1 /*
2 * Copyright (c) 2019, 2024, 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
71 // Repair the sender sp if the frame has been extended
72 if (sender_cb->is_nmethod()) {
73 sender_sp = f.repair_sender_sp(sender_sp, link_addr);
74 }
75
76 return sender_cb != nullptr
77 ? frame(sender_sp, sender_sp, *link_addr, sender_pc, sender_cb,
78 slot == -1 ? nullptr : sender_cb->oop_map_for_slot(slot, sender_pc),
79 false /* on_heap ? */)
80 : frame(sender_sp, sender_sp, *link_addr, sender_pc);
81 }
82
83 template<typename FKind>
84 frame FreezeBase::new_heap_frame(frame& f, frame& caller, int size_adjust) {
85 assert(FKind::is_instance(f), "");
86 assert(!caller.is_interpreted_frame()
87 || caller.unextended_sp() == (intptr_t*)caller.at(frame::interpreter_frame_last_sp_offset), "");
88
89 intptr_t *sp, *fp; // sp is really our unextended_sp
90 if (FKind::interpreted) {
91 assert((intptr_t*)f.at(frame::interpreter_frame_last_sp_offset) == nullptr
92 || f.unextended_sp() == (intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset), "");
93 intptr_t locals_offset = *f.addr_at(frame::interpreter_frame_locals_offset);
94 // If the caller.is_empty(), i.e. we're freezing into an empty chunk, then we set
95 // the chunk's argsize in finalize_freeze and make room for it above the unextended_sp
96 bool overlap_caller = caller.is_interpreted_frame() || caller.is_empty();
97 fp = caller.unextended_sp() - 1 - locals_offset + (overlap_caller ? ContinuationHelper::InterpretedFrame::stack_argsize(f) : 0);
98 sp = fp - (f.fp() - f.unextended_sp());
99 assert(sp <= fp, "");
100 assert(fp <= caller.unextended_sp(), "");
101 caller.set_sp(fp + frame::sender_sp_offset);
102
103 assert(_cont.tail()->is_in_chunk(sp), "");
104
105 frame hf(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
106 // copy relativized locals from the stack frame
107 *hf.addr_at(frame::interpreter_frame_locals_offset) = locals_offset;
108 return hf;
109 } else {
110 // We need to re-read fp out of the frame because it may be an oop and we might have
111 // had a safepoint in finalize_freeze, after constructing f.
112 fp = *(intptr_t**)(f.sp() - frame::sender_sp_offset);
113
114 int fsize = FKind::size(f);
115 sp = caller.unextended_sp() - fsize - size_adjust;
116 if (caller.is_interpreted_frame() && size_adjust == 0) {
117 // If the caller is interpreted, our stackargs are not supposed to overlap with it
118 // so we make more room by moving sp down by argsize
119 int argsize = FKind::stack_argsize(f);
120 sp -= argsize;
121 }
122 caller.set_sp(sp + fsize);
123
124 assert(_cont.tail()->is_in_chunk(sp), "");
125
126 return frame(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
127 }
128 }
129
130 void FreezeBase::adjust_interpreted_frame_unextended_sp(frame& f) {
131 assert((f.at(frame::interpreter_frame_last_sp_offset) != 0) || (f.unextended_sp() == f.sp()), "");
132 intptr_t* real_unextended_sp = (intptr_t*)f.at_relative_or_null(frame::interpreter_frame_last_sp_offset);
133 if (real_unextended_sp != nullptr) {
134 f.set_unextended_sp(real_unextended_sp); // can be null at a safepoint
135 }
136 }
137
138 inline void FreezeBase::prepare_freeze_interpreted_top_frame(frame& f) {
139 assert(f.interpreter_frame_last_sp() == nullptr, "should be null for top frame");
140 f.interpreter_frame_set_last_sp(f.unextended_sp());
141 }
142
143 inline void FreezeBase::relativize_interpreted_frame_metadata(const frame& f, const frame& hf) {
144 assert(hf.fp() == hf.unextended_sp() + (f.fp() - f.unextended_sp()), "");
145 assert((f.at(frame::interpreter_frame_last_sp_offset) != 0)
146 || (f.unextended_sp() == f.sp()), "");
147 assert(f.fp() > (intptr_t*)f.at_relative(frame::interpreter_frame_initial_sp_offset), "");
148
149 // on AARCH64, we may insert padding between the locals and the rest of the frame
150 // (see TemplateInterpreterGenerator::generate_normal_entry, and AbstractInterpreter::layout_activation)
151 // because we freeze the padding word (see recurse_freeze_interpreted_frame) in order to keep the same relativized
152 // locals value, we don't need to change the locals value here.
