Mercurial > hg > truffle
view src/cpu/x86/vm/methodHandles_x86.cpp @ 12169:29aa8936f03c
8023597: Optimize G1 barriers code for unsafe load_store
Summary: Avoid loading old values in G1 pre-barriers for inlined unsafe load_store nodes.
Reviewed-by: kvn, tonyp
Contributed-by: Martin Doerr <martin.doerr@sap.com>
author | kvn |
---|---|
date | Wed, 28 Aug 2013 11:22:43 +0200 |
parents | cc32ccaaf47f |
children | 190899198332 dc261f466b6d |
line wrap: on
line source
/* * Copyright (c) 1997, 2012, 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 "interpreter/interpreter.hpp" #include "interpreter/interpreterRuntime.hpp" #include "memory/allocation.inline.hpp" #include "prims/methodHandles.hpp" #define __ _masm-> #ifdef PRODUCT #define BLOCK_COMMENT(str) /* nothing */ #define STOP(error) stop(error) #else #define BLOCK_COMMENT(str) __ block_comment(str) #define STOP(error) block_comment(error); __ stop(error) #endif #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") void MethodHandles::load_klass_from_Class(MacroAssembler* _masm, Register klass_reg) { if (VerifyMethodHandles) verify_klass(_masm, klass_reg, SystemDictionary::WK_KLASS_ENUM_NAME(java_lang_Class), "MH argument is a Class"); __ movptr(klass_reg, Address(klass_reg, java_lang_Class::klass_offset_in_bytes())); } #ifdef ASSERT static int check_nonzero(const char* xname, int x) { assert(x != 0, err_msg("%s should be nonzero", xname)); return x; } #define NONZERO(x) check_nonzero(#x, x) #else //ASSERT #define NONZERO(x) (x) #endif //ASSERT #ifdef ASSERT void MethodHandles::verify_klass(MacroAssembler* _masm, Register obj, SystemDictionary::WKID klass_id, const char* error_message) { Klass** klass_addr = SystemDictionary::well_known_klass_addr(klass_id); KlassHandle klass = SystemDictionary::well_known_klass(klass_id); Register temp = rdi; Register temp2 = noreg; LP64_ONLY(temp2 = rscratch1); // used by MacroAssembler::cmpptr Label L_ok, L_bad; BLOCK_COMMENT("verify_klass {"); __ verify_oop(obj); __ testptr(obj, obj); __ jcc(Assembler::zero, L_bad); __ push(temp); if (temp2 != noreg) __ push(temp2); #define UNPUSH { if (temp2 != noreg) __ pop(temp2); __ pop(temp); } __ load_klass(temp, obj); __ cmpptr(temp, ExternalAddress((address) klass_addr)); __ jcc(Assembler::equal, L_ok); intptr_t super_check_offset = klass->super_check_offset(); __ movptr(temp, Address(temp, super_check_offset)); __ cmpptr(temp, ExternalAddress((address) klass_addr)); __ jcc(Assembler::equal, L_ok); UNPUSH; __ bind(L_bad); __ STOP(error_message); __ BIND(L_ok); UNPUSH; BLOCK_COMMENT("} verify_klass"); } void MethodHandles::verify_ref_kind(MacroAssembler* _masm, int ref_kind, Register member_reg, Register temp) { Label L; BLOCK_COMMENT("verify_ref_kind {"); __ movl(temp, Address(member_reg, NONZERO(java_lang_invoke_MemberName::flags_offset_in_bytes()))); __ shrl(temp, java_lang_invoke_MemberName::MN_REFERENCE_KIND_SHIFT); __ andl(temp, java_lang_invoke_MemberName::MN_REFERENCE_KIND_MASK); __ cmpl(temp, ref_kind); __ jcc(Assembler::equal, L); { char* buf = NEW_C_HEAP_ARRAY(char, 100, mtInternal); jio_snprintf(buf, 100, "verify_ref_kind expected %x", ref_kind); if (ref_kind == JVM_REF_invokeVirtual || ref_kind == JVM_REF_invokeSpecial) // could do this for all ref_kinds, but would explode assembly code size trace_method_handle(_masm, buf); __ STOP(buf); } BLOCK_COMMENT("} verify_ref_kind"); __ bind(L); } #endif //ASSERT void MethodHandles::jump_from_method_handle(MacroAssembler* _masm, Register method, Register temp, bool