1 /*
2 * Copyright (c) 2019, 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 #ifndef CPU_AARCH64_CONTINUATIONFREEZETHAW_AARCH64_INLINE_HPP
26 #define CPU_AARCH64_CONTINUATIONFREEZETHAW_AARCH64_INLINE_HPP
27
28 #include "code/codeBlob.inline.hpp"
29 #include "oops/stackChunkOop.inline.hpp"
30 #include "runtime/frame.hpp"
31 #include "runtime/frame.inline.hpp"
32
33
34 inline void patch_callee_link(const frame& f, intptr_t* fp) {
35 DEBUG_ONLY(intptr_t* orig = *ContinuationHelper::Frame::callee_link_address(f));
36 *ContinuationHelper::Frame::callee_link_address(f) = fp;
37 }
38
39 inline void patch_callee_link_relative(const frame& f, intptr_t* fp) {
40 intptr_t* la = (intptr_t*)ContinuationHelper::Frame::callee_link_address(f);
41 intptr_t new_value = fp - la;
42 *la = new_value;
43 }
44
45 ////// Freeze
46
47 // Fast path
48
49 inline void FreezeBase::patch_stack_pd(intptr_t* frame_sp, intptr_t* heap_sp) {
50 // copy the spilled fp from the heap to the stack
51 *(frame_sp - frame::sender_sp_offset) = *(heap_sp - frame::sender_sp_offset);
52 }
53
54 // Slow path
55
56 template<typename FKind>
57 inline frame FreezeBase::sender(const frame& f) {
58 assert(FKind::is_instance(f), "");
59 if (FKind::interpreted) {
60 return frame(f.sender_sp(), f.interpreter_frame_sender_sp(), f.link(), f.sender_pc());
61 }
62 intptr_t** link_addr = link_address<FKind>(f);
63
64 intptr_t* sender_sp = (intptr_t*)(link_addr + frame::sender_sp_offset); // f.unextended_sp() + (fsize/wordSize); //
65 address sender_pc = ContinuationHelper::return_address_at(sender_sp - 1);
66 assert(sender_sp != f.sp(), "must have changed");
67
68 int slot = 0;
69 CodeBlob* sender_cb = CodeCache::find_blob_and_oopmap(sender_pc, slot);
70 return sender_cb != nullptr
71 ? frame(sender_sp, sender_sp, *link_addr, sender_pc, sender_cb,
72 slot == -1 ? nullptr : sender_cb->oop_map_for_slot(slot, sender_pc),
73 false /* on_heap ? */)
74 : frame(sender_sp, sender_sp, *link_addr, sender_pc);
75 }
76
77 template<typename FKind>
78 frame FreezeBase::new_heap_frame(frame& f, frame& caller) {
79 assert(FKind::is_instance(f), "");
80 assert(!caller.is_interpreted_frame()
81 || caller.unextended_sp() == (intptr_t*)caller.at(frame::interpreter_frame_last_sp_offset), "");
82
83 intptr_t *sp, *fp; // sp is really our unextended_sp
84 if (FKind::interpreted) {
85 assert((intptr_t*)f.at(frame::interpreter_frame_last_sp_offset) == nullptr
86 || f.unextended_sp() == (intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset), "");
87 intptr_t locals_offset = *f.addr_at(frame::interpreter_frame_locals_offset);
88 // If the caller.is_empty(), i.e. we're freezing into an empty chunk, then we set
89 // the chunk's argsize in finalize_freeze and make room for it above the unextended_sp
90 bool overlap_caller = caller.is_interpreted_frame() || caller.is_empty();
91 fp = caller.unextended_sp() - 1 - locals_offset + (overlap_caller ? ContinuationHelper::InterpretedFrame::stack_argsize(f) : 0);
92 sp = fp - (f.fp() - f.unextended_sp());
93 assert(sp <= fp, "");
94 assert(fp <= caller.unextended_sp(), "");
95 caller.set_sp(fp + frame::sender_sp_offset);
96
97 assert(_cont.tail()->is_in_chunk(sp), "");
98
99 frame hf(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
100 // copy relativized locals from the stack frame
101 *hf.addr_at(frame::interpreter_frame_locals_offset) = locals_offset;
102 return hf;
103 } else {
104 // For a compiled frame we need to re-read fp out of the frame because it may be an
105 // oop and we might have had a safepoint in finalize_freeze, after constructing f.
