adds functionality, adds working asmjit

src/iss/arch/riscv_hart_m_p.h

@@ -320,6 +320,8 @@ protected:
 
         unsigned get_reg_num() override { return traits<BASE>::NUM_REGS; }
 
+        unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
+
         riscv_hart_m_p<BASE, FEAT>& arch;
     };
 
@@ -822,7 +824,8 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const acc
                 x |= 0x80; // set pll lock upon writing
                 return iss::Ok;
             } break;
-            default: {}
+            default: {
+            }
             }
         } break;
         case traits<BASE>::CSR: {

src/iss/arch/riscv_hart_msu_vp.h

@@ -371,6 +371,8 @@ protected:
 
         unsigned get_reg_num() override { return traits<BASE>::NUM_REGS; }
 
+        unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
+
         riscv_hart_msu_vp<BASE>& arch;
     };
 
@@ -802,7 +804,8 @@ iss::status riscv_hart_msu_vp<BASE>::write(const address_type type, const access
                 x |= 0x80; // set pll lock upon writing
                 return iss::Ok;
             } break;
-            default: {}
+            default: {
+            }
             }
         } break;
         case traits<BASE>::CSR: {
@@ -1225,9 +1228,9 @@ template <typename BASE> typename riscv_hart_msu_vp<BASE>::phys_addr_t riscv_har
                 break;
             } else if(!(pte & PTE_V) || (!(pte & PTE_R) && (pte & PTE_W))) {
                 break;
-            } else if(type == iss::access_type::FETCH ? !(pte & PTE_X)
+            } else if(type == iss::access_type::FETCH  ? !(pte & PTE_X)
-                                                      : type == iss::access_type::READ ? !(pte & PTE_R) && !(mxr && (pte & PTE_X))
+                      : type == iss::access_type::READ ? !(pte & PTE_R) && !(mxr && (pte & PTE_X))
-                                                                                       : !((pte & PTE_R) && (pte & PTE_W))) {
+                                                       : !((pte & PTE_R) && (pte & PTE_W))) {
                 break;
             } else if((ppn & ((reg_t(1) << ptshift) - 1)) != 0) {
                 break;

src/iss/arch/riscv_hart_mu_p.h

@@ -347,6 +347,8 @@ protected:
 
         unsigned get_reg_num() override { return traits<BASE>::NUM_REGS; }
 
+        unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
+
         riscv_hart_mu_p<BASE, FEAT>& arch;
     };
 
@@ -1005,7 +1007,8 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::write(const address_type type, const ac
                 x |= 0x80; // set pll lock upon writing
                 return iss::Ok;
             } break;
-            default: {}
+            default: {
+            }
             }
         } break;
         case traits<BASE>::CSR: {

src/vm/asmjit/helper_func.h

@@ -19,7 +19,7 @@ x86::Mem get_reg_ptr(jit_holder& jh, unsigned idx) {
     }
 }
 x86::Gp get_reg_for(jit_holder& jh, unsigned idx) {
-    // can check for regs in jh and return them instead of creating new ones
+    // TODO can check for regs in jh and return them instead of creating new ones
     switch(traits::reg_bit_widths[idx]) {
     case 8:
         return jh.cc.newInt8();
@@ -82,7 +82,9 @@ void gen_instr_prologue(jit_holder& jh, addr_t pc) {
     cc.mov(get_reg_ptr(jh, traits::PC), jh.next_pc);
 
     cc.comment("\n//*trap_state=*pending_trap;");
-    cc.mov(get_reg_ptr(jh, traits::PENDING_TRAP), jh.trap_state);
+    x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
+    cc.mov(current_trap_state, get_reg_ptr(jh, traits::TRAP_STATE));
+    cc.mov(get_reg_ptr(jh, traits::PENDING_TRAP), current_trap_state);
 
     cc.comment("\n//increment *next_pc");
     cc.mov(jh.next_pc, pc);
@@ -91,20 +93,20 @@ void gen_instr_epilogue(jit_holder& jh) {
     auto& cc = jh.cc;
 
     cc.comment("\n//if(*trap_state!=0) goto trap_entry;");
-    cc.test(jh.trap_state, jh.trap_state);
+    x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
-    cc.jnz(jh.trap_entry);
+    cc.mov(current_trap_state, get_reg_ptr(jh, traits::TRAP_STATE));
+    cc.cmp(current_trap_state, 0);
+    cc.jne(jh.trap_entry);
 
