/* src/vm/jit/x86_64/asmpart.S - Java-C interface functions for x86_64
- Copyright (C) 1996-2005, 2006 R. Grafl, A. Krall, C. Kruegel,
+ Copyright (C) 1996-2005, 2006, 2007 R. Grafl, A. Krall, C. Kruegel,
C. Oates, R. Obermaisser, M. Platter, M. Probst, S. Ring,
E. Steiner, C. Thalinger, D. Thuernbeck, P. Tomsich, C. Ullrich,
J. Wenninger, Institut f. Computersprachen - TU Wien
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
02110-1301, USA.
- Contact: cacao@cacaojvm.org
-
- Authors: Andreas Krall
- Reinhard Grafl
- Christian Thalinger
-
- Changes: Edwin Steiner
-
- $Id: asmpart.S 5142 2006-07-17 09:47:02Z twisti $
-
*/
#include "vm/jit/x86_64/arch.h"
#include "vm/jit/x86_64/md-abi.h"
#include "vm/jit/x86_64/md-asm.h"
-#include "vm/jit/x86_64/offsets.h"
#include "vm/jit/abi-asm.h"
#include "vm/jit/methodheader.h"
.globl asm_vm_call_method_float
.globl asm_vm_call_method_double
.globl asm_vm_call_method_exception_handler
-
- .globl asm_call_jit_compiler
+ .globl asm_vm_call_method_end
.globl asm_handle_exception
.globl asm_handle_nat_exception
.globl asm_abstractmethoderror
- .globl asm_wrapper_patcher
-
- .globl asm_replacement_out
- .globl asm_replacement_in
-
.globl asm_builtin_f2i
.globl asm_builtin_f2l
.globl asm_builtin_d2i
.globl asm_builtin_d2l
- .globl asm_criticalsections
- .globl asm_getclassvalues_atomic
+ .globl asm_compare_and_swap
+ .globl asm_memory_barrier
/********************* function asm_calljavafunction ***************************
.align 8
- .quad 0 /* catch type all */
- .quad 0 /* handler pc */
- .quad 0 /* end pc */
- .quad 0 /* start pc */
- .long 1 /* extable size */
- .long 0 /* ALIGNMENT PADDING */
.quad 0 /* line number table start */
.quad 0 /* line number table size */
- .long 0 /* ALIGNMENT PADDING */
.long 0 /* fltsave */
.long 0 /* intsave */
.long 0 /* isleaf */
- .long 0 /* IsSync */
.long 0 /* frame size */
.quad 0 /* codeinfo pointer */
mov s3,4*8(sp)
mov s4,5*8(sp)
- mov a0,itmp1 /* move method pointer for compiler */
- xor %rbp,%rbp /* set argument stack frame to zero */
+ mov a0,6*8(sp) /* store method PV */
- test a1,a1 /* maybe we have no args... */
- jle L_copy_done
+ mov sp,s0 /* save stack pointer */
- mov a1,itmp3 /* arg count */
- mov a2,itmp2 /* pointer to arg block */
+ mov a1,t0 /* address of data structure */
+ mov a2,itmp1 /* number of stack arguments */
- mov itmp2,%r14 /* save argument block pointer */
- mov itmp3,%r15 /* save argument count */
+ mov 0*8(t0),a0
+ mov 1*8(t0),a1
+ mov 2*8(t0),a2
+ mov 3*8(t0),a3
+ mov 4*8(t0),a4
+ mov 5*8(t0),a5
- sub $sizevmarg,itmp2 /* initialize pointer (smaller code) */
- add $1,itmp3 /* initialize argument count */
- xor %r12,%r12 /* initialize integer argument counter*/
- xor %r13,%r13 /* initialize float argument counter */
+ movq 6*8(t0),fa0
+ movq 7*8(t0),fa1
+ movq 8*8(t0),fa2
+ movq 9*8(t0),fa3
+ movq 10*8(t0),fa4
+ movq 11*8(t0),fa5
+ movq 12*8(t0),fa6
+ movq 13*8(t0),fa7
-L_register_copy:
