hdr, obj, hdr, thread, tmp, slow_case);
69 #else
70 // Implicit null check.
71 movptr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
72 // Lacking registers and thread on x86_32. Always take slow path.
73 jmp(slow_case);
74 #endif
75 } else if (LockingMode == LM_LEGACY) {
76 Label done;
77 // Load object header
78 movptr(hdr, Address(obj, hdr_offset));
79 // and mark it as unlocked
80 orptr(hdr, markWord::unlocked_value);
81 if (EnableValhalla) {
82 // Mask inline_type bit such that we go to the slow path if object is an inline type
83 andptr(hdr, ~((int) markWord::inline_type_bit_in_place));
84 }
85 // save unlocked object header into the displaced header location on the stack
86 movptr(Address(disp_hdr, 0), hdr);
87 // test if object header is still the same (i.e. unlocked), and if so, store the
88 // displaced header address in the object header - if it is not the same, get the
89 // object header instead
90 MacroAssembler::lock(); // must be immediately before cmpxchg!
91 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
92 // if the object header was the same, we're done
93 jcc(Assembler::equal, done);
94 // if the object header was not the same, it is now in the hdr register
95 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
96 //
97 // 1) (hdr & aligned_mask) == 0
98 // 2) rsp <= hdr
99 // 3) hdr <= rsp + page_size
100 //
101 // these 3 tests can be done by evaluating the following expression:
102 //
103 // (hdr - rsp) & (aligned_mask - page_size)
104 //
105 // assuming both the stack pointer and page_size have their least
106 // significant 2 bits cleared and page_size is a power of 2
107 subptr(hdr, rsp);
108 andptr(hdr, aligned_mask - (int)os::vm_page_size());
109 // for recursive locking, the result is zero => save it in the displaced header
110 // location (null in the displaced hdr location indicates recursive locking)
111 movptr(Address(disp_hdr, 0), hdr);
112 // otherwise we don't care about the result and handle locking via runtime call
113 jcc(Assembler::notZero, slow_case);
114 // done
115 bind(done);
116 inc_held_monitor_count();
117 }
118
119 return null_check_offset;
120 }
121
122 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
123 const int aligned_mask = BytesPerWord -1;
124 const int hdr_offset = oopDesc::mark_offset_in_bytes();
125 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
126 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
127 Label done;
128
129 if (LockingMode != LM_LIGHTWEIGHT) {
130 // load displaced header
131 movptr(hdr, Address(disp_hdr, 0));
132 // if the loaded hdr is null we had recursive locking
133 testptr(hdr, hdr);
134 // if we had recursive locking, we are done
135 jcc(Assembler::zero, done);
136 }
137
138 // load object
139 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
140 verify_oop(obj);
141
142 if (LockingMode == LM_LIGHTWEIGHT) {
143 #ifdef _LP64
144 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
145 #else
146 // Lacking registers and thread on x86_32. Always take slow path.
147 jmp(slow_case);
148 #endif
149 } else if (LockingMode == LM_LEGACY) {
150 // test if object header is pointing to the displaced header, and if so, restore
151 // the displaced header in the object - if the object header is not pointing to
152 // the displaced header, get the object header instead
153 MacroAssembler::lock(); // must be immediately before cmpxchg!
154 cmpxchgptr(hdr, Address(obj, hdr_offset));
155 // if the object header was not pointing to the displaced header,
156 // we do unlocking via runtime call
157 jcc(Assembler::notEqual, slow_case);
158 // done
159 bind(done);
160 dec_held_monitor_count();
161 }
162 }
163
164
165 // Defines obj, preserves var_size_in_bytes
166 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
167 if (UseTLAB) {
168 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
169 } else {
170 jmp(slow_case);
171 }
172 }
173
174
175 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
176 assert_different_registers(obj, klass, len, t1, t2);
177 if (UseCompactObjectHeaders || EnableValhalla) {
178 movptr(t1, Address(klass, Klass::prototype_header_offset()));
179 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
180 } else {
181 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
182 }
183 #ifdef _LP64
184 if (UseCompressedClassPointers) { // Take care not to kill klass
185 movptr(t1, klass);
186 encode_klass_not_null(t1, rscratch1);
187 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
188 } else if (!UseCompactObjectHeaders)
189 #endif
190 {
191 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
192 }
193
194 if (len->is_valid()) {
195 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
196 #ifdef _LP64
197 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
198 if (!is_aligned(base_offset, BytesPerWord)) {
199 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
200 // Clear gap/first 4 bytes following the length field.
