8290154: [JVMCI] partially implement JVMCI for RISC-V

Reviewed-by: ihse, dnsimon, yadongwang

make/autoconf/jvm-features.m4

@@ -275,6 +275,8 @@ AC_DEFUN_ONCE([JVM_FEATURES_CHECK_JVMCI],
       AC_MSG_RESULT([yes])
     elif test "x$OPENJDK_TARGET_CPU" = "xaarch64"; then
       AC_MSG_RESULT([yes])
+    elif test "x$OPENJDK_TARGET_CPU" = "xriscv64"; then
+      AC_MSG_RESULT([yes])
     else
       AC_MSG_RESULT([no, $OPENJDK_TARGET_CPU])
       AVAILABLE=false

src/hotspot/cpu/riscv/jvmciCodeInstaller_riscv.cpp

@@ -0,0 +1,124 @@
+/*
+ * Copyright (c) 2022, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "precompiled.hpp"
+#include "asm/macroAssembler.hpp"
+#include "jvmci/jvmci.hpp"
+#include "jvmci/jvmciCodeInstaller.hpp"
+#include "jvmci/jvmciRuntime.hpp"
+#include "jvmci/jvmciCompilerToVM.hpp"
+#include "jvmci/jvmciJavaClasses.hpp"
+#include "oops/oop.inline.hpp"
+#include "runtime/handles.inline.hpp"
+#include "runtime/jniHandles.hpp"
+#include "runtime/sharedRuntime.hpp"
+#include "vmreg_riscv.inline.hpp"
+
+jint CodeInstaller::pd_next_offset(NativeInstruction* inst, jint pc_offset, JVMCI_TRAPS) {
+  address pc = (address) inst;
+  if (inst->is_call()) {
+    return pc_offset + NativeCall::instruction_size;
+  } else if (inst->is_jump()) {
+    return pc_offset + NativeJump::instruction_size;
+  } else if (inst->is_movptr()) {
+    return pc_offset + NativeMovConstReg::movptr_instruction_size;
+  } else {
+    JVMCI_ERROR_0("unsupported type of instruction for call site");
+  }
+}
+
+void CodeInstaller::pd_patch_OopConstant(int pc_offset, Handle& obj, bool compressed, JVMCI_TRAPS) {
+  address pc = _instructions->start() + pc_offset;
+  jobject value = JNIHandles::make_local(obj());
+  MacroAssembler::patch_oop(pc, cast_from_oop<address>(obj()));
+  int oop_index = _oop_recorder->find_index(value);
+  RelocationHolder rspec = oop_Relocation::spec(oop_index);
+  _instructions->relocate(pc, rspec);
+}
+
+void CodeInstaller::pd_patch_MetaspaceConstant(int pc_offset, HotSpotCompiledCodeStream* stream, u1 tag, JVMCI_TRAPS) {
+  address pc = _instructions->start() + pc_offset;
+  if (tag == PATCH_NARROW_KLASS) {
+    narrowKlass narrowOop = record_narrow_metadata_reference(_instructions, pc, stream, tag, JVMCI_CHECK);
+    MacroAssembler::pd_patch_instruction_size(pc, (address) (long) narrowOop);
+    JVMCI_event_3("relocating (narrow metaspace constant) at " PTR_FORMAT "/0x%x", p2i(pc), narrowOop);
+  } else {
+    NativeMovConstReg* move = nativeMovConstReg_at(pc);
+    void* reference = record_metadata_reference(_instructions, pc, stream, tag, JVMCI_CHECK);
