1 /*
2 * Copyright (c) 2003, 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 #ifdef COMPILER2
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "code/SCCache.hpp"
29 #include "code/vmreg.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "opto/runtime.hpp"
32 #include "runtime/sharedRuntime.hpp"
33 #include "runtime/stubRoutines.hpp"
34 #include "runtime/vframeArray.hpp"
35 #include "utilities/globalDefinitions.hpp"
36 #include "vmreg_x86.inline.hpp"
37
38 class SimpleRuntimeFrame {
39
40 public:
41
42 // Most of the runtime stubs have this simple frame layout.
43 // This class exists to make the layout shared in one place.
44 // Offsets are for compiler stack slots, which are jints.
45 enum layout {
46 // The frame sender code expects that rbp will be in the "natural" place and
47 // will override any oopMap setting for it. We must therefore force the layout
48 // so that it agrees with the frame sender code.
49 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
50 rbp_off2,
51 return_off, return_off2,
52 framesize
53 };
54 };
55
56 #define __ masm->
57
58 //------------------------------generate_uncommon_trap_blob--------------------
59 UncommonTrapBlob* OptoRuntime::generate_uncommon_trap_blob() {
60 // Allocate space for the code
61 ResourceMark rm;
62 // Setup code generation tools
63 const char* name = OptoRuntime::stub_name(OptoStubId::uncommon_trap_id);
64 CodeBuffer buffer(name, 2048, 1024);
65 MacroAssembler* masm = new MacroAssembler(&buffer);
66
67 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
68
69 address start = __ pc();
70
71 // Push self-frame. We get here with a return address on the
72 // stack, so rsp is 8-byte aligned until we allocate our frame.
73 __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog!
74
75 // No callee saved registers. rbp is assumed implicitly saved
76 __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
77
78 // compiler left unloaded_class_index in j_rarg0 move to where the
79 // runtime expects it.
80 __ movl(c_rarg1, j_rarg0);
81
82 __ set_last_Java_frame(noreg, noreg, nullptr, rscratch1);
83
84 // Call C code. Need thread but NOT official VM entry
85 // crud. We cannot block on this call, no GC can happen. Call should
86 // capture callee-saved registers as well as return values.
87 // Thread is in rdi already.
88 //
89 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
90
91 __ mov(c_rarg0, r15_thread);
92 __ movl(c_rarg2, Deoptimization::Unpack_uncommon_trap);
93 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
94
95 // Set an oopmap for the call site
96 OopMapSet* oop_maps = new OopMapSet();
97 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
98
99 // location of rbp is known implicitly by the frame sender code
100
101 oop_maps->add_gc_map(__ pc() - start, map);
102
103 __ reset_last_Java_frame(false);
104
105 // Load UnrollBlock* into rdi
106 __ mov(rdi, rax);
107
108 #ifdef ASSERT
109 { Label L;
110 __ cmpptr(Address(rdi, Deoptimization::UnrollBlock::unpack_kind_offset()),
111 Deoptimization::Unpack_uncommon_trap);
112 __ jcc(Assembler::equal, L);
113 __ stop("OptoRuntime::generate_uncommon_trap_blob: expected Unpack_uncommon_trap");
114 __ bind(L);
115 }
116 #endif
117
118 // Pop all the frames we must move/replace.
119 //
120 // Frame picture (youngest to oldest)
121 // 1: self-frame (no frame link)
122 // 2: deopting frame (no frame link)
123 // 3: caller of deopting frame (could be compiled/interpreted).
124
125 // Pop self-frame. We have no frame, and must rely only on rax and rsp.
126 __ addptr(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog!
127
128 // Pop deoptimized frame (int)
129 __ movl(rcx, Address(rdi,
130 Deoptimization::UnrollBlock::
131 size_of_deoptimized_frame_offset()));
132 __ addptr(rsp, rcx);
133
134 // rsp should be pointing at the return address to the caller (3)
135
136 // Pick up the initial fp we should save
137 // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
138 __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset()));
139
140 #ifdef ASSERT
141 // Compilers generate code that bang the stack by as much as the
142 // interpreter would need. So this stack banging should never
143 // trigger a fault. Verify that it does not on non product builds.
