1 /*
  2  * Copyright (c) 1999, 2021, 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 "gc/shared/barrierSet.hpp"
 29 #include "gc/shared/barrierSetAssembler.hpp"
 30 #include "gc/shared/collectedHeap.hpp"
 31 #include "gc/shared/tlab_globals.hpp"
 32 #include "interpreter/interpreter.hpp"
 33 #include "oops/arrayOop.hpp"
 34 #include "oops/markWord.hpp"
 35 #include "runtime/basicLock.hpp"
 36 #include "runtime/frame.inline.hpp"
 37 #include "runtime/os.hpp"
 38 #include "runtime/sharedRuntime.hpp"
 39 #include "runtime/stubRoutines.hpp"
 40 
 41 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 42   const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
 43   const int aligned_mask = BytesPerWord -1;
 44   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 45   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 46   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 47   Label done;
 48   int null_check_offset = -1;
 49 
 50   verify_oop(obj);
 51 
 52   // save object being locked into the BasicObjectLock
 53   movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
 54 
 55   null_check_offset = offset();
 56 
 57   if (DiagnoseSyncOnValueBasedClasses != 0) {
 58     load_klass(hdr, obj, rklass_decode_tmp);
 59     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 60     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 61     jcc(Assembler::notZero, slow_case);
 62   }
 63 
 64   // Load object header
 65   movptr(hdr, Address(obj, hdr_offset));
 66   // and mark it as unlocked
 67   orptr(hdr, markWord::unlocked_value);
 68   if (EnableValhalla) {
 69     // Mask inline_type bit such that we go to the slow path if object is an inline type
 70     andptr(hdr, ~((int) markWord::inline_type_bit_in_place));
 71   }
 72   // save unlocked object header into the displaced header location on the stack
 73   movptr(Address(disp_hdr, 0), hdr);
 74   // test if object header is still the same (i.e. unlocked), and if so, store the
 75   // displaced header address in the object header - if it is not the same, get the
 76   // object header instead
 77   MacroAssembler::lock(); // must be immediately before cmpxchg!
 78   cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 79   // if the object header was the same, we're done
 80   jcc(Assembler::equal, done);
 81   // if the object header was not the same, it is now in the hdr register
 82   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 83   //
 84   // 1) (hdr & aligned_mask) == 0
 85   // 2) rsp <= hdr
 86   // 3) hdr <= rsp + page_size
 87   //
 88   // these 3 tests can be done by evaluating the following expression:
 89   //
 90   // (hdr - rsp) & (aligned_mask - page_size)
 91   //
 92   // assuming both the stack pointer and page_size have their least
 93   // significant 2 bits cleared and page_size is a power of 2
 94   subptr(hdr, rsp);
 95   andptr(hdr, aligned_mask - os::vm_page_size());
 96   // for recursive locking, the result is zero => save it in the displaced header
 97   // location (NULL in the displaced hdr location indicates recursive locking)
 98   movptr(Address(disp_hdr, 0), hdr);
 99   // otherwise we don't care about the result and handle locking via runtime call
100   jcc(Assembler::notZero, slow_case);
101   // done
102   bind(done);
103   return null_check_offset;
104 }
105 
106 
107 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
108   const int aligned_mask = BytesPerWord -1;
109   const int hdr_offset = oopDesc::mark_offset_in_bytes();
110   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
111   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
112   Label done;
113 
114   // load displaced header
115   movptr(hdr, Address(disp_hdr, 0));
116   // if the loaded hdr is NULL we had recursive locking
117   testptr(hdr, hdr);
118   // if we had recursive locking, we are done
119   jcc(Assembler::zero, done);
120   // load object
121   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
122 
123   verify_oop(obj);
124   // test if object header is pointing to the displaced header, and if so, restore
125   // the displaced header in the object - if the object header is not pointing to
126   // the displaced header, get the object header instead
127   MacroAssembler::lock(); // must be immediately before cmpxchg!
128   cmpxchgptr(hdr, Address(obj, hdr_offset));
129   // if the object header was not pointing to the displaced header,
130   // we do unlocking via runtime call
131   jcc(Assembler::notEqual, slow_case);
132   // done
133   bind(done);
134 }
135 
136 
137 // Defines obj, preserves var_size_in_bytes
138 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
139   if (UseTLAB) {
140     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
141   } else {
142     eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
143   }
144 }
145 
146 
147 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
148   assert_different_registers(obj, klass, len);
149   Register tmp_encode_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
150   if (EnableValhalla) {
151     // Need to copy markWord::prototype header for klass
152     assert_different_registers(obj, klass, len, t1, t2);
153     movptr(t1, Address(klass, Klass::prototype_header_offset()));
154     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
155   } else {
156     // This assumes that all prototype bits fit in an int32_t
157     movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markWord::prototype().value());
158   }
159 #ifdef _LP64
160   if (UseCompressedClassPointers) { // Take care not to kill klass
161     movptr(t1, klass);
162     encode_klass_not_null(t1, tmp_encode_klass);
163     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
164   } else
165 #endif
166   {
167     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
168   }
169 
170   if (len->is_valid()) {
171     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
172   }
173 #ifdef _LP64
174   else if (UseCompressedClassPointers) {
175     xorptr(t1, t1);
176     store_klass_gap(obj, t1);
177   }
178 #endif
179 }
180 
181 
182 // preserves obj, destroys len_in_bytes
183 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
184   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
185   Label done;
186 
187   // len_in_bytes is positive and ptr sized
188   subptr(len_in_bytes, hdr_size_in_bytes);
189   jcc(Assembler::zero, done);
190   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
191   bind(done);
192 }
193 
194 
195 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
196   assert(obj == rax, "obj must be in rax, for cmpxchg");
197   assert_different_registers(obj, t1, t2); // XXX really?