+    move->set_data((intptr_t) reference);
+    JVMCI_event_3("relocating (metaspace constant) at " PTR_FORMAT "/" PTR_FORMAT, p2i(pc), p2i(reference));
+  }
+}
+
+void CodeInstaller::pd_patch_DataSectionReference(int pc_offset, int data_offset, JVMCI_TRAPS) {
+  address pc = _instructions->start() + pc_offset;
+  address dest = _constants->start() + data_offset;
+  _instructions->relocate(pc, section_word_Relocation::spec((address) dest, CodeBuffer::SECT_CONSTS));
+  JVMCI_event_3("relocating at " PTR_FORMAT " (+%d) with destination at %d", p2i(pc), pc_offset, data_offset);
+}
+
+void CodeInstaller::pd_relocate_ForeignCall(NativeInstruction* inst, jlong foreign_call_destination, JVMCI_TRAPS) {
+  address pc = (address) inst;
+  if (inst->is_jal()) {
+    NativeCall* call = nativeCall_at(pc);
+    call->set_destination((address) foreign_call_destination);
+    _instructions->relocate(call->instruction_address(), runtime_call_Relocation::spec());
+  } else if (inst->is_jump()) {
+    NativeJump* jump = nativeJump_at(pc);
+    jump->set_jump_destination((address) foreign_call_destination);
+    _instructions->relocate(jump->instruction_address(), runtime_call_Relocation::spec());
+  } else if (inst->is_movptr()) {
+    NativeMovConstReg* movptr = nativeMovConstReg_at(pc);
+    movptr->set_data((intptr_t) foreign_call_destination);
+    _instructions->relocate(movptr->instruction_address(), runtime_call_Relocation::spec());
+  } else {
+    JVMCI_ERROR("unknown call or jump instruction at " PTR_FORMAT, p2i(pc));
+  }
+  JVMCI_event_3("relocating (foreign call) at " PTR_FORMAT, p2i(inst));
+}
+
+void CodeInstaller::pd_relocate_JavaMethod(CodeBuffer &cbuf, methodHandle& method, jint pc_offset, JVMCI_TRAPS) {
+  Unimplemented();
+}
+
+void CodeInstaller::pd_relocate_poll(address pc, jint mark, JVMCI_TRAPS) {
+  Unimplemented();
+}
+
+// convert JVMCI register indices (as used in oop maps) to HotSpot registers
+VMReg CodeInstaller::get_hotspot_reg(jint jvmci_reg, JVMCI_TRAPS) {
+  if (jvmci_reg < Register::number_of_registers) {
+    return as_Register(jvmci_reg)->as_VMReg();
+  } else {
+    jint floatRegisterNumber = jvmci_reg - Register::number_of_registers;
+    if (floatRegisterNumber >= 0 && floatRegisterNumber < FloatRegister::number_of_registers) {
+      return as_FloatRegister(floatRegisterNumber)->as_VMReg();
+    }
+    JVMCI_ERROR_NULL("invalid register number: %d", jvmci_reg);
+  }
+}
+
+bool CodeInstaller::is_general_purpose_reg(VMReg hotspotRegister) {
+  return !(hotspotRegister->is_FloatRegister() || hotspotRegister->is_VectorRegister());
+}

