1 /*
  2  * Copyright (c) 1999, 2024, 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 "precompiled.hpp"
 26 #include "c1/c1_MacroAssembler.hpp"
 27 #include "c1/c1_Runtime1.hpp"
 28 #include "code/compiledIC.hpp"
 29 #include "compiler/compilerDefinitions.inline.hpp"
 30 #include "gc/shared/barrierSet.hpp"
 31 #include "gc/shared/barrierSetAssembler.hpp"
 32 #include "gc/shared/collectedHeap.hpp"
 33 #include "gc/shared/tlab_globals.hpp"
 34 #include "interpreter/interpreter.hpp"
 35 #include "oops/arrayOop.hpp"
 36 #include "oops/markWord.hpp"
 37 #include "runtime/basicLock.hpp"
 38 #include "runtime/frame.inline.hpp"
 39 #include "runtime/globals.hpp"
 40 #include "runtime/os.hpp"
 41 #include "runtime/sharedRuntime.hpp"
 42 #include "runtime/stubRoutines.hpp"
 43 #include "utilities/checkedCast.hpp"
 44 #include "utilities/globalDefinitions.hpp"
 45 
 46 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
 47   const int aligned_mask = BytesPerWord -1;
 48   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 49   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 50   assert_different_registers(hdr, obj, disp_hdr, tmp);
 51   int null_check_offset = -1;
 52 
 53   verify_oop(obj);
 54 
 55   // save object being locked into the BasicObjectLock
 56   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 57 
 58   null_check_offset = offset();
 59 
 60   if (DiagnoseSyncOnValueBasedClasses != 0) {
 61     load_klass(hdr, obj, rscratch1);
 62     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 63     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 64     jcc(Assembler::notZero, slow_case);
 65   }
 66 
 67   if (LockingMode == LM_LIGHTWEIGHT) {
 68 #ifdef _LP64
 69     const Register thread = r15_thread;
 70 #else
 71     const Register thread = disp_hdr;
 72     get_thread(thread);
 73 #endif
 74     lightweight_lock(obj, hdr, thread, tmp, slow_case);
 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   }
117 
118   inc_held_monitor_count();
119 
120   return null_check_offset;
121 }
122 
123 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
124   const int aligned_mask = BytesPerWord -1;
125   const int hdr_offset = oopDesc::mark_offset_in_bytes();
126   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
127   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
128   Label done;
129 
130   if (LockingMode != LM_LIGHTWEIGHT) {
131     // load displaced header
132     movptr(hdr, Address(disp_hdr, 0));
133     // if the loaded hdr is null we had recursive locking
134     testptr(hdr, hdr);
135     // if we had recursive locking, we are done
136     jcc(Assembler::zero, done);
137   }
138 
139   // load object
140   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
141   verify_oop(obj);
142 
143   if (LockingMode == LM_LIGHTWEIGHT) {
144 #ifdef _LP64
145     lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
146 #else
147     // This relies on the implementation of lightweight_unlock being able to handle
148     // that the reg_rax and thread Register parameters may alias each other.
149     get_thread(disp_hdr);
150     lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
151 #endif
152   } else if (LockingMode == LM_LEGACY) {
153     // test if object header is pointing to the displaced header, and if so, restore
154     // the displaced header in the object - if the object header is not pointing to