- add $sizevmarg,itmp2 /* goto next argument block */
- dec itmp3 /* argument count - 1 */
- jz L_register_copy_done
- andb $0x02,offvmargtype(itmp2) /* is this a float/double type? */
- jnz L_register_handle_float /* yes, handle it */
+ cmp $0,itmp1l
+ je L_asm_vm_call_method_stack_copy_done
- cmp $INT_ARG_CNT,%r12 /* are we out of integer argument */
- je L_register_copy /* register? yes, next loop */
+ mov itmp1,itmp2
+ add $1,itmp2 /* keep stack 16-byte aligned */
+ and $0xfffffffffffffffe,itmp2
+ shl $3,itmp2 /* calculate stack size */
+ sub itmp2,sp /* create stack frame */
+ mov sp,itmp2 /* temporary stack pointer */
- lea jumptable_integer(%rip),%rbp
- mov 0(%rbp,%r12,8),%rbx
- inc %r12 /* integer argument counter + 1 */
- jmp *%rbx
+L_asm_vm_call_method_stack_copy_loop:
+ mov 14*8(t0),itmp3 /* load argument */
+ mov itmp3,0(itmp2) /* store argument on stack */
-L_register_handle_float:
- cmp $FLT_ARG_CNT,%r13 /* are we out of float argument */
- je L_register_copy /* register? yes, next loop */
+ sub $1,itmp1l /* subtract 1 argument */
+ add $8,t0 /* set address of next argument */
+ add $8,itmp2 /* increase SP */
- lea jumptable_float(%rip),%rbp
- mov 0(%rbp,%r13,8),%rbx
- inc %r13 /* float argument counter + 1 */
- jmp *%rbx
-
-L_register_copy_done:
- mov %r15,%rbp /* calculate remaining arguments */
- sub %r12,%rbp /* - integer arguments in registers */
- sub %r13,%rbp /* - float arguments in registers */
- jle L_copy_done /* are all assigned to registers? */
-
- and $0xfffffffffffffffe,%rbp /* keep stack 16-byte aligned */
- shl $3,%rbp /* calculate stack size */
- sub %rbp,sp /* stack frame for arguments */
- mov sp,%rbx /* use %rbx as temp sp */
-
- sub $sizevmarg,%r14 /* initialize pointer (smaller code) */
- add $1,%r15 /* initialize argument count */
-
-L_stack_copy_loop:
- add $sizevmarg,%r14 /* goto next argument block */
- dec %r15 /* are there any arguments left? */
- jz L_copy_done /* no test needed after dec */
-
- andb $0x02,offvmargtype(%r14) /* is this a float/double type? */
- jnz L_stack_handle_float
- dec %r12 /* arguments assigned to registers */
- jge L_stack_copy_loop
- jmp L_stack_copy
-
-L_stack_handle_float:
- dec %r13 /* arguments assigned to registers */
- jge L_stack_copy_loop
-
-L_stack_copy:
- mov offvmargdata(%r14),itmp3 /* copy s8 argument onto stack */
- mov itmp3,0(%rbx)
- add $8,%rbx /* increase sp to next argument */
- jmp L_stack_copy_loop
-
-L_copy_done:
- /* itmp1 still contains method pointer*/
- lea L_asm_call_jit_compiler(%rip),mptr
- mov sp,itmp3 /* calculate the old stack pointer */
- add bp,itmp3
- mov mptr,6*8(itmp3)
- lea (6*8-256)(itmp3),mptr /* We subtract 256 to force the next */
+ cmp $0,itmp1l
+ jg L_asm_vm_call_method_stack_copy_loop
+
+L_asm_vm_call_method_stack_copy_done:
+ lea (6*8-256)(s0),mptr /* We subtract 256 to force the next */
/* move instruction to have a 32-bit */
/* offset. */
- mov (0*8+256)(mptr),itmp3 /* method call as in Java */
- call *itmp3 /* call JIT compiler */
