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 //////// Thaw
204 
205 // Fast path
206 
207 inline void ThawBase::prefetch_chunk_pd(void* start, int size) {
208   size <<= LogBytesPerWord;
209   Prefetch::read(start, size);
210   Prefetch::read(start, size - 64);
211 }
212 
213 template <typename ConfigT>
214 inline void Thaw<ConfigT>::patch_caller_links(intptr_t* sp, intptr_t* bottom) {
215   // Fast path depends on !PreserveFramePointer. See can_thaw_fast().
216   assert(!PreserveFramePointer, "Frame pointers need to be fixed");
217 }
218 
219 // Slow path
220 
221 inline frame ThawBase::new_entry_frame() {
222   intptr_t* sp = _cont.entrySP();
223   return frame(sp, sp, _cont.entryFP(), _cont.entryPC()); // TODO PERF: This finds code blob and computes deopt state
224 }
225 
226 template<typename FKind> frame ThawBase::new_stack_frame(const frame& hf, frame& caller, bool bottom) {
227   assert(FKind::is_instance(hf), "");
228   // The values in the returned frame object will be written into the callee's stack in patch.
229 
230   if (FKind::interpreted) {
231     intptr_t* heap_sp = hf.unextended_sp();
232     // If caller is interpreted it already made room for the callee arguments
233     int overlap = caller.is_interpreted_frame() ? ContinuationHelper::InterpretedFrame::stack_argsize(hf) : 0;
234     const int fsize = (int)(ContinuationHelper::InterpretedFrame::frame_bottom(hf) - hf.unextended_sp() - overlap);
235     intptr_t* frame_sp = caller.unextended_sp() - fsize;
236     intptr_t* fp = frame_sp + (hf.fp() - heap_sp);
237     if ((intptr_t)fp % frame::frame_alignment != 0) {
238       fp--;
239       frame_sp--;
240       log_develop_trace(continuations)("Adding internal interpreted frame alignment");
241     }
242     DEBUG_ONLY(intptr_t* unextended_sp = fp + *hf.addr_at(frame::interpreter_frame_last_sp_offset);)
243     assert(frame_sp == unextended_sp, "");
244     caller.set_sp(fp + frame::sender_sp_offset);
245     frame f(frame_sp, frame_sp, fp, hf.pc());
246     // we need to set the locals so that the caller of new_stack_frame() can call
247     // ContinuationHelper::InterpretedFrame::frame_bottom
248     // copy relativized locals from the heap frame
249     *f.addr_at(frame::interpreter_frame_locals_offset) = *hf.addr_at(frame::interpreter_frame_locals_offset);
250     assert((intptr_t)f.fp() % frame::frame_alignment == 0, "");
251     return f;
252   } else {
253     int fsize = FKind::size(hf);
254     intptr_t* frame_sp = caller.unextended_sp() - fsize;
255     if (bottom || caller.is_interpreted_frame()) {
256       int argsize = FKind::stack_argsize(hf);
257 
258       fsize += argsize;
259       frame_sp   -= argsize;
260       caller.set_sp(caller.sp() - argsize);
261       assert(caller.sp() == frame_sp + (fsize-argsize), "");
262 
263       frame_sp = align(hf, frame_sp, caller, bottom);
264     }
265 
266     assert(hf.cb() != nullptr, "");
267     assert(hf.oop_map() != nullptr, "");
268     intptr_t* fp;
269     if (PreserveFramePointer) {
270       // we need to recreate a "real" frame pointer, pointing into the stack
271       fp = frame_sp + FKind::size(hf) - frame::sender_sp_offset;
272     } else {
273       fp = FKind::stub || FKind::native
274         ? frame_sp + fsize - frame::sender_sp_offset // fp always points to the address below the pushed return pc. We need correct address.
275         : *(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.
276     }
277     return frame(frame_sp, frame_sp, fp, hf.pc(), hf.cb(), hf.oop_map(), false); // TODO PERF : this computes deopt state; is it necessary?
278   }
279 }
280 
281 inline intptr_t* ThawBase::align(const frame& hf, intptr_t* frame_sp, frame& caller, bool bottom) {
282 #ifdef _LP64
283   if (((intptr_t)frame_sp & 0xf) != 0) {
284     assert(caller.is_interpreted_frame() || (bottom && hf.compiled_frame_stack_argsize() % 2 != 0), "");
285     frame_sp--;
286     caller.set_sp(caller.sp() - 1);
287   }
288   assert(is_aligned(frame_sp, frame::frame_alignment), "");
289 #endif
290 
291   return frame_sp;
292 }
293 
294 inline void ThawBase::patch_pd(frame& f, const frame& caller) {
295   patch_callee_link(caller, caller.fp());
296 }
297 
298 inline void ThawBase::patch_pd(frame& f, intptr_t* caller_sp) {
299   intptr_t* fp = caller_sp - frame::sender_sp_offset;
300   patch_callee_link(f, fp);
301 }
302 
303 inline intptr_t* ThawBase::push_cleanup_continuation() {
304   frame enterSpecial = new_entry_frame();
305   intptr_t* sp = enterSpecial.sp();
306 
307   sp[-1] = (intptr_t)ContinuationEntry::cleanup_pc();
308   sp[-2] = (intptr_t)enterSpecial.fp();
309 
310   log_develop_trace(continuations, preempt)("push_cleanup_continuation initial sp: " INTPTR_FORMAT " final sp: " INTPTR_FORMAT, p2i(sp + 2 * frame::metadata_words), p2i(sp));
311   return sp;
312 }
313 
314 inline void ThawBase::derelativize_interpreted_frame_metadata(const frame& hf, const frame& f) {
315   // Make sure that last_sp is kept relativized.
316   assert((intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset) == f.unextended_sp(), "");
317 
318   // Make sure that monitor_block_top is still relativized.
319   assert(f.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
320 
321   // Make sure that extended_sp is kept relativized.
322   DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
323   DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;) // see comment in relativize_interpreted_frame_metadata()
324   assert((intptr_t*)f.at_relative(frame::interpreter_frame_extended_sp_offset) < f.unextended_sp() + extra_space, "");
325 }
326 
327 #endif // CPU_AARCH64_CONTINUATIONFREEZETHAW_AARCH64_INLINE_HPP