106 // For stub/native frames the value is not used while frozen, and will be constructed again
107 // when thawing the frame (see ThawBase::new_stack_frame). We use a special bad address to
108 // help with debugging, particularly when inspecting frames and identifying invalid accesses.
109 fp = FKind::compiled ? *(intptr_t**)(f.sp() - frame::sender_sp_offset) : (intptr_t*)badAddressVal;
110
111 int fsize = FKind::size(f);
112 sp = caller.unextended_sp() - fsize;
113 if (caller.is_interpreted_frame()) {
114 // If the caller is interpreted, our stackargs are not supposed to overlap with it
115 // so we make more room by moving sp down by argsize
116 int argsize = FKind::stack_argsize(f);
117 sp -= argsize;
118 }
119 caller.set_sp(sp + fsize);
120
121 assert(_cont.tail()->is_in_chunk(sp), "");
122
123 return frame(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
124 }
125 }
126
127 void FreezeBase::adjust_interpreted_frame_unextended_sp(frame& f) {
128 assert((f.at(frame::interpreter_frame_last_sp_offset) != 0) || (f.unextended_sp() == f.sp()), "");
129 intptr_t* real_unextended_sp = (intptr_t*)f.at_relative_or_null(frame::interpreter_frame_last_sp_offset);
130 if (real_unextended_sp != nullptr) {
131 f.set_unextended_sp(real_unextended_sp); // can be null at a safepoint
132 }
133 }
134
135 inline void FreezeBase::prepare_freeze_interpreted_top_frame(frame& f) {
136 assert(f.interpreter_frame_last_sp() == nullptr, "should be null for top frame");
137 f.interpreter_frame_set_last_sp(f.unextended_sp());
138 }
139
140 inline void FreezeBase::relativize_interpreted_frame_metadata(const frame& f, const frame& hf) {
141 assert(hf.fp() == hf.unextended_sp() + (f.fp() - f.unextended_sp()), "");
142 assert((f.at(frame::interpreter_frame_last_sp_offset) != 0)
143 || (f.unextended_sp() == f.sp()), "");
144 assert(f.fp() > (intptr_t*)f.at_relative(frame::interpreter_frame_initial_sp_offset), "");
145
146 // on AARCH64, we may insert padding between the locals and the rest of the frame
147 // (see TemplateInterpreterGenerator::generate_normal_entry, and AbstractInterpreter::layout_activation)
148 // because we freeze the padding word (see recurse_freeze_interpreted_frame) in order to keep the same relativized
149 // locals value, we don't need to change the locals value here.
150
151 // Make sure that last_sp is already relativized.
152 assert((intptr_t*)hf.at_relative(frame::interpreter_frame_last_sp_offset) == hf.unextended_sp(), "");
153
154 // Make sure that monitor_block_top is already relativized.
155 assert(hf.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
156
157 // extended_sp is already relativized by TemplateInterpreterGenerator::generate_normal_entry or
158 // AbstractInterpreter::layout_activation
159
160 // The interpreter native wrapper code adds space in the stack equal to size_of_parameters()
161 // after the fixed part of the frame. For wait0 this is equal to 3 words (this + long parameter).
162 // We adjust by this size since otherwise the saved last sp will be less than the extended_sp.
163 DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
164 DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;)
165
166 assert((hf.fp() - hf.unextended_sp()) == (f.fp() - f.unextended_sp()), "");
167 assert(hf.unextended_sp() == (intptr_t*)hf.at(frame::interpreter_frame_last_sp_offset), "");
168 assert(hf.unextended_sp() <= (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
169 assert(hf.unextended_sp() + extra_space > (intptr_t*)hf.at(frame::interpreter_frame_extended_sp_offset), "");
170 assert(hf.fp() > (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
171 assert(hf.fp() <= (intptr_t*)hf.at(frame::interpreter_frame_locals_offset), "");
172 }
173
174 inline void FreezeBase::set_top_frame_metadata_pd(const frame& hf) {
175 stackChunkOop chunk = _cont.tail();
176 assert(chunk->is_in_chunk(hf.sp() - 1), "");
177 assert(chunk->is_in_chunk(hf.sp() - frame::sender_sp_offset), "");
178
179 *(hf.sp() - 1) = (intptr_t)hf.pc();
180
181 intptr_t* fp_addr = hf.sp() - frame::sender_sp_offset;
182 *fp_addr = hf.is_interpreted_frame() ? (intptr_t)(hf.fp() - fp_addr)
183 : (intptr_t)hf.fp();
184 }
185
186 inline void FreezeBase::patch_pd(frame& hf, const frame& caller) {
187 if (caller.is_interpreted_frame()) {
188 assert(!caller.is_empty(), "");
189 patch_callee_link_relative(caller, caller.fp());
190 } else {
191 // If we're the bottom-most frame frozen in this freeze, the caller might have stayed frozen in the chunk,
192 // and its oop-containing fp fixed. We've now just overwritten it, so we must patch it back to its value
193 // as read from the chunk.
