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_RISCV_CONTINUATIONFREEZETHAW_RISCV_INLINE_HPP
 26 #define CPU_RISCV_CONTINUATIONFREEZETHAW_RISCV_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 - 2) = *(heap_sp - 2);
 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   intptr_t** link_addr = link_address<FKind>(f);
 64   intptr_t* sender_sp = (intptr_t*)(link_addr + 2); //  f.unextended_sp() + (fsize/wordSize); //
 65   address sender_pc = (address) *(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> frame FreezeBase::new_heap_frame(frame& f, frame& caller) {
 78   assert(FKind::is_instance(f), "");
 79   assert(!caller.is_interpreted_frame()
 80     || caller.unextended_sp() == (intptr_t*)caller.at(frame::interpreter_frame_last_sp_offset), "");
 81 
 82   intptr_t *sp, *fp; // sp is really our unextended_sp
 83   if (FKind::interpreted) {
 84     assert((intptr_t*)f.at(frame::interpreter_frame_last_sp_offset) == nullptr
 85       || f.unextended_sp() == (intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset), "");
 86     intptr_t locals_offset = *f.addr_at(frame::interpreter_frame_locals_offset);
 87     // If the caller.is_empty(), i.e. we're freezing into an empty chunk, then we set
 88     // the chunk's argsize in finalize_freeze and make room for it above the unextended_sp
 89     bool overlap_caller = caller.is_interpreted_frame() || caller.is_empty();
 90     fp = caller.unextended_sp() - 1 - locals_offset + (overlap_caller ? ContinuationHelper::InterpretedFrame::stack_argsize(f) : 0);
 91     sp = fp - (f.fp() - f.unextended_sp());
 92     assert(sp <= fp, "");
 93     assert(fp <= caller.unextended_sp(), "");
 94     caller.set_sp(fp + frame::sender_sp_offset);
 95 
 96     assert(_cont.tail()->is_in_chunk(sp), "");
 97 
 98     frame hf(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
 99     *hf.addr_at(frame::interpreter_frame_locals_offset) = locals_offset;
100     return hf;
101   } else {
102     // We need to re-read fp out of the frame because it may be an oop and we might have
103     // had a safepoint in finalize_freeze, after constructing f.
104     fp = *(intptr_t**)(f.sp() - 2);
105 
106     int fsize = FKind::size(f);
107     sp = caller.unextended_sp() - fsize;
108     if (caller.is_interpreted_frame()) {
109       // If the caller is interpreted, our stackargs are not supposed to overlap with it
110       // so we make more room by moving sp down by argsize
111       int argsize = FKind::stack_argsize(f);
112       sp -= argsize;
113     }
114     caller.set_sp(sp + fsize);
115 
116     assert(_cont.tail()->is_in_chunk(sp), "");
117 
118     return frame(sp, sp, fp, f.pc(), nullptr, nullptr, true /* on_heap */);
119   }
120 }
121 
122 void FreezeBase::adjust_interpreted_frame_unextended_sp(frame& f) {
123   assert((f.at(frame::interpreter_frame_last_sp_offset) != 0) || (f.unextended_sp() == f.sp()), "");
124   intptr_t* real_unextended_sp = (intptr_t*)f.at_relative_or_null(frame::interpreter_frame_last_sp_offset);
125   if (real_unextended_sp != nullptr) {
126     f.set_unextended_sp(real_unextended_sp); // can be null at a safepoint
127   }
128 }
129 
130 inline void FreezeBase::prepare_freeze_interpreted_top_frame(frame& f) {
131   assert(f.interpreter_frame_last_sp() == nullptr, "should be null for top frame");
132   f.interpreter_frame_set_last_sp(f.unextended_sp());
133 }
134 
135 inline void FreezeBase::relativize_interpreted_frame_metadata(const frame& f, const frame& hf) {
136   assert(hf.fp() == hf.unextended_sp() + (f.fp() - f.unextended_sp()), "");
137   assert((f.at(frame::interpreter_frame_last_sp_offset) != 0)
138     || (f.unextended_sp() == f.sp()), "");
139   assert(f.fp() > (intptr_t*)f.at_relative(frame::interpreter_frame_initial_sp_offset), "");
140 
141   // On RISCV, we may insert padding between the locals and the rest of the frame
142   // (see TemplateInterpreterGenerator::generate_normal_entry, and AbstractInterpreter::layout_activation)
143   // because we freeze the padding word (see recurse_freeze_interpreted_frame) in order to keep the same relativized
144   // locals value, we don't need to change the locals value here.
145 
146   // Make sure that last_sp is already relativized.
147   assert((intptr_t*)hf.at_relative(frame::interpreter_frame_last_sp_offset) == hf.unextended_sp(), "");
148 
149   // Make sure that monitor_block_top is already relativized.