+ mov (0*8+256)(mptr),itmp3 /* load PV */
+ call *itmp3
- add bp,sp /* remove argument stack frame if any */
+ mov s0,sp /* restore SP */
L_asm_vm_call_method_return:
mov 0*8(sp),%rbx /* restore callee saved registers */
call builtin_throw_exception@PLT
jmp L_asm_vm_call_method_return
-
-jumptable_integer:
- .quad handle_a0
- .quad handle_a1
- .quad handle_a2
- .quad handle_a3
- .quad handle_a4
- .quad handle_a5
-
-handle_a0:
- mov offvmargdata(itmp2),a0
- jmp L_register_copy
-handle_a1:
- mov offvmargdata(itmp2),a1
- jmp L_register_copy
-handle_a2:
- mov offvmargdata(itmp2),a2
- jmp L_register_copy
-handle_a3:
- mov offvmargdata(itmp2),a3
- jmp L_register_copy
-handle_a4:
- mov offvmargdata(itmp2),a4
- jmp L_register_copy
-handle_a5:
- mov offvmargdata(itmp2),a5
- jmp L_register_copy
-
-
-jumptable_float:
- .quad handle_fa0
- .quad handle_fa1
- .quad handle_fa2
- .quad handle_fa3
- .quad handle_fa4
- .quad handle_fa5
- .quad handle_fa6
- .quad handle_fa7
-
-handle_fa0:
- movq offvmargdata(itmp2),fa0
- jmp L_register_copy
-handle_fa1:
- movq offvmargdata(itmp2),fa1
- jmp L_register_copy
-handle_fa2:
- movq offvmargdata(itmp2),fa2
- jmp L_register_copy
-handle_fa3:
- movq offvmargdata(itmp2),fa3
- jmp L_register_copy
-handle_fa4:
- movq offvmargdata(itmp2),fa4
- jmp L_register_copy
-handle_fa5:
- movq offvmargdata(itmp2),fa5
- jmp L_register_copy
-handle_fa6:
- movq offvmargdata(itmp2),fa6
- jmp L_register_copy
-handle_fa7:
- movq offvmargdata(itmp2),fa7
- jmp L_register_copy
-
-
-/****************** function asm_call_jit_compiler *****************************
-* *
-* invokes the compiler for untranslated JavaVM methods. *
-* *
-* Register R0 contains a pointer to the method info structure (prepared *
-* by createcompilerstub). Using the return address in R26 and the *
-* offset in the LDA instruction or using the value in methodptr R28 the *
-* patching address for storing the method address can be computed: *
-* *
-* method address was either loaded using *
-* *
-* i386_mov_imm_reg(a, REG_ITMP2) ; invokestatic/special *
-* i386_call_reg(REG_ITMP2) *
-* *
-* or *
-* *
-* i386_mov_membase_reg(REG_SP, 0, REG_ITMP2) ; invokevirtual/interface *
-* i386_mov_membase_reg(REG_ITMP2, OFFSET(, vftbl), REG_ITMP3) *
-* i386_mov_membase_reg(REG_ITMP3, OFFSET(vftbl, table[0]) + \ *
-* sizeof(methodptr) * m->vftblindex, REG_ITMP1) *
-* i386_call_reg(REG_ITMP1) *
-* *
-* in the static case the method pointer can be computed using the *
-* return address and the lda function following the jmp instruction *
-* *
-*******************************************************************************/
-
-asm_call_jit_compiler:
-L_asm_call_jit_compiler: /* required for PIC code */
- sub $(ARG_CNT+1)*8,sp /* +1: keep stack 16-byte aligned */
-
- SAVE_ARGUMENT_REGISTERS(0)
-
- mov itmp1,a0 /* pass methodinfo pointer */
- mov mptr,a1 /* pass method pointer */
- mov sp,a2 /* pass java sp */
- add $(1+ARG_CNT+1)*8,a2