-    // Does this need to be done after every single instruction?
+    // TODO: Does not need to be done for every instruction, only when needed
     cc.comment("\n//write back regs to mem");
     write_reg_to_mem(jh, jh.pc, traits::PC);
     write_reg_to_mem(jh, jh.next_pc, traits::NEXT_PC);
-    write_reg_to_mem(jh, jh.trap_state, traits::TRAP_STATE);
 }
 void gen_block_prologue(jit_holder& jh) override {
 
     jh.pc = load_reg_from_mem(jh, traits::PC);
     jh.next_pc = load_reg_from_mem(jh, traits::NEXT_PC);
-    jh.trap_state = load_reg_from_mem(jh, traits::TRAP_STATE);
 }
 void gen_block_epilogue(jit_holder& jh) override {
     x86::Compiler& cc = jh.cc;
@@ -112,7 +114,12 @@ void gen_block_epilogue(jit_holder& jh) override {
     cc.ret(jh.next_pc);
 
     cc.bind(jh.trap_entry);
-    cc.comment("\n//enter_trap(core_ptr, *trap_state, *pc, 0);");
+    cc.comment("\n//Prepare for enter_trap;");
+    // Make sure cached values are written back
+    cc.comment("\n//write back regs to mem");
+    write_reg_to_mem(jh, jh.pc, traits::PC);
+    write_reg_to_mem(jh, jh.next_pc, traits::NEXT_PC);
+    this->gen_sync(jh, POST_SYNC, -1);
 
     x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
     cc.mov(current_trap_state, get_reg_ptr(jh, traits::TRAP_STATE));
@@ -121,23 +128,43 @@ void gen_block_epilogue(jit_holder& jh) override {
     cc.mov(current_pc, get_reg_ptr(jh, traits::PC));
 
     x86::Gp instr = cc.newInt32("instr");
-    cc.mov(instr, 0);
+    cc.mov(instr, 0); // this is not correct
+    cc.comment("\n//enter trap call;");
     InvokeNode* call_enter_trap;
     cc.invoke(&call_enter_trap, &enter_trap, FuncSignatureT<uint64_t, void*, uint64_t, uint64_t, uint64_t>());
     call_enter_trap->setArg(0, jh.arch_if_ptr);
     call_enter_trap->setArg(1, current_trap_state);
     call_enter_trap->setArg(2, current_pc);
     call_enter_trap->setArg(3, instr);
+
+    x86::Gp current_next_pc = get_reg_for(jh, traits::NEXT_PC);
+    cc.mov(current_next_pc, get_reg_ptr(jh, traits::NEXT_PC));
+    cc.mov(jh.next_pc, current_next_pc);
+
     cc.comment("\n//*last_branch = std::numeric_limits<uint32_t>::max();");
     cc.mov(get_reg_ptr(jh, traits::LAST_BRANCH), std::numeric_limits<uint32_t>::max());
     cc.comment("\n//return *next_pc;");
     cc.ret(jh.next_pc);
 }
-// TODO implement
+/*
-
+    inline void raise(uint16_t trap_id, uint16_t cause){
-void gen_raise(jit_holder& jh, uint16_t trap_id, uint16_t cause) { jh.cc.comment("//gen_raise"); }
+        auto trap_val =  0x80ULL << 24 | (cause << 16) | trap_id;
-void gen_wait(jit_holder& jh, unsigned type) { jh.cc.comment("//gen_wait"); }
+        this->core.reg.trap_state = trap_val;
-void gen_leave(jit_holder& jh, unsigned lvl) { jh.cc.comment("//gen_leave"); }
+        this->template get_reg<uint32_t>(traits::NEXT_PC) = std::numeric_limits<uint32_t>::max();
+    }
+*/
+inline void gen_raise(jit_holder& jh, uint16_t trap_id, uint16_t cause) {
+    auto& cc = jh.cc;
+    cc.comment("//gen_raise");
+    auto tmp1 = get_reg_for(jh, traits::TRAP_STATE);
+    cc.mov(tmp1, 0x80ULL << 24 | (cause << 16) | trap_id);
+    cc.mov(get_reg_ptr(jh, traits::TRAP_STATE), tmp1);
+    auto tmp2 = get_reg_for(jh, traits::NEXT_PC);
+    cc.mov(tmp2, std::numeric_limits<uint32_t>::max());
+    cc.mov(get_reg_ptr(jh, traits::NEXT_PC), tmp2);
+}
+inline void gen_wait(jit_holder& jh, unsigned type) { jh.cc.comment("//gen_wait"); }
+inline void gen_leave(jit_holder& jh, unsigned lvl) { jh.cc.comment("//gen_leave"); }
 