for_compiler_entry) { assert(method == rbx, "interpreter calling convention"); __ verify_method_ptr(method); if (!for_compiler_entry && JvmtiExport::can_post_interpreter_events()) { Label run_compiled_code; // JVMTI events, such as single-stepping, are implemented partly by avoiding running // compiled code in threads for which the event is enabled. Check here for // interp_only_mode if these events CAN be enabled. #ifdef _LP64 Register rthread = r15_thread; #else Register rthread = temp; __ get_thread(rthread); #endif // interp_only is an int, on little endian it is sufficient to test the byte only // Is a cmpl faster? __ cmpb(Address(rthread, JavaThread::interp_only_mode_offset()), 0); __ jccb(Assembler::zero, run_compiled_code); __ jmp(Address(method, Method::interpreter_entry_offset())); __ BIND(run_compiled_code); } const ByteSize entry_offset = for_compiler_entry ? Method::from_compiled_offset() : Method::from_interpreted_offset(); __ jmp(Address(method, entry_offset)); } void MethodHandles::jump_to_lambda_form(MacroAssembler* _masm, Register recv, Register method_temp, Register temp2, bool for_compiler_entry) { BLOCK_COMMENT("jump_to_lambda_form {"); // This is the initial entry point of a lazy method handle. // After type checking, it picks up the invoker from the LambdaForm. assert_different_registers(recv, method_temp, temp2); assert(recv != noreg, "required register"); assert(method_temp == rbx, "required register for loading method"); //NOT_PRODUCT({ FlagSetting fs(TraceMethodHandles, true); trace_method_handle(_masm, "LZMH"); }); // Load the invoker, as MH -> MH.form -> LF.vmentry __ verify_oop(recv); __ load_heap_oop(method_temp, Address(recv, NONZERO(java_lang_invoke_MethodHandle::form_offset_in_bytes()))); __ verify_oop(method_temp); __ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_LambdaForm::vmentry_offset_in_bytes()))); __ verify_oop(method_temp); // the following assumes that a Method* is normally compressed in the vmtarget field: __ movptr(method_temp, Address(method_temp, NONZERO(java_lang_invoke_MemberName::vmtarget_offset_in_bytes()))); if (VerifyMethodHandles && !for_compiler_entry) { // make sure recv is already on stack __ movptr(temp2, Address(method_temp, Method::const_offset())); __ load_sized_value(temp2, Address(temp2, ConstMethod::size_of_parameters_offset()), sizeof(u2), /*is_signed*/ false); // assert(sizeof(u2) == sizeof(Method::_size_of_parameters), ""); Label L; __ cmpptr(recv, __ argument_address(temp2, -1)); __ jcc(Assembler::equal, L); __ movptr(rax, __ argument_address(temp2, -1)); __ STOP("receiver not on stack"); __ BIND(L); } jump_from_method_handle(_masm, method_temp, temp2, for_compiler_entry); BLOCK_COMMENT("} jump_to_lambda_form"); } // Code generation address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm, vmIntrinsics::ID iid) { const bool not_for_compiler_entry = false; // this is the interpreter entry assert(is_signature_polymorphic(iid), "expected invoke iid"); if (iid == vmIntrinsics::_invokeGeneric || iid == vmIntrinsics::_compiledLambdaForm) { // Perhaps surprisingly, the symbolic references visible to Java are not directly used. // They are linked to Java-generated adapters via MethodHandleNatives.linkMethod. // They all allow an appendix argument. __ hlt(); // empty stubs make SG sick return NULL; } // rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted) // rbx: Method* // rdx: argument locator (parameter slot count, added to rsp) // rcx: used as temp to hold mh or receiver // rax, rdi: garbage temps, blown away Register rdx_argp = rdx; // argument list ptr, live on error paths Register rax_temp = rax; Register rcx_mh = rcx; // MH receiver; dies quickly and is recycled Register rbx_method = rbx; // eventual target of this invocation // here's where control starts out: __ align(CodeEntryAlignment); address entry_point = __ pc(); if (VerifyMethodHandles) { Label L; BLOCK_COMMENT("verify_intrinsic_id {"); __ cmpb(Address(rbx_method, Method::intrinsic_id_offset_in_bytes()), (int) iid); __ jcc(Assembler::equal, L); if (iid == vmIntrinsics::_linkToVirtual || iid == vmIntrinsics::_linkToSpecial) { // could do this for all kinds, but would explode assembly code size trace_method_handle(_masm, "bad Method*::intrinsic_id"); } __ STOP("bad Method*::intrinsic_id"); __ bind(L); BLOCK_COMMENT("} verify_intrinsic_id"); } // First task: Find out how big the argument list is. Address rdx_first_arg_addr; int ref_kind = signature_polymorphic_intrinsic_ref_kind(iid); assert(ref_kind != 0 || iid == vmIntrinsics::_invokeBasic, "must be _invokeBasic or a linkTo intrinsic"); if (ref_kind == 0 || MethodHandles::ref_kind_has_receiver(ref_kind)) { __ movptr(rdx_argp, Address(rbx_method, Method::const_offset())); __ load_sized_value(rdx_argp, Address(rdx_argp, ConstMethod::size_of_parameters_offset()), sizeof(u2), /*is_signed*/ false); // assert(sizeof(u2) == sizeof(Method::_size_of_parameters), ""); rdx_first_arg_addr = __ argument_address(rdx_argp, -1); } else { DEBUG_ONLY(rdx_argp = noreg); } if (!is_signature_polymorphic_static(iid)) { __ movptr(rcx_mh, rdx_first_arg_addr); DEBUG_ONLY(rdx_argp = noreg); } // rdx_first_arg_addr is live! trace_method_handle_interpreter_entry(_masm, iid); if (iid == vmIntrinsics::_invokeBasic) { generate_method_handle_dispatch(_masm, iid, rcx_mh, noreg, not_for_compiler_entry); } else { // Adjust argument list by popping the trailing MemberName argument. Register rcx_recv = noreg; if (MethodHandles::ref_kind_has_receiver(ref_kind)) { // Load the receiver (not the MH; the actual MemberName's receiver) up from the interpreter stack. __ movptr(rcx_recv = rcx, rdx_first_arg_addr); } DEBUG_ONLY(rdx_argp = noreg); Register rbx_member = rbx_method; // MemberName ptr; incoming method ptr is dead now __ pop(rax_temp); // return address __ pop(rbx_member); // extract last argument __ push(rax_temp); // re-push return address generate_method_handle_dispatch(_masm, iid, rcx_recv, rbx_member, not_for_compiler_entry); } return entry_point; } void MethodHandles::generate_method_handle_dispatch(MacroAssembler* _masm, vmIntrinsics::ID iid, Register receiver_reg, Register member_reg, bool for_compiler_entry) { assert(is_signature_polymorphic(iid), "expected invoke iid"); Register rbx_method = rbx; // eventual target of this invocation // temps used in this code are not used in *either* compiled or interpreted calling sequences #ifdef _LP64 Register temp1 = rscratch1; Register temp2 = rscratch2; Register temp3 = rax; if (for_compiler_entry) { assert(receiver_reg == (iid == vmIntrinsics::_linkToStatic ? noreg : j_rarg0), "only valid assignment"); assert_different_registers(temp1, j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5); assert_different_registers(temp2, j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5); assert_different_registers(temp3, j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5); } #else Register temp1 = (for_compiler_entry ? rsi : rdx); Register temp2 = rdi; Register temp3 = rax; if (for_compiler_entry) { assert(receiver_reg == (iid == vmIntrinsics::_linkToStatic ? noreg : rcx), "only valid assignment"); assert_different_registers(temp1, rcx, rdx); assert_different_registers(temp2, rcx, rdx); assert_different_registers(temp3, rcx, rdx); } #endif else { assert_different_registers(temp1, temp2, temp3, saved_last_sp_register()); // don't trash lastSP } assert_different_registers(temp1, temp2, temp3, receiver_reg); assert_different_registers(temp1, temp2, temp3, member_reg); if (iid == vmIntrinsics::_invokeBasic) { // indirect through MH.form.vmentry.vmtarget jump_to_lambda_form(_masm, receiver_reg, rbx_method, temp1, for_compiler_entry); } else { // The method is a member invoker used by direct method handles. if (VerifyMethodHandles) { // make sure the trailing argument really is a MemberName (caller responsibility) verify_klass(_masm, member_reg, SystemDictionary::WK_KLASS_ENUM_NAME(java_lang_invoke_MemberName), "MemberName required for invokeVirtual etc."); } Address member_clazz( member_reg, NONZERO(java_lang_invoke_MemberName::clazz_offset_in_bytes())); Address member_vmindex( member_reg, NONZERO(java_lang_invoke_MemberName::vmindex_offset_in_bytes())); Address member_vmtarget( member_reg, NONZERO(java_lang_invoke_MemberName::vmtarget_offset_in_bytes())); Register temp1_recv_klass = temp1; if (iid != vmIntrinsics::_linkToStatic) { __ verify_oop(receiver_reg); if (iid == vmIntrinsics::_linkToSpecial) { // Don't actually load the klass; just null-check the receiver. __ null_check(receiver_reg); } else { // load receiver klass itself __ null_check(receiver_reg, oopDesc::klass_offset_in_bytes()); __ load_klass(temp1_recv_klass, receiver_reg); __ verify_klass_ptr(temp1_recv_klass); } BLOCK_COMMENT("check_receiver {"); // The receiver for the MemberName must be in receiver_reg. // Check the receiver against the MemberName.clazz if (VerifyMethodHandles && iid == vmIntrinsics::_linkToSpecial) { // Did not load it above... __ load_klass(temp1_recv_klass, receiver_reg); __ verify_klass_ptr(temp1_recv_klass); } if (VerifyMethodHandles && iid != vmIntrinsics::_linkToInterface) { Label L_ok; Register temp2_defc = temp2; __ load_heap_oop(temp2_defc, member_clazz); load_klass_from_Class(_masm, temp2_defc); __ verify_klass_ptr(temp2_defc); __ check_klass_subtype(temp1_recv_klass, temp2_defc, temp3, L_ok); // If we get here, the type check failed! __ STOP("receiver class disagrees with MemberName.clazz"); __ bind(L_ok); } BLOCK_COMMENT("} check_receiver"); } if (iid == vmIntrinsics::_linkToSpecial || iid == vmIntrinsics::_linkToStatic) { DEBUG_ONLY(temp1_recv_klass = noreg); // these guys didn't load the recv_klass } // Live registers at this point: // member_reg - MemberName that was the trailing argument // temp1_recv_klass - klass of stacked receiver, if needed // rsi/r13 - interpreter linkage (if interpreted) // rcx, rdx, rsi, rdi, r8, r8 - compiler arguments (if compiled) Label L_incompatible_class_change_error; switch (iid) { case vmIntrinsics::_linkToSpecial: if (VerifyMethodHandles) { verify_ref_kind(_masm, JVM_REF_invokeSpecial, member_reg, temp3); } __ movptr(rbx_method, member_vmtarget); break; case vmIntrinsics::_linkToStatic: if (VerifyMethodHandles) { verify_ref_kind(_masm, JVM_REF_invokeStatic, member_reg, temp3); } __ movptr(rbx_method, member_vmtarget); break; case vmIntrinsics::_linkToVirtual: { // same as TemplateTable::invokevirtual, // minus the CP setup and profiling: if (VerifyMethodHandles) { verify_ref_kind(_masm, JVM_REF_invokeVirtual, member_reg, temp3); } // pick