153
154 // Make sure that last_sp is already relativized.
155 assert((intptr_t*)hf.at_relative(frame::interpreter_frame_last_sp_offset) == hf.unextended_sp(), "");
156
157 // Make sure that monitor_block_top is already relativized.
158 assert(hf.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
159
160 // extended_sp is already relativized by TemplateInterpreterGenerator::generate_normal_entry or
161 // AbstractInterpreter::layout_activation
162
163 // The interpreter native wrapper code adds space in the stack equal to size_of_parameters()
164 // after the fixed part of the frame. For wait0 this is equal to 3 words (this + long parameter).
165 // We adjust by this size since otherwise the saved last sp will be less than the extended_sp.
166 DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
167 DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;)
168
169 assert((hf.fp() - hf.unextended_sp()) == (f.fp() - f.unextended_sp()), "");
170 assert(hf.unextended_sp() == (intptr_t*)hf.at(frame::interpreter_frame_last_sp_offset), "");
171 assert(hf.unextended_sp() <= (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
172 assert(hf.unextended_sp() + extra_space > (intptr_t*)hf.at(frame::interpreter_frame_extended_sp_offset), "");
173 assert(hf.fp() > (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
174 assert(hf.fp() <= (intptr_t*)hf.at(frame::interpreter_frame_locals_offset), "");
175 }
176
177 inline void FreezeBase::set_top_frame_metadata_pd(const frame& hf) {
178 stackChunkOop chunk = _cont.tail();
179 assert(chunk->is_in_chunk(hf.sp() - 1), "");
180 assert(chunk->is_in_chunk(hf.sp() - frame::sender_sp_offset), "");
181
182 *(hf.sp() - 1) = (intptr_t)hf.pc();
183
184 intptr_t* fp_addr = hf.sp() - frame::sender_sp_offset;
185 *fp_addr = hf.is_interpreted_frame() ? (intptr_t)(hf.fp() - fp_addr)
186 : (intptr_t)hf.fp();
187 }
188
189 inline void FreezeBase::patch_pd(frame& hf, const frame& caller, bool is_bottom_frame) {
190 if (caller.is_interpreted_frame()) {
191 assert(!caller.is_empty(), "");
192 patch_callee_link_relative(caller, caller.fp());
193 } else if (is_bottom_frame && caller.pc() != nullptr) {
194 assert(caller.is_compiled_frame(), "");
195 // If we're the bottom-most frame frozen in this freeze, the caller might have stayed frozen in the chunk,
196 // and its oop-containing fp fixed. We've now just overwritten it, so we must patch it back to its value
197 // as read from the chunk.
198 patch_callee_link(caller, caller.fp());
199 }
200 }
201
202 //////// Thaw
203
204 // Fast path
205
206 inline void ThawBase::prefetch_chunk_pd(void* start, int size) {
207 size <<= LogBytesPerWord;
208 Prefetch::read(start, size);
209 Prefetch::read(start, size - 64);
210 }
211
212 template <typename ConfigT>
213 inline void Thaw<ConfigT>::patch_caller_links(intptr_t* sp, intptr_t* bottom) {
214 // Fast path depends on !PreserveFramePointer. See can_thaw_fast().
215 assert(!PreserveFramePointer, "Frame pointers need to be fixed");
216 }
217
218 // Slow path
219
220 inline frame ThawBase::new_entry_frame() {
221 intptr_t* sp = _cont.entrySP();
222 return frame(sp, sp, _cont.entryFP(), _cont.entryPC()); // TODO PERF: This finds code blob and computes deopt state
223 }
224
225 template<typename FKind> frame ThawBase::new_stack_frame(const frame& hf, frame& caller, bool bottom, int size_adjust) {
226 assert(FKind::is_instance(hf), "");
227 // The values in the returned frame object will be written into the callee's stack in patch.