155     // the displaced header, get the object header instead
156     MacroAssembler::lock(); // must be immediately before cmpxchg!
157     cmpxchgptr(hdr, Address(obj, hdr_offset));
158     // if the object header was not pointing to the displaced header,
159     // we do unlocking via runtime call
160     jcc(Assembler::notEqual, slow_case);
161     // done
162   }
163   bind(done);
164   dec_held_monitor_count();
165 }
166 
167 
168 // Defines obj, preserves var_size_in_bytes
169 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
170   if (UseTLAB) {
171     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
172   } else {
173     jmp(slow_case);
174   }
175 }
176 
177 
178 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
179   assert_different_registers(obj, klass, len);
180   if (EnableValhalla) {
181     // Need to copy markWord::prototype header for klass
182     assert_different_registers(obj, klass, len, t1, t2);
183     movptr(t1, Address(klass, Klass::prototype_header_offset()));
184     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
185   } else {
186     // This assumes that all prototype bits fit in an int32_t
187     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
188   }
189 #ifdef _LP64
190   if (UseCompressedClassPointers) { // Take care not to kill klass
191     movptr(t1, klass);
192     encode_klass_not_null(t1, rscratch1);
193     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
194   } else
195 #endif
196   {
197     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
198   }
199 
200   if (len->is_valid()) {
201     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
202   }
203 #ifdef _LP64
204   else if (UseCompressedClassPointers) {
205     xorptr(t1, t1);
206     store_klass_gap(obj, t1);
207   }
208 #endif
209 }
210 
211 
212 // preserves obj, destroys len_in_bytes
213 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
214   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
215   Label done;
216 
217   // len_in_bytes is positive and ptr sized
218   subptr(len_in_bytes, hdr_size_in_bytes);
219   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
220   bind(done);
221 }
222 
223 
224 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
225   assert(obj == rax, "obj must be in rax, for cmpxchg");
226   assert_different_registers(obj, t1, t2); // XXX really?
227   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
228 
229   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
230 
231   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
232 }
233 
234 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) {
235   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
236          "con_size_in_bytes is not multiple of alignment");
237   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
238 
239   initialize_header(obj, klass, noreg, t1, t2);
240 
241   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
242     // clear rest of allocated space
243     const Register t1_zero = t1;
244     const Register index = t2;
245     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
246     if (var_size_in_bytes != noreg) {
247       mov(index, var_size_in_bytes);
248       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
249     } else if (con_size_in_bytes <= threshold) {
250       // use explicit null stores
251       // code size = 2 + 3*n bytes (n = number of fields to clear)
252       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
253       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
254         movptr(Address(obj, i), t1_zero);
255     } else if (con_size_in_bytes > hdr_size_in_bytes) {
256       // use loop to null out the fields
257       // code size = 16 bytes for even n (n = number of fields to clear)
258       // initialize last object field first if odd number of fields
259       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
260       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
261       // initialize last object field if constant size is odd
262       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
263         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
264       // initialize remaining object fields: rdx is a multiple of 2
265       { Label loop;
266         bind(loop);
267         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
268                t1_zero);
269         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
270                t1_zero);)
271         decrement(index);
272         jcc(Assembler::notZero, loop);
273       }
274     }
275   }
276 
277   if (CURRENT_ENV->dtrace_alloc_probes()) {
278     assert(obj == rax, "must be");
279     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
280   }
281 
282   verify_oop(obj);
283 }
284 
285 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
286   assert(obj == rax, "obj must be in rax, for cmpxchg");
287   assert_different_registers(obj, len, t1, t2, klass);
288 
289   // determine alignment mask
290   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
291 
292   // check for negative or excessive length
293   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
294   jcc(Assembler::above, slow_case);
295 
296   const Register arr_size = t2; // okay to be the same
297   // align object end
298   movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
299   lea(arr_size, Address(arr_size, len, f));
300   andptr(arr_size, ~MinObjAlignmentInBytesMask);
301 
302   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
303 
304   initialize_header(obj, klass, len, t1, t2);
305 
306   // clear rest of allocated space
307   const Register len_zero = len;
308   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
309 
310   if (CURRENT_ENV->dtrace_alloc_probes()) {
311     assert(obj == rax, "must be");
312     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
313   }
314 
315   verify_oop(obj);
316 }
317 
318 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) {
319   push(rbp);
320   if (PreserveFramePointer) {
321     mov(rbp, rsp);
322   }
323 #if !defined(_LP64) && defined(COMPILER2)
324   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
325       // c2 leaves fpu stack dirty. Clean it on entry
326       empty_FPU_stack();
327     }
328 #endif // !_LP64 && COMPILER2
329   decrement(rsp, frame_size_in_bytes);
330 
331   if (needs_stack_repair) {
332     // Save stack increment (also account for fixed framesize and rbp)
333     assert((sp_inc & (StackAlignmentInBytes-1)) == 0, "stack increment not aligned");
334     int real_frame_size = sp_inc + frame_size_in_bytes + wordSize;
335     movptr(Address(rsp, frame_size_in_bytes - wordSize), real_frame_size);
336   }
337   if (reset_orig_pc) {
338     // Zero orig_pc to detect deoptimization during buffering in the entry points
339     movptr(Address(rsp, sp_offset_for_orig_pc), 0);
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 doing an enter(). This matches the
346   // ordering of C2's stack overflow check / rsp decrement and allows
347   // the SharedRuntime stack overflow handling to be consistent
348   // between the two compilers.