201 xorl(t1, t1);
202 movl(Address(obj, base_offset), t1);
203 }
204 #endif
205 }
206 #ifdef _LP64
207 else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
208 xorptr(t1, t1);
209 store_klass_gap(obj, t1);
210 }
211 #endif
212 }
213
214
215 // preserves obj, destroys len_in_bytes
216 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
217 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
218 Label done;
219
220 // len_in_bytes is positive and ptr sized
221 subptr(len_in_bytes, hdr_size_in_bytes);
222 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
223 bind(done);
224 }
225
226
227 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
228 assert(obj == rax, "obj must be in rax, for cmpxchg");
229 assert_different_registers(obj, t1, t2); // XXX really?
230 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
231
232 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
233
234 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
235 }
236
237 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) {
238 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
239 "con_size_in_bytes is not multiple of alignment");
240 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
241 if (UseCompactObjectHeaders) {
242 assert(hdr_size_in_bytes == 8, "check object headers size");
243 }
244 initialize_header(obj, klass, noreg, t1, t2);
245
246 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
247 // clear rest of allocated space
248 const Register t1_zero = t1;
249 const Register index = t2;
250 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
251 if (var_size_in_bytes != noreg) {
252 mov(index, var_size_in_bytes);
253 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
254 } else if (con_size_in_bytes <= threshold) {
255 // use explicit null stores
256 // code size = 2 + 3*n bytes (n = number of fields to clear)
257 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
258 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
259 movptr(Address(obj, i), t1_zero);
260 } else if (con_size_in_bytes > hdr_size_in_bytes) {
261 // use loop to null out the fields
262 // code size = 16 bytes for even n (n = number of fields to clear)
263 // initialize last object field first if odd number of fields
264 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
265 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
266 // initialize last object field if constant size is odd
267 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
268 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
269 // initialize remaining object fields: rdx is a multiple of 2
270 { Label loop;
271 bind(loop);
272 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
273 t1_zero);
274 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
275 t1_zero);)
276 decrement(index);
277 jcc(Assembler::notZero, loop);
278 }
279 }
280 }
281
282 if (CURRENT_ENV->dtrace_alloc_probes()) {
283 assert(obj == rax, "must be");
284 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
285 }
286
287 verify_oop(obj);
288 }
289
290 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case, bool zero_array) {
291 assert(obj == rax, "obj must be in rax, for cmpxchg");
292 assert_different_registers(obj, len, t1, t2, klass);
293
294 // determine alignment mask
295 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
296
297 // check for negative or excessive length
298 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
299 jcc(Assembler::above, slow_case);
300
301 const Register arr_size = t2; // okay to be the same
302 // align object end
303 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
304 lea(arr_size, Address(arr_size, len, f));
305 andptr(arr_size, ~MinObjAlignmentInBytesMask);
306
307 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
308
309 initialize_header(obj, klass, len, t1, t2);
310
311 // clear rest of allocated space
312 if (zero_array) {
313 const Register len_zero = len;
314 // Align-up to word boundary, because we clear the 4 bytes potentially
315 // following the length field in initialize_header().
316 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
317 initialize_body(obj, arr_size, base_offset, len_zero);
318 }
319
320 if (CURRENT_ENV->dtrace_alloc_probes()) {
321 assert(obj == rax, "must be");
322 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
323 }
324
325 verify_oop(obj);
326 }
327
328 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) {
329 push(rbp);
330 if (PreserveFramePointer) {
331 mov(rbp, rsp);
332 }
333 decrement(rsp, frame_size_in_bytes);
334
335 if (needs_stack_repair) {
336 // Save stack increment (also account for fixed framesize and rbp)
337 assert((sp_inc & (StackAlignmentInBytes-1)) == 0, "stack increment not aligned");
338 int real_frame_size = sp_inc + frame_size_in_bytes + wordSize;
339 movptr(Address(rsp, frame_size_in_bytes - wordSize), real_frame_size);
340 }
341 if (reset_orig_pc) {
342 // Zero orig_pc to detect deoptimization during buffering in the entry points
343 movptr(Address(rsp, sp_offset_for_orig_pc), 0);
344 }
345 }
346
347 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) {
348 // Make sure there is enough stack space for this method's activation.