228
229 if (FKind::interpreted) {
230 intptr_t* heap_sp = hf.unextended_sp();
231 // If caller is interpreted it already made room for the callee arguments
232 int overlap = caller.is_interpreted_frame() ? ContinuationHelper::InterpretedFrame::stack_argsize(hf) : 0;
233 const int fsize = (int)(ContinuationHelper::InterpretedFrame::frame_bottom(hf) - hf.unextended_sp() - overlap);
234 intptr_t* frame_sp = caller.unextended_sp() - fsize;
235 intptr_t* fp = frame_sp + (hf.fp() - heap_sp);
236 if ((intptr_t)fp % frame::frame_alignment != 0) {
237 fp--;
238 frame_sp--;
239 log_develop_trace(continuations)("Adding internal interpreted frame alignment");
240 }
241 DEBUG_ONLY(intptr_t* unextended_sp = fp + *hf.addr_at(frame::interpreter_frame_last_sp_offset);)
242 assert(frame_sp == unextended_sp, "");
243 caller.set_sp(fp + frame::sender_sp_offset);
244 frame f(frame_sp, frame_sp, fp, hf.pc());
245 // we need to set the locals so that the caller of new_stack_frame() can call
246 // ContinuationHelper::InterpretedFrame::frame_bottom
247 // copy relativized locals from the heap frame
248 *f.addr_at(frame::interpreter_frame_locals_offset) = *hf.addr_at(frame::interpreter_frame_locals_offset);
249 assert((intptr_t)f.fp() % frame::frame_alignment == 0, "");
250 return f;
251 } else {
252 int fsize = FKind::size(hf);
253 intptr_t* frame_sp = caller.unextended_sp() - fsize - size_adjust;
254 if (bottom || caller.is_interpreted_frame()) {
255 if (size_adjust == 0) {
256 int argsize = FKind::stack_argsize(hf);
257 frame_sp -= argsize;
258 }
259 frame_sp = align(hf, frame_sp, caller, bottom);
260 }
261 caller.set_sp(frame_sp + fsize);
262 assert(is_aligned(frame_sp, frame::frame_alignment), "");
263
264 assert(hf.cb() != nullptr, "");
265 assert(hf.oop_map() != nullptr, "");
266 intptr_t* fp;
267 if (PreserveFramePointer) {
268 // we need to recreate a "real" frame pointer, pointing into the stack
269 fp = frame_sp + fsize - frame::sender_sp_offset;
270 } else {
271 fp = FKind::stub || FKind::native
272 ? frame_sp + fsize - frame::sender_sp_offset // fp always points to the address below the pushed return pc. We need correct address.
273 : *(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.
274 }
275 return frame(frame_sp, frame_sp, fp, hf.pc(), hf.cb(), hf.oop_map(), false); // TODO PERF : this computes deopt state; is it necessary?
276 }
277 }
278
279 inline intptr_t* ThawBase::align(const frame& hf, intptr_t* frame_sp, frame& caller, bool bottom) {
280 #ifdef _LP64
281 if (((intptr_t)frame_sp & 0xf) != 0) {
282 assert(caller.is_interpreted_frame() || (bottom && hf.compiled_frame_stack_argsize() % 2 != 0), "");
283 frame_sp--;
284 }
285 assert(is_aligned(frame_sp, frame::frame_alignment), "");
286 #endif
287 return frame_sp;
288 }
289
290 inline void ThawBase::patch_pd(frame& f, const frame& caller) {
291 if (caller.is_interpreted_frame() || PreserveFramePointer) {
292 patch_callee_link(caller, caller.fp());
293 }
294 }
295
296 inline void ThawBase::patch_pd(frame& f, intptr_t* caller_sp) {
297 intptr_t* fp = caller_sp - frame::sender_sp_offset;
298 patch_callee_link(f, fp);
299 }
300
301 inline intptr_t* ThawBase::push_cleanup_continuation() {
302 frame enterSpecial = new_entry_frame();
303 intptr_t* sp = enterSpecial.sp();
304
305 sp[-1] = (intptr_t)ContinuationEntry::cleanup_pc();
306 sp[-2] = (intptr_t)enterSpecial.fp();
307
308 log_develop_trace(continuations, preempt)("push_cleanup_continuation initial sp: " INTPTR_FORMAT " final sp: " INTPTR_FORMAT, p2i(sp + 2 * frame::metadata_words), p2i(sp));
309 return sp;
310 }
311
312 inline void ThawBase::derelativize_interpreted_frame_metadata(const frame& hf, const frame& f) {
313 // Make sure that last_sp is kept relativized.
314 assert((intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset) == f.unextended_sp(), "");
315
316 // Make sure that monitor_block_top is still relativized.
317 assert(f.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
318
319 // Make sure that extended_sp is kept relativized.
320 DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
321 DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;) // see comment in relativize_interpreted_frame_metadata()
322 assert((intptr_t*)f.at_relative(frame::interpreter_frame_extended_sp_offset) < f.unextended_sp() + extra_space, "");
323 }
324
325 #endif // CPU_AARCH64_CONTINUATIONFREEZETHAW_AARCH64_INLINE_HPP