194 patch_callee_link(caller, caller.fp());
195 }
196 }
197
198 inline void FreezeBase::patch_pd_unused(intptr_t* sp) {
199 intptr_t* fp_addr = sp - frame::sender_sp_offset;
200 *fp_addr = badAddressVal;
201 }
202
203 inline intptr_t* AnchorMark::anchor_mark_set_pd() {
204 intptr_t* sp = _top_frame.sp();
205 if (_top_frame.is_interpreted_frame()) {
206 // In case the top frame is interpreted we need to set up the anchor using
207 // the last_sp saved in the frame (remove possible alignment added while
208 // thawing, see ThawBase::finish_thaw()). We also clear last_sp to match
209 // the behavior when calling the VM from the interpreter (we check for this
210 // in FreezeBase::prepare_freeze_interpreted_top_frame, which can be reached
211 // if preempting again at redo_vmcall()).
212 _last_sp_from_frame = _top_frame.interpreter_frame_last_sp();
213 assert(_last_sp_from_frame != nullptr, "");
214 _top_frame.interpreter_frame_set_last_sp(nullptr);
215 if (sp != _last_sp_from_frame) {
216 // We need to move up return pc and fp. They will be read next in
217 // set_anchor() and set as _last_Java_pc and _last_Java_fp respectively.
218 _last_sp_from_frame[-1] = (intptr_t)_top_frame.pc();
219 _last_sp_from_frame[-2] = (intptr_t)_top_frame.fp();
220 }
221 _is_interpreted = true;
222 sp = _last_sp_from_frame;
223 }
224 return sp;
225 }
226
227 inline void AnchorMark::anchor_mark_clear_pd() {
228 if (_is_interpreted) {
229 // Restore last_sp_from_frame and possibly overwritten pc.
230 _top_frame.interpreter_frame_set_last_sp(_last_sp_from_frame);
231 intptr_t* sp = _top_frame.sp();
232 if (sp != _last_sp_from_frame) {
233 sp[-1] = (intptr_t)_top_frame.pc();
234 }
235 }
236 }
237
238 //////// Thaw
239
240 // Fast path
241
242 inline void ThawBase::prefetch_chunk_pd(void* start, int size) {
243 size <<= LogBytesPerWord;
244 Prefetch::read(start, size);
245 Prefetch::read(start, size - 64);
246 }
247
248 template <typename ConfigT>
249 inline void Thaw<ConfigT>::patch_caller_links(intptr_t* sp, intptr_t* bottom) {
250 // Fast path depends on !PreserveFramePointer. See can_thaw_fast().
251 assert(!PreserveFramePointer, "Frame pointers need to be fixed");
252 }
253
254 // Slow path
255
256 inline frame ThawBase::new_entry_frame() {
257 intptr_t* sp = _cont.entrySP();
258 return frame(sp, sp, _cont.entryFP(), _cont.entryPC()); // TODO PERF: This finds code blob and computes deopt state
259 }
260
261 template<typename FKind> frame ThawBase::new_stack_frame(const frame& hf, frame& caller, bool bottom) {
262 assert(FKind::is_instance(hf), "");
263 // The values in the returned frame object will be written into the callee's stack in patch.