- mov (ARG_CNT+1)*8(sp),a3 /* pass ra to java function */
- call jit_asm_compile@PLT
-
- RESTORE_ARGUMENT_REGISTERS(0)
-
- add $(ARG_CNT+1)*8,sp /* remove stack frame */
-
- test v0,v0 /* check for exception */
- je L_asm_call_jit_compiler_exception
-
- jmp *v0 /* ...and now call the new method */
-
-L_asm_call_jit_compiler_exception:
- call exceptions_get_and_clear_exception@PLT
- pop xpc /* delete return address */
- sub $3,xpc /* faulting address is ra - 3 */
- jmp L_asm_handle_exception
+asm_vm_call_method_end:
+ nop
/* asm_handle_exception ********************************************************
mov t0,4*8(sp) /* save maybe-leaf flag */
mov xpc,a0 /* exception pc */
- call codegen_findmethod@PLT
+ call codegen_get_pv_from_pc@PLT
mov v0,2*8(sp) /* save data segment pointer */
mov 0*8(sp),a0 /* pass exception pointer */
jmp L_asm_handle_exception
-/* asm_wrapper_patcher *********************************************************
-
- XXX
-
- Stack layout:
- 40 return address
- 32 pointer to virtual java_objectheader
- 24 machine code (which is patched back later)
- 16 unresolved class/method/field reference
- 8 data segment displacement from load instructions
- 0 pointer to patcher function
- -8 bp
-
-*******************************************************************************/
-
-asm_wrapper_patcher:
- push bp /* save base pointer */
- mov sp,bp /* move actual sp to bp */
- sub $((3+ARG_CNT+TMP_CNT)*8+sizestackframeinfo),sp
- and $0xfffffffffffffff0,sp /* align sp to 16-byte (this is for */
- /* leaf functions) */
-
- SAVE_ARGUMENT_REGISTERS(3)
- SAVE_TEMPORARY_REGISTERS(3+ARG_CNT)
-
- mov itmp1,0*8(sp) /* save itmp1 and itmp2 */
- mov itmp2,1*8(sp) /* can be used by some instructions */
-
- mov bp,a0 /* pass SP of patcher stub */
- add $(1*8),a0
- mov $0,a1 /* pass PV (if NULL, use findmethod) */
- mov $0,a2 /* pass RA (it's on the stack) */
- call patcher_wrapper@PLT
- mov v0,2*8(sp) /* save return value */
-
- RESTORE_ARGUMENT_REGISTERS(3)
- RESTORE_TEMPORARY_REGISTERS(3+ARG_CNT)
-
- mov 0*8(sp),itmp1 /* restore itmp1 and itmp2 */
- mov 1*8(sp),itmp2 /* can be used by some instructions */
- mov 2*8(sp),itmp3 /* restore return value */
-
- mov bp,sp /* restore original sp */
- pop bp /* restore bp */
- add $(5*8),sp /* remove patcher stackframe, keep RA */
-
- test itmp3,itmp3 /* exception thrown? */
- jne L_asm_wrapper_patcher_exception
- ret /* call new patched code */
-
-L_asm_wrapper_patcher_exception:
- mov itmp3,xptr /* get exception */
- pop xpc /* get and remove return address */
- jmp L_asm_handle_exception
-
-
-/* asm_replacement_out *********************************************************
-
- This code is jumped to from the replacement-out stubs that are executed
- when a thread reaches an activated replacement point.
-
- The purpose of asm_replacement_out is to read out the parts of the
- execution state that cannot be accessed from C code, store this state,
- and then call the C function replace_me.
