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