349   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
350   generate_stack_overflow_check(bang_size_in_bytes);
351 
352   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
353 
354   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
355   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
356   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
357 
358   if (verified_inline_entry_label != nullptr) {
359     // Jump here from the scalarized entry points that already created the frame.
360     bind(*verified_inline_entry_label);
361   }
362 }
363 
364 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
365   if (breakAtEntry || VerifyFPU) {
366     // Verified Entry first instruction should be 5 bytes long for correct
367     // patching by patch_verified_entry().
368     //
369     // Breakpoint and VerifyFPU have one byte first instruction.
370     // Also first instruction will be one byte "push(rbp)" if stack banging
371     // code is not generated (see build_frame() above).
372     // For all these cases generate long instruction first.
373     fat_nop();
374   }
375   if (breakAtEntry) int3();
376   // build frame
377   IA32_ONLY( verify_FPU(0, "method_entry"); )
378 }
379 
380 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) {
381   assert(InlineTypePassFieldsAsArgs, "sanity");
382   // Make sure there is enough stack space for this method's activation.
383   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
384   generate_stack_overflow_check(bang_size_in_bytes);
385 
386   GrowableArray<SigEntry>* sig    = ces->sig();
387   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
388   VMRegPair* regs      = ces->regs();
389   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
390   int args_on_stack    = ces->args_on_stack();
391   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
392 
393   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
394   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
395   int args_passed = sig->length();
396   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
397 
398   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
399   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
400 
401   // The runtime call might safepoint, make sure nmethod entry barrier is executed
402   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
403   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
404   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
405 
406   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
407   movptr(rbx, (intptr_t)(ces->method()));
408   if (is_inline_ro_entry) {
409     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_no_receiver_id)));
410   } else {
411     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_id)));
412   }
413   int rt_call_offset = offset();
414 
415   // Remove the temp frame
416   addptr(rsp, frame_size_in_bytes);
417   pop(rbp);
418 
419   // Check if we need to extend the stack for packing
420   int sp_inc = 0;
421   if (args_on_stack > args_on_stack_cc) {
422     sp_inc = extend_stack_for_inline_args(args_on_stack);
423   }
424 
425   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
426                       args_passed_cc, args_on_stack_cc, regs_cc, // from
427                       args_passed, args_on_stack, regs,          // to
428                       sp_inc, rax);
429 
430   // Create the real frame. Below jump will then skip over the stack banging and frame
431   // setup code in the verified_inline_entry (which has a different real_frame_size).
432   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
433 
434   jmp(verified_inline_entry_label);
435   return rt_call_offset;
436 }
437 
438 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
439   // rbp, + 0: link
440   //     + 1: return address
441   //     + 2: argument with offset 0
442   //     + 3: argument with offset 1
443   //     + 4: ...
444 
445   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
446 }
447 
448 #ifndef PRODUCT
449 
450 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
451   if (!VerifyOops) return;
452   verify_oop_addr(Address(rsp, stack_offset));
453 }
454 
455 void C1_MacroAssembler::verify_not_null_oop(Register r) {
456   if (!VerifyOops) return;
457   Label not_null;
458   testptr(r, r);
459   jcc(Assembler::notZero, not_null);
460   stop("non-null oop required");
461   bind(not_null);
462   verify_oop(r);
463 }
464 
465 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
466 #ifdef ASSERT
467   if (inv_rax) movptr(rax, 0xDEAD);
468   if (inv_rbx) movptr(rbx, 0xDEAD);
469   if (inv_rcx) movptr(rcx, 0xDEAD);
470   if (inv_rdx) movptr(rdx, 0xDEAD);
471   if (inv_rsi) movptr(rsi, 0xDEAD);
472   if (inv_rdi) movptr(rdi, 0xDEAD);
473 #endif
474 }
475 
476 #endif // ifndef PRODUCT