264
265 if (FKind::interpreted) {
266 intptr_t* heap_sp = hf.unextended_sp();
267 // If caller is interpreted it already made room for the callee arguments
268 int overlap = caller.is_interpreted_frame() ? ContinuationHelper::InterpretedFrame::stack_argsize(hf) : 0;
269 const int fsize = (int)(ContinuationHelper::InterpretedFrame::frame_bottom(hf) - hf.unextended_sp() - overlap);
270 intptr_t* frame_sp = caller.unextended_sp() - fsize;
271 intptr_t* fp = frame_sp + (hf.fp() - heap_sp);
272 if ((intptr_t)fp % frame::frame_alignment != 0) {
273 fp--;
274 frame_sp--;
275 log_develop_trace(continuations)("Adding internal interpreted frame alignment");
276 }
277 DEBUG_ONLY(intptr_t* unextended_sp = fp + *hf.addr_at(frame::interpreter_frame_last_sp_offset);)
278 assert(frame_sp == unextended_sp, "");
279 caller.set_sp(fp + frame::sender_sp_offset);
280 frame f(frame_sp, frame_sp, fp, hf.pc());
281 // we need to set the locals so that the caller of new_stack_frame() can call
282 // ContinuationHelper::InterpretedFrame::frame_bottom
283 // copy relativized locals from the heap frame
284 *f.addr_at(frame::interpreter_frame_locals_offset) = *hf.addr_at(frame::interpreter_frame_locals_offset);
285 assert((intptr_t)f.fp() % frame::frame_alignment == 0, "");
286 return f;
287 } else {
288 int fsize = FKind::size(hf);
289 intptr_t* frame_sp = caller.unextended_sp() - fsize;
290 if (bottom || caller.is_interpreted_frame()) {
291 int argsize = FKind::stack_argsize(hf);
292
293 fsize += argsize;
294 frame_sp -= argsize;
295 caller.set_sp(caller.sp() - argsize);
296 assert(caller.sp() == frame_sp + (fsize-argsize), "");
297
298 frame_sp = align(hf, frame_sp, caller, bottom);
299 }
300
301 assert(hf.cb() != nullptr, "");
302 assert(hf.oop_map() != nullptr, "");
303 intptr_t* fp;
304 if (PreserveFramePointer) {
305 // we need to recreate a "real" frame pointer, pointing into the stack
306 fp = frame_sp + FKind::size(hf) - frame::sender_sp_offset;
307 } else {
308 fp = FKind::stub || FKind::native
309 ? frame_sp + fsize - frame::sender_sp_offset // fp always points to the address below the pushed return pc. We need correct address.
310 : *(intptr_t**)(hf.sp() - frame::sender_sp_offset); // we need to re-read fp because it may be an oop and we might have fixed the frame.
311 }
312 return frame(frame_sp, frame_sp, fp, hf.pc(), hf.cb(), hf.oop_map(), false); // TODO PERF : this computes deopt state; is it necessary?
313 }
314 }
315
316 inline intptr_t* ThawBase::align(const frame& hf, intptr_t* frame_sp, frame& caller, bool bottom) {
317 #ifdef _LP64
318 if (((intptr_t)frame_sp & 0xf) != 0) {
319 assert(caller.is_interpreted_frame() || (bottom && hf.compiled_frame_stack_argsize() % 2 != 0), "");
320 frame_sp--;
321 caller.set_sp(caller.sp() - 1);
322 }
323 assert(is_aligned(frame_sp, frame::frame_alignment), "");
324 #endif
325
326 return frame_sp;
327 }
328
329 inline void ThawBase::patch_pd(frame& f, const frame& caller) {
330 patch_callee_link(caller, caller.fp());
331 }
332
333 inline void ThawBase::patch_pd(frame& f, intptr_t* caller_sp) {
334 intptr_t* fp = caller_sp - frame::sender_sp_offset;
335 patch_callee_link(f, fp);
336 }
337
338 inline intptr_t* ThawBase::push_cleanup_continuation() {
339 frame enterSpecial = new_entry_frame();
340 intptr_t* sp = enterSpecial.sp();
341
342 // We only need to set the return pc. rfp will be restored back in gen_continuation_enter().
343 sp[-1] = (intptr_t)ContinuationEntry::cleanup_pc();
344 return sp;
345 }
346
347 inline intptr_t* ThawBase::push_preempt_adapter() {
348 frame enterSpecial = new_entry_frame();
349 intptr_t* sp = enterSpecial.sp();
350
351 // We only need to set the return pc. rfp will be restored back in generate_cont_preempt_stub().
352 sp[-1] = (intptr_t)StubRoutines::cont_preempt_stub();
353 return sp;
354 }
355
356 inline void ThawBase::derelativize_interpreted_frame_metadata(const frame& hf, const frame& f) {
357 // Make sure that last_sp is kept relativized.
358 assert((intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset) == f.unextended_sp(), "");
359
360 // Make sure that monitor_block_top is still relativized.
361 assert(f.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
362
363 // Make sure that extended_sp is kept relativized.
364 DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
365 DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;) // see comment in relativize_interpreted_frame_metadata()
366 assert((intptr_t*)f.at_relative(frame::interpreter_frame_extended_sp_offset) < f.unextended_sp() + extra_space, "");
367 }
368
369 #endif // CPU_AARCH64_CONTINUATIONFREEZETHAW_AARCH64_INLINE_HPP