-
- Stack layout:
- 8 start of stack inside method to replace
- 0 rplpoint * info on the replacement point that was reached
-
-*******************************************************************************/
-
-/* some room to accomodate changes of the stack frame size during replacement */
- /* XXX we should find a cleaner solution here */
-#define REPLACEMENT_ROOM 512
-
-asm_replacement_out:
- /* create stack frame */
- sub $(sizeexecutionstate + REPLACEMENT_ROOM),sp
-
- /* save registers in execution state */
- mov %rax,(RAX*8+offes_intregs)(sp)
- mov %rbx,(RBX*8+offes_intregs)(sp)
- mov %rcx,(RCX*8+offes_intregs)(sp)
- mov %rdx,(RDX*8+offes_intregs)(sp)
- mov %rsi,(RSI*8+offes_intregs)(sp)
- mov %rdi,(RDI*8+offes_intregs)(sp)
- mov %rbp,(RBP*8+offes_intregs)(sp)
- movq $0 ,(RSP*8+offes_intregs)(sp) /* not used */
- mov %r8 ,(R8 *8+offes_intregs)(sp)
- mov %r9 ,(R9 *8+offes_intregs)(sp)
- mov %r10,(R10*8+offes_intregs)(sp)
- mov %r11,(R11*8+offes_intregs)(sp)
- mov %r12,(R12*8+offes_intregs)(sp)
- mov %r13,(R13*8+offes_intregs)(sp)
- mov %r14,(R14*8+offes_intregs)(sp)
- mov %r15,(R15*8+offes_intregs)(sp)
-
- movq %xmm0 ,(XMM0 *8+offes_fltregs)(sp)
- movq %xmm1 ,(XMM1 *8+offes_fltregs)(sp)
- movq %xmm2 ,(XMM2 *8+offes_fltregs)(sp)
- movq %xmm3 ,(XMM3 *8+offes_fltregs)(sp)
- movq %xmm4 ,(XMM4 *8+offes_fltregs)(sp)
- movq %xmm5 ,(XMM5 *8+offes_fltregs)(sp)
- movq %xmm6 ,(XMM6 *8+offes_fltregs)(sp)
- movq %xmm7 ,(XMM7 *8+offes_fltregs)(sp)
- movq %xmm8 ,(XMM8 *8+offes_fltregs)(sp)
- movq %xmm9 ,(XMM9 *8+offes_fltregs)(sp)
- movq %xmm10,(XMM10*8+offes_fltregs)(sp)
- movq %xmm11,(XMM11*8+offes_fltregs)(sp)
- movq %xmm12,(XMM12*8+offes_fltregs)(sp)
- movq %xmm13,(XMM13*8+offes_fltregs)(sp)
- movq %xmm14,(XMM14*8+offes_fltregs)(sp)
- movq %xmm15,(XMM15*8+offes_fltregs)(sp)
-
- /* calculate sp of method */
- mov sp,itmp1
- add $(sizeexecutionstate + REPLACEMENT_ROOM + 8),itmp1
- mov itmp1,(offes_sp)(sp)
-
- /* pv must be looked up via AVL tree */
- movq $0,(offes_pv)(sp)
-
- /* call replace_me */
- mov -8(itmp1),a0 /* rplpoint * */
- mov sp,a1 /* arg1: execution state */
- call replace_me@PLT /* call C function replace_me */
- call abort@PLT /* NEVER REACHED */
-
-/* asm_replacement_in **********************************************************
-
- This code writes the given execution state and jumps to the replacement
- code.
-
- This function never returns!