src/hotspot/cpu/riscv/sharedRuntime_riscv.cpp

@@ -56,6 +56,9 @@
 #include "adfiles/ad_riscv.hpp"
 #include "opto/runtime.hpp"
 #endif
+#if INCLUDE_JVMCI
+#include "jvmci/jvmciJavaClasses.hpp"
+#endif
 
 #define __ masm->
 
@@ -208,6 +211,9 @@ void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
 #ifdef COMPILER2
   __ pop_CPU_state(_save_vectors, Matcher::scalable_vector_reg_size(T_BYTE));
 #else
+#if !INCLUDE_JVMCI
+  assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
+#endif
   __ pop_CPU_state(_save_vectors);
 #endif
   __ leave();
@@ -493,6 +499,18 @@ void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
   // Pre-load the register-jump target early, to schedule it better.
   __ ld(t1, Address(xmethod, in_bytes(Method::from_compiled_offset())));
 
+#if INCLUDE_JVMCI
+  if (EnableJVMCI) {
+    // check if this call should be routed towards a specific entry point
+    __ ld(t0, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
+    Label no_alternative_target;
+    __ beqz(t0, no_alternative_target);
+    __ mv(t1, t0);
+    __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
+    __ bind(no_alternative_target);
+  }
+#endif // INCLUDE_JVMCI
+
   // Now generate the shuffle code.
   for (int i = 0; i < total_args_passed; i++) {
     if (sig_bt[i] == T_VOID) {
@@ -1675,6 +1693,11 @@ void SharedRuntime::generate_deopt_blob() {
   ResourceMark rm;
   // Setup code generation tools
   int pad = 0;
+#if INCLUDE_JVMCI
+  if (EnableJVMCI) {
+    pad += 512; // Increase the buffer size when compiling for JVMCI
+  }
+#endif
   CodeBuffer buffer("deopt_blob", 2048 + pad, 1024);
   MacroAssembler* masm = new MacroAssembler(&buffer);
   int frame_size_in_words = -1;
@@ -1726,6 +1749,12 @@ void SharedRuntime::generate_deopt_blob() {
   __ j(cont);
 
   int reexecute_offset = __ pc() - start;
+#if INCLUDE_JVMCI && !defined(COMPILER1)
+  if (EnableJVMCI && UseJVMCICompiler) {
+    // JVMCI does not use this kind of deoptimization
+    __ should_not_reach_here();
+  }
+#endif
 
   // Reexecute case
   // return address is the pc describes what bci to do re-execute at
@@ -1736,6 +1765,44 @@ void SharedRuntime::generate_deopt_blob() {
   __ mvw(xcpool, Deoptimization::Unpack_reexecute); // callee-saved
   __ j(cont);
 
+#if INCLUDE_JVMCI
+  Label after_fetch_unroll_info_call;
+  int implicit_exception_uncommon_trap_offset = 0;
+  int uncommon_trap_offset = 0;
+
+  if (EnableJVMCI) {
+    implicit_exception_uncommon_trap_offset = __ pc() - start;
+
+    __ ld(ra, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
+    __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
+
+    uncommon_trap_offset = __ pc() - start;
+
+    // Save everything in sight.
+    reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
+    // fetch_unroll_info needs to call last_java_frame()
+    Label retaddr;
+    __ set_last_Java_frame(sp, noreg, retaddr, t0);
+
+    __ lw(c_rarg1, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
+    __ mvw(t0, -1);
+    __ sw(t0, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
+
+    __ mvw(xcpool, (int32_t)Deoptimization::Unpack_reexecute);
+    __ mv(c_rarg0, xthread);
+    __ orrw(c_rarg2, zr, xcpool); // exec mode
+    int32_t offset = 0;
+    __ la_patchable(t0, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)), offset);
+    __ jalr(x1, t0, offset);
+    __ bind(retaddr);
+    oop_maps->add_gc_map( __ pc()-start, map->deep_copy());
+
+    __ reset_last_Java_frame(false);
+
+    __ j(after_fetch_unroll_info_call);
+  } // EnableJVMCI
+#endif // INCLUDE_JVMCI
+
   int exception_offset = __ pc() - start;
 
   // Prolog for exception case
@@ -1829,6 +1896,12 @@ void SharedRuntime::generate_deopt_blob() {
 
   __ reset_last_Java_frame(false);
 
+#if INCLUDE_JVMCI
+  if (EnableJVMCI) {
+    __ bind(after_fetch_unroll_info_call);
+  }
+#endif
+
   // Load UnrollBlock* into x15
   __ mv(x15, x10);
 
@@ -1984,6 +2057,12 @@ void SharedRuntime::generate_deopt_blob() {
   _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
   assert(_deopt_blob != NULL, "create deoptimization blob fail!");
   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
+#if INCLUDE_JVMCI
+  if (EnableJVMCI) {
+    _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
+    _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
+  }
+#endif
 }
 
 // Number of stack slots between incoming argument block and the start of