 enum operation { add, sub, band, bor, bxor, shl, sar, shr };
 
@@ -342,24 +369,12 @@ x86::Gp gen_operation(jit_holder& jh, binary_operation op, x86::Gp a) {
     return a;
 }
 
-/* template <typename T>
-inline typename std::enable_if_t<std::is_unsigned<T>::value, x86::Gp> gen_ext(jit_holder& jh, T val, unsigned size, bool
-is_signed) const { auto val_reg = get_reg_for(jh, sizeof(val)*8); auto tmp = get_reg_for(jh, size); jh.cc.mov(val_reg,
-val); if(is_signed) jh.cc.movsx(tmp, val_reg); else jh.cc.movzx(tmp,val_reg); return tmp;
-}
-
-template <typename T>
-inline typename std::enable_if_t<std::is_signed<T>::value, x86::Gp> gen_ext(jit_holder& jh, T val, unsigned size, bool
-is_signed) const { auto val_reg = get_reg_for(jh, sizeof(val)*8); auto tmp = get_reg_for(jh, size); jh.cc.mov(val_reg,
-val); if(is_signed) jh.cc.movsx(tmp, val_reg); else jh.cc.movzx(tmp,val_reg); return tmp;
-} */
 template <typename T, typename = std::enable_if_t<std::is_integral<T>::value>>
 inline x86::Gp gen_ext(jit_holder& jh, T val, unsigned size, bool is_signed) {
-    auto val_reg = get_reg_for(jh, sizeof(val) * 8);
+    auto val_reg = get_reg_for(jh, sizeof(val) * 8, is_signed);
     jh.cc.mov(val_reg, val);
     return gen_ext(jh, val_reg, size, is_signed);
 }
-// explicit Gp size cast
 inline x86::Gp gen_ext(jit_holder& jh, x86::Gp val, unsigned size, bool is_signed) {
     auto& cc = jh.cc;
     if(is_signed) {
@@ -398,7 +413,6 @@ inline x86::Gp gen_ext(jit_holder& jh, x86::Gp val, unsigned size, bool is_signe
         throw std::runtime_error("Invalid size in gen_ext");
     }
 }
-
 inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, uint32_t length) {
     x86::Compiler& cc = jh.cc;
     auto ret_reg = cc.newInt32();
@@ -447,31 +461,18 @@ inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, uint3
     invokeNode->setArg(2, mem_type_reg);
     invokeNode->setArg(3, addr);
     invokeNode->setArg(4, val_ptr);
+    cc.cmp(ret_reg, 0);
+    cc.jne(jh.trap_entry);
 