198   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
199 
200   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
201 
202   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
203 }
204 
205 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) {
206   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
207          "con_size_in_bytes is not multiple of alignment");
208   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
209 
210   initialize_header(obj, klass, noreg, t1, t2);
211 
212   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
213     // clear rest of allocated space
214     const Register t1_zero = t1;
215     const Register index = t2;
216     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
217     if (var_size_in_bytes != noreg) {
218       mov(index, var_size_in_bytes);
219       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
220     } else if (con_size_in_bytes <= threshold) {
221       // use explicit null stores
222       // code size = 2 + 3*n bytes (n = number of fields to clear)
223       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
224       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
225         movptr(Address(obj, i), t1_zero);
226     } else if (con_size_in_bytes > hdr_size_in_bytes) {
227       // use loop to null out the fields
228       // code size = 16 bytes for even n (n = number of fields to clear)
229       // initialize last object field first if odd number of fields
230       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
231       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
232       // initialize last object field if constant size is odd
233       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
234         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
235       // initialize remaining object fields: rdx is a multiple of 2
236       { Label loop;
237         bind(loop);
238         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
239                t1_zero);
240         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
241                t1_zero);)
242         decrement(index);
243         jcc(Assembler::notZero, loop);
244       }
245     }
246   }
247 
248   if (CURRENT_ENV->dtrace_alloc_probes()) {
249     assert(obj == rax, "must be");
250     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
251   }
252 
253   verify_oop(obj);
254 }
255 
256 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
257   assert(obj == rax, "obj must be in rax, for cmpxchg");
258   assert_different_registers(obj, len, t1, t2, klass);
259 
260   // determine alignment mask
261   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
262 
263   // check for negative or excessive length
264   cmpptr(len, (int32_t)max_array_allocation_length);
265   jcc(Assembler::above, slow_case);
266 
267   const Register arr_size = t2; // okay to be the same
268   // align object end
269   movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
270   lea(arr_size, Address(arr_size, len, f));
271   andptr(arr_size, ~MinObjAlignmentInBytesMask);
272 
273   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
274 
275   initialize_header(obj, klass, len, t1, t2);
276 
277   // clear rest of allocated space
278   const Register len_zero = len;
279   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
280 
281   if (CURRENT_ENV->dtrace_alloc_probes()) {
282     assert(obj == rax, "must be");
283     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
284   }
285 
286   verify_oop(obj);
287 }
288 
289 
290 
291 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
292   verify_oop(receiver);
293   // explicit NULL check not needed since load from [klass_offset] causes a trap
294   // check against inline cache
295   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
296   int start_offset = offset();
297   Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
298 
299   if (UseCompressedClassPointers) {
300     load_klass(rscratch1, receiver, tmp_load_klass);
301     cmpptr(rscratch1, iCache);
302   } else {
303     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
304   }
305   // if icache check fails, then jump to runtime routine
306   // Note: RECEIVER must still contain the receiver!
307   jump_cc(Assembler::notEqual,
308           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
309   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
310   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
311 }
312 
313 void C1_MacroAssembler::build_frame_helper(int frame_size_in_bytes, int sp_inc, bool needs_stack_repair) {
314   push(rbp);
315   if (PreserveFramePointer) {
316     mov(rbp, rsp);
317   }
318 #if !defined(_LP64) && defined(COMPILER2)
319   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
320       // c2 leaves fpu stack dirty. Clean it on entry
321       empty_FPU_stack();
322     }
323 #endif // !_LP64 && COMPILER2
324   decrement(rsp, frame_size_in_bytes);
325 
326   if (needs_stack_repair) {
327     // Save stack increment (also account for fixed framesize and rbp)
328     assert((sp_inc & (StackAlignmentInBytes-1)) == 0, "stack increment not aligned");
329     int real_frame_size = sp_inc + frame_size_in_bytes + wordSize;
330     movptr(Address(rsp, frame_size_in_bytes - wordSize), real_frame_size);
331   }
332 }
333 
334 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) {
335   if (has_scalarized_args) {
336     // Initialize orig_pc to detect deoptimization during buffering in the entry points
337     movptr(Address(rsp, sp_offset_for_orig_pc - frame_size_in_bytes - wordSize), 0);
338   }
339   if (!needs_stack_repair && verified_inline_entry_label != NULL) {
340     bind(*verified_inline_entry_label);
341   }
342   // Make sure there is enough stack space for this method's activation.
