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