144 __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset()));
145 __ bang_stack_size(rbx, rcx);
146 #endif
147
148 // Load address of array of frame pcs into rcx (address*)
149 __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset()));
150
151 // Trash the return pc
152 __ addptr(rsp, wordSize);
153
154 // Load address of array of frame sizes into rsi (intptr_t*)
155 __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock:: frame_sizes_offset()));
156
157 // Counter
158 __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock:: number_of_frames_offset())); // (int)
159
160 // Now adjust the caller's stack to make up for the extra locals but
161 // record the original sp so that we can save it in the skeletal
162 // interpreter frame and the stack walking of interpreter_sender
163 // will get the unextended sp value and not the "real" sp value.
164
165 const Register sender_sp = r8;
166
167 __ mov(sender_sp, rsp);
168 __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock:: caller_adjustment_offset())); // (int)
169 __ subptr(rsp, rbx);
170
171 // Push interpreter frames in a loop
172 Label loop;
173 __ bind(loop);
174 __ movptr(rbx, Address(rsi, 0)); // Load frame size
175 __ subptr(rbx, 2 * wordSize); // We'll push pc and rbp by hand
176 __ pushptr(Address(rcx, 0)); // Save return address
177 __ enter(); // Save old & set new rbp
178 __ subptr(rsp, rbx); // Prolog
179 __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize),
180 sender_sp); // Make it walkable
181 // This value is corrected by layout_activation_impl
182 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
183 __ mov(sender_sp, rsp); // Pass sender_sp to next frame
184 __ addptr(rsi, wordSize); // Bump array pointer (sizes)
185 __ addptr(rcx, wordSize); // Bump array pointer (pcs)
186 __ decrementl(rdx); // Decrement counter
187 __ jcc(Assembler::notZero, loop);
188 __ pushptr(Address(rcx, 0)); // Save final return address
189
190 // Re-push self-frame
191 __ enter(); // Save old & set new rbp
192 __ subptr(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt);
193 // Prolog
194
195 // Use rbp because the frames look interpreted now
196 // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
197 // Don't need the precise return PC here, just precise enough to point into this code blob.
198 address the_pc = __ pc();
199 __ set_last_Java_frame(noreg, rbp, the_pc, rscratch1);
200
201 // Call C code. Need thread but NOT official VM entry
202 // crud. We cannot block on this call, no GC can happen. Call should
203 // restore return values to their stack-slots with the new SP.
204 // Thread is in rdi already.
205 //
206 // BasicType unpack_frames(JavaThread* thread, int exec_mode);
207
208 __ andptr(rsp, -(StackAlignmentInBytes)); // Align SP as required by ABI
209 __ mov(c_rarg0, r15_thread);
210 __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap);
211 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
212
213 // Set an oopmap for the call site
214 // Use the same PC we used for the last java frame
215 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
216
217 // Clear fp AND pc
218 __ reset_last_Java_frame(true);
219
220 // Pop self-frame.
221 __ leave(); // Epilog
222
223 // Jump to interpreter
224 __ ret(0);
225
226 // Make sure all code is generated
227 masm->flush();
228
229 return UncommonTrapBlob::create(&buffer, oop_maps,
230 SimpleRuntimeFrame::framesize >> 1);
231 }
232
233 //------------------------------generate_exception_blob---------------------------
234 // creates exception blob at the end
235 // Using exception blob, this code is jumped from a compiled method.
236 // (see emit_exception_handler in x86_64.ad file)
237 //
238 // Given an exception pc at a call we call into the runtime for the
239 // handler in this method. This handler might merely restore state
240 // (i.e. callee save registers) unwind the frame and jump to the
241 // exception handler for the nmethod if there is no Java level handler
242 // for the nmethod.
243 //
244 // This code is entered with a jmp.
245 //
246 // Arguments:
247 // rax: exception oop
248 // rdx: exception pc
249 //
250 // Results:
251 // rax: exception oop
252 // rdx: exception pc in caller or ???
253 // destination: exception handler of caller
254 //
255 // Note: the exception pc MUST be at a call (precise debug information)
256 // Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved.