150   assert(hf.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
151 
152   // extended_sp is already relativized by TemplateInterpreterGenerator::generate_normal_entry or
153   // AbstractInterpreter::layout_activation
154 
155   // The interpreter native wrapper code adds space in the stack equal to size_of_parameters()
156   // after the fixed part of the frame. For wait0 this is equal to 3 words (this + long parameter).
157   // We adjust by this size since otherwise the saved last sp will be less than the extended_sp.
158   DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
159   DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;)
160 
161   assert((hf.fp() - hf.unextended_sp()) == (f.fp() - f.unextended_sp()), "");
162   assert(hf.unextended_sp() == (intptr_t*)hf.at(frame::interpreter_frame_last_sp_offset), "");
163   assert(hf.unextended_sp() <= (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
164   assert(hf.unextended_sp() + extra_space >  (intptr_t*)hf.at(frame::interpreter_frame_extended_sp_offset), "");
165   assert(hf.fp()            >  (intptr_t*)hf.at(frame::interpreter_frame_initial_sp_offset), "");
166 #ifdef ASSERT
167   if (f.interpreter_frame_method()->max_locals() > 0) {
168     assert(hf.fp()          <= (intptr_t*)hf.at(frame::interpreter_frame_locals_offset), "");
169   }
170 #endif
171 }
172 
173 inline void FreezeBase::set_top_frame_metadata_pd(const frame& hf) {
174   stackChunkOop chunk = _cont.tail();
175   assert(chunk->is_in_chunk(hf.sp() - 1), "");
176   assert(chunk->is_in_chunk(hf.sp() - 2), "");
177 
178   *(hf.sp() - 1) = (intptr_t)hf.pc();
179 
180   intptr_t* fp_addr = hf.sp() - 2;
181   *fp_addr = hf.is_interpreted_frame() ? (intptr_t)(hf.fp() - fp_addr)
182                                        : (intptr_t)hf.fp();
183 }
184 
185 inline void FreezeBase::patch_pd(frame& hf, const frame& caller) {
186   if (caller.is_interpreted_frame()) {
187     assert(!caller.is_empty(), "");
188     patch_callee_link_relative(caller, caller.fp());
189   } else {
190     // If we're the bottom-most frame frozen in this freeze, the caller might have stayed frozen in the chunk,
191     // and its oop-containing fp fixed. We've now just overwritten it, so we must patch it back to its value
192     // as read from the chunk.
193     patch_callee_link(caller, caller.fp());
194   }
195 }
196 
197 inline intptr_t* AnchorMark::anchor_mark_set_pd() {
198   return _top_frame.sp();
199 }
200 
201 inline void AnchorMark::anchor_mark_clear_pd() {
202 }
203 
204 //////// Thaw
205 
206 // Fast path
207 
208 inline void ThawBase::prefetch_chunk_pd(void* start, int size) {
209   size <<= LogBytesPerWord;
210   Prefetch::read(start, size);
211   Prefetch::read(start, size - 64);
212 }
213 
214 template <typename ConfigT>
215 inline void Thaw<ConfigT>::patch_caller_links(intptr_t* sp, intptr_t* bottom) {
216   // Fast path depends on !PreserveFramePointer. See can_thaw_fast().
217   assert(!PreserveFramePointer, "Frame pointers need to be fixed");
218 }
219 
220 // Slow path
221 
222 inline frame ThawBase::new_entry_frame() {
223   intptr_t* sp = _cont.entrySP();
224   // TODO PERF: This finds code blob and computes deopt state
225   return frame(sp, sp, _cont.entryFP(), _cont.entryPC());
226 }
227 
228 template<typename FKind> frame ThawBase::new_stack_frame(const frame& hf, frame& caller, bool bottom) {
229   assert(FKind::is_instance(hf), "");
230   // The values in the returned frame object will be written into the callee's stack in patch.