-
- C prototype:
- void asm_replacement_in(executionstate *es);
-
-*******************************************************************************/
-
-asm_replacement_in:
- mov a0,%rbp /* executionstate *es */
-
- /* set new sp */
- mov (offes_sp)(%rbp),%rsp
-
- /* store address of new code */
- push (offes_pc)(%rbp)
-
- /* copy registers from execution state */
- movq (XMM0 *8+offes_fltregs)(%rbp),%xmm0
- movq (XMM1 *8+offes_fltregs)(%rbp),%xmm1
- movq (XMM2 *8+offes_fltregs)(%rbp),%xmm2
- movq (XMM3 *8+offes_fltregs)(%rbp),%xmm3
- movq (XMM4 *8+offes_fltregs)(%rbp),%xmm4
- movq (XMM5 *8+offes_fltregs)(%rbp),%xmm5
- movq (XMM6 *8+offes_fltregs)(%rbp),%xmm6
- movq (XMM7 *8+offes_fltregs)(%rbp),%xmm7
- movq (XMM8 *8+offes_fltregs)(%rbp),%xmm8
- movq (XMM9 *8+offes_fltregs)(%rbp),%xmm9
- movq (XMM10*8+offes_fltregs)(%rbp),%xmm10
- movq (XMM11*8+offes_fltregs)(%rbp),%xmm11
- movq (XMM12*8+offes_fltregs)(%rbp),%xmm12
- movq (XMM13*8+offes_fltregs)(%rbp),%xmm13
- movq (XMM14*8+offes_fltregs)(%rbp),%xmm14
- movq (XMM15*8+offes_fltregs)(%rbp),%xmm15
-
- mov (RAX*8+offes_intregs)(%rbp),%rax
- mov (RBX*8+offes_intregs)(%rbp),%rbx
- mov (RCX*8+offes_intregs)(%rbp),%rcx
- mov (RDX*8+offes_intregs)(%rbp),%rdx
- mov (RSI*8+offes_intregs)(%rbp),%rsi
- mov (RDI*8+offes_intregs)(%rbp),%rdi
- mov (R8 *8+offes_intregs)(%rbp),%r8
- mov (R9 *8+offes_intregs)(%rbp),%r9
- mov (R10*8+offes_intregs)(%rbp),%r10
- mov (R11*8+offes_intregs)(%rbp),%r11
- mov (R12*8+offes_intregs)(%rbp),%r12
- mov (R13*8+offes_intregs)(%rbp),%r13
- mov (R14*8+offes_intregs)(%rbp),%r14
- mov (R15*8+offes_intregs)(%rbp),%r15
-
- mov (RBP*8+offes_intregs)(%rbp),%rbp
-
- /* jump to new code */
- ret
-
-
/* asm_builtin_x2x *************************************************************
* *
* Wrapper functions for float to int corner cases *
ret
-asm_getclassvalues_atomic:
-_crit_restart:
-_crit_begin:
- movl offbaseval(a0),itmp1l
- movl offdiffval(a0),itmp2l
- movl offbaseval(a1),itmp3l
-_crit_end:
- movl itmp1l,offcast_super_baseval(a2)
- movl itmp2l,offcast_super_diffval(a2)
- movl itmp3l,offcast_sub_baseval(a2)
+/* asm_compare_and_swap ********************************************************
+
+ Does an atomic compare and swap. Required for the lock
+ implementation.
+
+*******************************************************************************/
+
+asm_compare_and_swap:
+ mov a1,v0 /* v0 is %rax */
+ lock cmpxchg a2,(a0)
ret
- .data
-
-asm_criticalsections:
-#if defined(ENABLE_THREADS)
- .quad _crit_begin
- .quad _crit_end
- .quad _crit_restart
-#endif
- .quad 0
+
+/* asm_memory_barrier **********************************************************
+
+ A memory barrier for the Java Memory Model.
+
+*******************************************************************************/
+
+asm_memory_barrier:
+ mfence
+ ret
-/* Disable exec-stacks, required for Gentoo ***********************************/
+/* disable exec-stacks ********************************************************/
-#if defined(__GCC__) && defined(__ELF__)
- .section .note.GNU-stack,"",@progbits
+#if defined(__linux__) && defined(__ELF__)
+ .section .note.GNU-stack,"",%progbits
#endif