257 //
258
259 ExceptionBlob* OptoRuntime::generate_exception_blob() {
260 assert(!OptoRuntime::is_callee_saved_register(RDX_num), "");
261 assert(!OptoRuntime::is_callee_saved_register(RAX_num), "");
262 assert(!OptoRuntime::is_callee_saved_register(RCX_num), "");
263
264 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
265
266 // Allocate space for the code
267 ResourceMark rm;
268 // Setup code generation tools
269 const char* name = OptoRuntime::stub_name(OptoStubId::exception_id);
270 CodeBuffer buffer(name, 2048, 1024);
271
272 int pc_offset = 0;
273 if (SCCache::load_exception_blob(&buffer, &pc_offset)) {
274 OopMapSet* oop_maps = new OopMapSet();
275 oop_maps->add_gc_map(pc_offset, new OopMap(SimpleRuntimeFrame::framesize, 0));
276
277 // Set exception blob
278 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
279 return _exception_blob;
280 }
281
282 MacroAssembler* masm = new MacroAssembler(&buffer);
283 address start = __ pc();
284
285 // Exception pc is 'return address' for stack walker
286 __ push(rdx);
287 __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog
288
289 // Save callee-saved registers. See x86_64.ad.
290
291 // rbp is an implicitly saved callee saved register (i.e., the calling
292 // convention will save/restore it in the prolog/epilog). Other than that
293 // there are no callee save registers now that adapter frames are gone.
294
295 __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
296
297 // Store exception in Thread object. We cannot pass any arguments to the
298 // handle_exception call, since we do not want to make any assumption
299 // about the size of the frame where the exception happened in.
300 // c_rarg0 is either rdi (Linux) or rcx (Windows).
301 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()),rax);
302 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
303
304 // This call does all the hard work. It checks if an exception handler
305 // exists in the method.
306 // If so, it returns the handler address.
307 // If not, it prepares for stack-unwinding, restoring the callee-save
308 // registers of the frame being removed.
309 //
310 // address OptoRuntime::handle_exception_C(JavaThread* thread)
311
312 // At a method handle call, the stack may not be properly aligned
313 // when returning with an exception.
314 address the_pc = __ pc();
315 __ set_last_Java_frame(noreg, noreg, the_pc, rscratch1);
316 __ mov(c_rarg0, r15_thread);
317 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
318 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
319
320 // Set an oopmap for the call site. This oopmap will only be used if we
321 // are unwinding the stack. Hence, all locations will be dead.
322 // Callee-saved registers will be the same as the frame above (i.e.,
323 // handle_exception_stub), since they were restored when we got the
324 // exception.
325
326 OopMapSet* oop_maps = new OopMapSet();
327
328 pc_offset = the_pc - start;
329 oop_maps->add_gc_map(pc_offset, new OopMap(SimpleRuntimeFrame::framesize, 0));
330
331 __ reset_last_Java_frame(false);
332
333 // Restore callee-saved registers
334
335 // rbp is an implicitly saved callee-saved register (i.e., the calling
336 // convention will save restore it in prolog/epilog) Other than that
337 // there are no callee save registers now that adapter frames are gone.
338
339 __ movptr(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt));
340
341 __ addptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog
342 __ pop(rdx); // No need for exception pc anymore
343
344 // rax: exception handler
345
346 // We have a handler in rax (could be deopt blob).
347 __ mov(r8, rax);
348
349 // Get the exception oop
350 __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
351 // Get the exception pc in case we are deoptimized
352 __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
353 #ifdef ASSERT
354 __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), NULL_WORD);
355 __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), NULL_WORD);
356 #endif
357 // Clear the exception oop so GC no longer processes it as a root.
358 __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), NULL_WORD);
359
360 // rax: exception oop
361 // r8: exception handler
362 // rdx: exception pc
363 // Jump to handler
364
365 __ jmp(r8);
366
367 // Make sure all code is generated
368 masm->flush();
369
370 SCCache::store_exception_blob(&buffer, pc_offset);
371 // Set exception blob
372 return ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
373 }
374 #endif // COMPILER2