349 // Note that we do this before doing an enter(). This matches the
350 // ordering of C2's stack overflow check / rsp decrement and allows
351 // the SharedRuntime stack overflow handling to be consistent
352 // between the two compilers.
353 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
354 generate_stack_overflow_check(bang_size_in_bytes);
355
356 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
357
358 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
359 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
360 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
361
362 if (verified_inline_entry_label != nullptr) {
363 // Jump here from the scalarized entry points that already created the frame.
364 bind(*verified_inline_entry_label);
365 }
366 }
367
368 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
369 if (breakAtEntry || VerifyFPU) {
370 // Verified Entry first instruction should be 5 bytes long for correct
371 // patching by patch_verified_entry().
372 //
373 // Breakpoint and VerifyFPU have one byte first instruction.
374 // Also first instruction will be one byte "push(rbp)" if stack banging
375 // code is not generated (see build_frame() above).
376 // For all these cases generate long instruction first.
377 fat_nop();
378 }
379 if (breakAtEntry) int3();
380 // build frame
381 IA32_ONLY( verify_FPU(0, "method_entry"); )
382 }
383
384 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) {
385 assert(InlineTypePassFieldsAsArgs, "sanity");
386 // Make sure there is enough stack space for this method's activation.
387 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
388 generate_stack_overflow_check(bang_size_in_bytes);
389
390 GrowableArray<SigEntry>* sig = ces->sig();
391 GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
392 VMRegPair* regs = ces->regs();
393 VMRegPair* regs_cc = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
394 int args_on_stack = ces->args_on_stack();
395 int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
396
397 assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
398 BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
399 int args_passed = sig->length();
400 int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
401
402 // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
403 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
404
405 // The runtime call might safepoint, make sure nmethod entry barrier is executed
406 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
407 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
408 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
409
410 // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
411 movptr(rbx, (intptr_t)(ces->method()));
412 if (is_inline_ro_entry) {
413 call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_no_receiver_id)));
414 } else {
415 call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_id)));
416 }
417 int rt_call_offset = offset();
418
419 // Remove the temp frame
420 addptr(rsp, frame_size_in_bytes);
421 pop(rbp);
422
423 // Check if we need to extend the stack for packing
424 int sp_inc = 0;
425 if (args_on_stack > args_on_stack_cc) {
426 sp_inc = extend_stack_for_inline_args(args_on_stack);
427 }
428
429 shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
430 args_passed_cc, args_on_stack_cc, regs_cc, // from
431 args_passed, args_on_stack, regs, // to
432 sp_inc, rax);
433
434 // Create the real frame. Below jump will then skip over the stack banging and frame
435 // setup code in the verified_inline_entry (which has a different real_frame_size).
436 build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
437
438 jmp(verified_inline_entry_label);
439 return rt_call_offset;
440 }
441
442 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
443 // rbp, + 0: link
444 // + 1: return address
445 // + 2: argument with offset 0
446 // + 3: argument with offset 1
447 // + 4: ...
448
449 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
450 }
451
452 #ifndef PRODUCT
453
454 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
455 if (!VerifyOops) return;
456 verify_oop_addr(Address(rsp, stack_offset));
457 }
458
459 void C1_MacroAssembler::verify_not_null_oop(Register r) {
460 if (!VerifyOops) return;
461 Label not_null;
462 testptr(r, r);
463 jcc(Assembler::notZero, not_null);
464 stop("non-null oop required");
465 bind(not_null);
466 verify_oop(r);
467 }
468
469 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
470 #ifdef ASSERT
471 if (inv_rax) movptr(rax, 0xDEAD);
472 if (inv_rbx) movptr(rbx, 0xDEAD);
473 if (inv_rcx) movptr(rcx, 0xDEAD);
474 if (inv_rdx) movptr(rdx, 0xDEAD);
475 if (inv_rsi) movptr(rsi, 0xDEAD);
476 if (inv_rdi) movptr(rdi, 0xDEAD);
477 #endif
478 }
479
480 #endif // ifndef PRODUCT