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