     cc.mov(val_reg, x86::ptr_64(val_ptr));
     cc.and_(val_reg, mask);
-    cc.cmp(ret_reg, 0);
-    cc.jne(jh.trap_entry);
     return val_reg;
 }
 inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp length) {
-    uint32_t length_val = 0;
+    throw std::runtime_error("Invalid gen_read_mem");
-    auto length_ptr = jh.cc.newIntPtr();
-    jh.cc.mov(length_ptr, &length_val);
-    jh.cc.mov(x86::ptr_32(length_ptr), length);
-
-    return gen_read_mem(jh, type, addr, length);
 }
 inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, uint64_t addr, x86::Gp length) {
-    auto addr_reg = jh.cc.newInt64();
+    throw std::runtime_error("Invalid gen_read_mem");
-    jh.cc.mov(addr_reg, addr);
-
-    uint32_t length_val = 0;
-    auto length_ptr = jh.cc.newIntPtr();
-    jh.cc.mov(length_ptr, &length_val);
-    jh.cc.mov(x86::ptr_32(length_ptr), length);
-
-    return gen_read_mem(jh, type, addr_reg, length_val);
 }
 inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, uint64_t addr, uint32_t length) {
     auto addr_reg = jh.cc.newInt64();
@@ -479,32 +480,38 @@ inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, uint64_t addr, uint
 
     return gen_read_mem(jh, type, addr_reg, length);
 }
-inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, int64_t val) {
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, int64_t val, uint32_t length) {
-    auto val_reg = jh.cc.newInt64();
+    auto val_reg = get_reg_for(jh, length * 8, true);
     jh.cc.mov(val_reg, val);
-    gen_write_mem(jh, type, addr, val_reg);
+    gen_write_mem(jh, type, addr, val_reg, length);
 }
-inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp val) {
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp val, uint32_t length) {
     x86::Compiler& cc = jh.cc;
-
+    assert(val.size() == length);
     auto mem_type_reg = cc.newInt32();
     jh.cc.mov(mem_type_reg, type);
     auto space_reg = cc.newInt32();
     jh.cc.mov(space_reg, static_cast<uint16_t>(iss::address_type::VIRTUAL));
     auto ret_reg = cc.newInt32();
     InvokeNode* invokeNode;
-
+    switch(length) {
-    if(val.isGpb()) {
+    case 1:
         cc.invoke(&invokeNode, &write_mem1, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint8_t>());
-    } else if(val.isGpw()) {
-        cc.invoke(&invokeNode, &write_mem2, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint16_t>());
-    } else if(val.isGpd()) {
-        cc.invoke(&invokeNode, &write_mem4, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint32_t>());
-    } else if(val.isGpq()) {
-        cc.invoke(&invokeNode, &write_mem8, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint64_t>());
-    } else
-        throw std::runtime_error("Invalid register size in gen_write_mem");
 
+        break;
+    case 2:
+        cc.invoke(&invokeNode, &write_mem2, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint16_t>());
+        break;
+    case 4:
+        cc.invoke(&invokeNode, &write_mem4, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint32_t>());
+        break;
+    case 8:
+        cc.invoke(&invokeNode, &write_mem8, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint64_t>());
+
+        break;
+    default:
+        throw std::runtime_error("Invalid register size in gen_ext");
+    }
     invokeNode->setRet(0, ret_reg);
     invokeNode->setArg(0, jh.arch_if_ptr);
     invokeNode->setArg(1, space_reg);
@@ -515,16 +522,16 @@ inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp
     cc.cmp(ret_reg, 0);
     cc.jne(jh.trap_entry);
 }
-inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, x86::Gp val) {
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, x86::Gp val, uint32_t length) {
-    auto addr_reg = jh.cc.newInt64();
+    auto addr_reg = jh.cc.newUInt64();
     jh.cc.mov(addr_reg, addr);
-    gen_write_mem(jh, type, addr_reg, val);
+    gen_write_mem(jh, type, addr_reg, val, length);
 }
-inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, int64_t val) {
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, int64_t val, uint32_t length) {
-    auto val_reg = jh.cc.newInt64();
+    auto val_reg = get_reg_for(jh, length * 8, true);
     jh.cc.mov(val_reg, val);
 
-    auto addr_reg = jh.cc.newInt64();
+    auto addr_reg = jh.cc.newUInt64();
     jh.cc.mov(addr_reg, addr);
-    gen_write_mem(jh, type, addr_reg, val_reg);
+    gen_write_mem(jh, type, addr_reg, val_reg, length);
 }