out the vtable index from the MemberName, and then we can discard it: Register temp2_index = temp2; __ movptr(temp2_index, member_vmindex); if (VerifyMethodHandles) { Label L_index_ok; __ cmpl(temp2_index, 0); __ jcc(Assembler::greaterEqual, L_index_ok); __ STOP("no virtual index"); __ BIND(L_index_ok); } // Note: The verifier invariants allow us to ignore MemberName.clazz and vmtarget // at this point. And VerifyMethodHandles has already checked clazz, if needed. // get target Method* & entry point __ lookup_virtual_method(temp1_recv_klass, temp2_index, rbx_method); break; } case vmIntrinsics::_linkToInterface: { // same as TemplateTable::invokeinterface // (minus the CP setup and profiling, with different argument motion) if (VerifyMethodHandles) { verify_ref_kind(_masm, JVM_REF_invokeInterface, member_reg, temp3); } Register temp3_intf = temp3; __ load_heap_oop(temp3_intf, member_clazz); load_klass_from_Class(_masm, temp3_intf); __ verify_klass_ptr(temp3_intf); Register rbx_index = rbx_method; __ movptr(rbx_index, member_vmindex); if (VerifyMethodHandles) { Label L; __ cmpl(rbx_index, 0); __ jcc(Assembler::greaterEqual, L); __ STOP("invalid vtable index for MH.invokeInterface"); __ bind(L); } // given intf, index, and recv klass, dispatch to the implementation method __ lookup_interface_method(temp1_recv_klass, temp3_intf, // note: next two args must be the same: rbx_index, rbx_method, temp2, L_incompatible_class_change_error); break; } default: fatal(err_msg_res("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid))); break; } // Live at this point: // rbx_method // rsi/r13 (if interpreted) // After figuring out which concrete method to call, jump into it. // Note that this works in the interpreter with no data motion. // But the compiled version will require that rcx_recv be shifted out. __ verify_method_ptr(rbx_method); jump_from_method_handle(_masm, rbx_method, temp1, for_compiler_entry); if (iid == vmIntrinsics::_linkToInterface) { __ bind(L_incompatible_class_change_error); __ jump(RuntimeAddress(StubRoutines::throw_IncompatibleClassChangeError_entry())); } } } #ifndef PRODUCT void trace_method_handle_stub(const char* adaptername, oop mh, intptr_t* saved_regs, intptr_t* entry_sp) { // called as a leaf from native code: do not block the JVM! bool has_mh = (strstr(adaptername, "/static") == NULL && strstr(adaptername, "linkTo") == NULL); // static linkers don't have MH const char* mh_reg_name = has_mh ? "rcx_mh" : "rcx"; tty->print_cr("MH %s %s="PTR_FORMAT" sp="PTR_FORMAT, adaptername, mh_reg_name, mh, entry_sp); if (Verbose) { tty->print_cr("Registers:"); const int saved_regs_count = RegisterImpl::number_of_registers; for (int i = 0; i < saved_regs_count; i++) { Register r = as_Register(i); // The registers are stored in reverse order on the stack (by pusha). tty->print("%3s=" PTR_FORMAT, r->name(), saved_regs[((saved_regs_count - 1) - i)]); if ((i + 1) % 4 == 0) { tty->cr(); } else { tty->print(", "); } } tty->cr(); { // dumping last frame with frame::describe JavaThread* p = JavaThread::active(); ResourceMark rm; PRESERVE_EXCEPTION_MARK; // may not be needed by safer and unexpensive here FrameValues values; // Note: We want to allow trace_method_handle from any call site. // While trace_method_handle creates a frame, it may be entered // without a PC on the stack top (e.g. not just after a call). // Walking that frame could lead to failures due to that invalid PC. // => carefully detect that frame when doing the stack