231 
232   if (FKind::interpreted) {
233     intptr_t* heap_sp = hf.unextended_sp();
234     // If caller is interpreted it already made room for the callee arguments
235     int overlap = caller.is_interpreted_frame() ? ContinuationHelper::InterpretedFrame::stack_argsize(hf) : 0;
236     const int fsize = (int)(ContinuationHelper::InterpretedFrame::frame_bottom(hf) - hf.unextended_sp() - overlap);
237     intptr_t* frame_sp = caller.unextended_sp() - fsize;
238     intptr_t* fp = frame_sp + (hf.fp() - heap_sp);
239     if ((intptr_t)fp % frame::frame_alignment != 0) {
240       fp--;
241       frame_sp--;
242       log_develop_trace(continuations)("Adding internal interpreted frame alignment");
243     }
244     DEBUG_ONLY(intptr_t* unextended_sp = fp + *hf.addr_at(frame::interpreter_frame_last_sp_offset);)
245     assert(frame_sp == unextended_sp, "");
246     caller.set_sp(fp + frame::sender_sp_offset);
247     frame f(frame_sp, frame_sp, fp, hf.pc());
248     // we need to set the locals so that the caller of new_stack_frame() can call
249     // ContinuationHelper::InterpretedFrame::frame_bottom
250     // copy relativized locals from the heap frame
251     *f.addr_at(frame::interpreter_frame_locals_offset) = *hf.addr_at(frame::interpreter_frame_locals_offset);
252     assert((intptr_t)f.fp() % frame::frame_alignment == 0, "");
253     return f;
254   } else {
255     int fsize = FKind::size(hf);
256     intptr_t* frame_sp = caller.unextended_sp() - fsize;
257     if (bottom || caller.is_interpreted_frame()) {
258       int argsize = FKind::stack_argsize(hf);
259 
260       fsize += argsize;
261       frame_sp -= argsize;
262       caller.set_sp(caller.sp() - argsize);
263       assert(caller.sp() == frame_sp + (fsize-argsize), "");
264 
265       frame_sp = align(hf, frame_sp, caller, bottom);
266     }
267 
268     assert(hf.cb() != nullptr, "");
269     assert(hf.oop_map() != nullptr, "");
270     intptr_t* fp;
271     if (PreserveFramePointer) {
272       // we need to recreate a "real" frame pointer, pointing into the stack
273       fp = frame_sp + FKind::size(hf) - frame::sender_sp_offset;
274     } else {
275       fp = FKind::stub || FKind::native
276         // fp always points to the address above the pushed return pc. We need correct address.
277         ? frame_sp + fsize - frame::sender_sp_offset
278         // we need to re-read fp because it may be an oop and we might have fixed the frame.
279         : *(intptr_t**)(hf.sp() - 2);
280     }
281     // TODO PERF : this computes deopt state; is it necessary?
282     return frame(frame_sp, frame_sp, fp, hf.pc(), hf.cb(), hf.oop_map(), false);
283   }
284 }
285 
286 inline intptr_t* ThawBase::align(const frame& hf, intptr_t* frame_sp, frame& caller, bool bottom) {
287 #ifdef _LP64
288   if (((intptr_t)frame_sp & 0xf) != 0) {
289     assert(caller.is_interpreted_frame() || (bottom && hf.compiled_frame_stack_argsize() % 2 != 0), "");
290     frame_sp--;
291     caller.set_sp(caller.sp() - 1);
292   }
293   assert(is_aligned(frame_sp, frame::frame_alignment), "");
294 #endif
295 
296   return frame_sp;
297 }
298 
299 inline void ThawBase::patch_pd(frame& f, const frame& caller) {
300   patch_callee_link(caller, caller.fp());
301 }
302 
303 inline void ThawBase::patch_pd(frame& f, intptr_t* caller_sp) {
304   intptr_t* fp = caller_sp - frame::sender_sp_offset;
305   patch_callee_link(f, fp);
306 }
307 
308 inline intptr_t* ThawBase::push_cleanup_continuation() {
309   frame enterSpecial = new_entry_frame();
310   intptr_t* sp = enterSpecial.sp();
311 
312   sp[-1] = (intptr_t)ContinuationEntry::cleanup_pc();
313   sp[-2] = (intptr_t)enterSpecial.fp();
314 
315   log_develop_trace(continuations, preempt)("push_cleanup_continuation initial sp: " INTPTR_FORMAT " final sp: " INTPTR_FORMAT, p2i(sp + 2 * frame::metadata_words), p2i(sp));
316   return sp;
317 }
318 
319 inline intptr_t* ThawBase::push_preempt_adapter() {
320   Unimplemented();
321   return nullptr;
322 }
323 
324 inline void ThawBase::derelativize_interpreted_frame_metadata(const frame& hf, const frame& f) {
325   // Make sure that last_sp is kept relativized.
326   assert((intptr_t*)f.at_relative(frame::interpreter_frame_last_sp_offset) == f.unextended_sp(), "");
327 
328   // Make sure that monitor_block_top is still relativized.
329   assert(f.at_absolute(frame::interpreter_frame_monitor_block_top_offset) <= frame::interpreter_frame_initial_sp_offset, "");
330 
331   DEBUG_ONLY(Method* m = hf.interpreter_frame_method();)
332   DEBUG_ONLY(int extra_space = m->is_object_wait0() ? m->size_of_parameters() : 0;) // see comment in relativize_interpreted_frame_metadata()
333 
334   // Make sure that extended_sp is kept relativized.
335   assert((intptr_t*)f.at_relative(frame::interpreter_frame_extended_sp_offset) < f.unextended_sp() + extra_space, "");
336 }
337 
338 #endif // CPU_RISCV_CONTINUATIONFREEZETHAW_RISCV_INLINE_HPP