343   // Note that we do this before doing an enter(). This matches the
344   // ordering of C2's stack overflow check / rsp decrement and allows
345   // the SharedRuntime stack overflow handling to be consistent
346   // between the two compilers.
347   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
348   generate_stack_overflow_check(bang_size_in_bytes);
349 
350   build_frame_helper(frame_size_in_bytes, 0, needs_stack_repair);
351 
352   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
353   bs->nmethod_entry_barrier(this);
354 
355   if (needs_stack_repair && verified_inline_entry_label != NULL) {
356     // Jump here from the scalarized entry points that require additional stack space
357     // for packing scalarized arguments and therefore already created the frame.
358     bind(*verified_inline_entry_label);
359   }
360 }
361 
362 void C1_MacroAssembler::verified_entry() {
363   if (C1Breakpoint || VerifyFPU) {
364     // Verified Entry first instruction should be 5 bytes long for correct
365     // patching by patch_verified_entry().
366     //
367     // C1Breakpoint and VerifyFPU have one byte first instruction.
368     // Also first instruction will be one byte "push(rbp)" if stack banging
369     // code is not generated (see build_frame() above).
370     // For all these cases generate long instruction first.
371     fat_nop();
372   }
373   if (C1Breakpoint)int3();
374   // build frame
375   IA32_ONLY( verify_FPU(0, "method_entry"); )
376 }
377 
378 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) {
379   assert(InlineTypePassFieldsAsArgs, "sanity");
380   // Make sure there is enough stack space for this method's activation.
381   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
382   generate_stack_overflow_check(bang_size_in_bytes);
383 
384   GrowableArray<SigEntry>* sig    = &ces->sig();
385   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? &ces->sig_cc_ro() : &ces->sig_cc();
386   VMRegPair* regs      = ces->regs();
387   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
388   int args_on_stack    = ces->args_on_stack();
389   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
390 
391   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
392   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
393   int args_passed = sig->length();
394   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
395 
396   // Check if we need to extend the stack for packing
397   int sp_inc = 0;
398   if (args_on_stack > args_on_stack_cc) {
399     sp_inc = extend_stack_for_inline_args(args_on_stack);
400   }
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_inc, ces->c1_needs_stack_repair());
404 
405   // Initialize orig_pc to detect deoptimization during buffering in below runtime call
406   movptr(Address(rsp, sp_offset_for_orig_pc), 0);
407 
408   // The runtime call might safepoint, make sure nmethod entry barrier is executed
409   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
410   bs->nmethod_entry_barrier(this);
411 
412   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
413   movptr(rbx, (intptr_t)(ces->method()));
414   if (is_inline_ro_entry) {
415     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_no_receiver_id)));
416   } else {
417     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_id)));
418   }
419   int rt_call_offset = offset();
420 
421   // Remove the temp frame
422   addptr(rsp, frame_size_in_bytes);
423   pop(rbp);
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   if (ces->c1_needs_stack_repair()) {
431     // Create the real frame. Below jump will then skip over the stack banging and frame
432     // setup code in the verified_inline_entry (which has a different real_frame_size).
433     build_frame_helper(frame_size_in_bytes, sp_inc, true);
434   }
435 
436   jmp(verified_inline_entry_label);
437   return rt_call_offset;
438 }
439 
440 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
441   // rbp, + 0: link
442   //     + 1: return address
443   //     + 2: argument with offset 0
444   //     + 3: argument with offset 1
445   //     + 4: ...
446 
447   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
448 }
449 
450 #ifndef PRODUCT
451 
452 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
453   if (!VerifyOops) return;
454   verify_oop_addr(Address(rsp, stack_offset));
455 }
456 
457 void C1_MacroAssembler::verify_not_null_oop(Register r) {
458   if (!VerifyOops) return;
459   Label not_null;
460   testptr(r, r);
461   jcc(Assembler::notZero, not_null);
462   stop("non-null oop required");
463   bind(not_null);
464   verify_oop(r);
465 }
466 
467 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
468 #ifdef ASSERT
469   if (inv_rax) movptr(rax, 0xDEAD);
470   if (inv_rbx) movptr(rbx, 0xDEAD);
471   if (inv_rcx) movptr(rcx, 0xDEAD);
472   if (inv_rdx) movptr(rdx, 0xDEAD);
473   if (inv_rsi) movptr(rsi, 0xDEAD);
474   if (inv_rdi) movptr(rdi, 0xDEAD);
475 #endif
476 }
477 
478 #endif // ifndef PRODUCT