walking // Current C frame frame cur_frame = os::current_frame(); // Robust search of trace_calling_frame (independant of inlining). // Assumes saved_regs comes from a pusha in the trace_calling_frame. assert(cur_frame.sp() < saved_regs, "registers not saved on stack ?"); frame trace_calling_frame = os::get_sender_for_C_frame(&cur_frame); while (trace_calling_frame.fp() < saved_regs) { trace_calling_frame = os::get_sender_for_C_frame(&trace_calling_frame); } // safely create a frame and call frame::describe intptr_t *dump_sp = trace_calling_frame.sender_sp(); intptr_t *dump_fp = trace_calling_frame.link(); bool walkable = has_mh; // whether the traced frame shoud be walkable if (walkable) { // The previous definition of walkable may have to be refined // if new call sites cause the next frame constructor to start // failing. Alternatively, frame constructors could be // modified to support the current or future non walkable // frames (but this is more intrusive and is not considered as // part of this RFE, which will instead use a simpler output). frame dump_frame = frame(dump_sp, dump_fp); dump_frame.describe(values, 1); } else { // Stack may not be walkable (invalid PC above FP): // Add descriptions without building a Java frame to avoid issues values.describe(-1, dump_fp, "fp for #1 <not parsed, cannot trust pc>"); values.describe(-1, dump_sp, "sp for #1"); } values.describe(-1, entry_sp, "raw top of stack"); tty->print_cr("Stack layout:"); values.print(p); } if (has_mh && mh->is_oop()) { mh->print(); if (java_lang_invoke_MethodHandle::is_instance(mh)) { if (java_lang_invoke_MethodHandle::form_offset_in_bytes() != 0) java_lang_invoke_MethodHandle::form(mh)->print(); } } } } // The stub wraps the arguments in a struct on the stack to avoid // dealing with the different calling conventions for passing 6 // arguments. struct MethodHandleStubArguments { const char* adaptername; oopDesc* mh; intptr_t* saved_regs; intptr_t* entry_sp; }; void trace_method_handle_stub_wrapper(MethodHandleStubArguments* args) { trace_method_handle_stub(args->adaptername, args->mh, args->saved_regs, args->entry_sp); } void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) { if (!TraceMethodHandles) return; BLOCK_COMMENT("trace_method_handle {"); __ enter(); __ andptr(rsp, -16); // align stack if needed for FPU state __ pusha(); __ mov(rbx, rsp); // for retreiving saved_regs // Note: saved_regs must be in the entered frame for the // robust stack walking implemented in trace_method_handle_stub. // save FP result, valid at some call sites (adapter_opt_return_float, ...) __ increment(rsp, -2 * wordSize); if (UseSSE >= 2) { __ movdbl(Address(rsp, 0), xmm0); } else if (UseSSE == 1) { __ movflt(Address(rsp, 0), xmm0); } else { __ fst_d(Address(rsp, 0)); } // Incoming state: // rcx: method handle // // To avoid calling convention issues, build a record on the stack // and pass the pointer to that instead. __ push(rbp); // entry_sp (with extra align space) __ push(rbx); // pusha saved_regs __ push(rcx); // mh __ push(rcx); // slot for adaptername __ movptr(Address(rsp, 0), (intptr_t) adaptername); __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub_wrapper), rsp); __ increment(rsp, sizeof(MethodHandleStubArguments)); if (UseSSE >= 2) { __ movdbl(xmm0, Address(rsp, 0)); } else if (UseSSE == 1) { __ movflt(xmm0, Address(rsp, 0)); } else { __ fld_d(Address(rsp, 0)); } __ increment(rsp, 2 * wordSize); __ popa(); __ leave(); BLOCK_COMMENT("} trace_method_handle"); } #endif //PRODUCT