/*
* %CopyrightBegin%
*
* Copyright Ericsson AB 2004-2011. All Rights Reserved.
*
* The contents of this file are subject to the Erlang Public License,
* Version 1.1, (the "License"); you may not use this file except in
* compliance with the License. You should have received a copy of the
* Erlang Public License along with this software. If not, it can be
* retrieved online at http://www.erlang.org/.
*
* Software distributed under the License is distributed on an "AS IS"
* basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
* the License for the specific language governing rights and limitations
* under the License.
*
* %CopyrightEnd%
*/
#define ASM
#include "hipe_amd64_asm.h"
#include "hipe_literals.h"
#include "hipe_mode_switch.h"
/*
* Note: the mode-switch entry points in hipe_amd64_glue.S have
* the same names as in hipe_x86_glue.S. This is intentional,
* as it allows using hipe_x86_glue.h with AMD64.
*/
/*
* Set up frame on C stack,
* save C callee-save registers,
* retrieve the process pointer from the parameters from C,
* SWITCH_C_TO_ERLANG.
*
* The end of the frame must be 16-byte aligned, otherwise
* calls to C may break. %rsp+8 is 16-byte aligned on entry,
* and six registers are to be saved, so a seventh word is
* added to make the resulting %rsp 16-byte aligned.
*/
#define ENTER_FROM_C \
/* save C callee-save registers on the C stack */ \
subq $(7*8), %rsp; \
movq %r15, 40(%rsp); \
movq %r14, 32(%rsp); \
movq %r13, 24(%rsp); \
movq %r12, 16(%rsp); \
movq %rbx, 8(%rsp); \
movq %rbp, (%rsp); \
/* get the process pointer */ \
movq %rdi, P; \
/* switch to native stack */ \
SWITCH_C_TO_ERLANG
TEXT
/*
* int x86_call_to_native(Process *p);
* Emulated code recursively calls native code.
*/
.align 4
GLOBAL(CSYM(x86_call_to_native))
GLOBAL(ASYM(nbif_return))
CSYM(x86_call_to_native):
ENTER_FROM_C
/* get argument registers */
LOAD_ARG_REGS
/* call the target */
NSP_CALL(*P_NCALLEE(P))
/*
* We export this return address so that hipe_mode_switch() can discover
* when native code tailcalls emulated code.
*
* This is where native code returns to emulated code.
*/
ASYM(nbif_return):
movq %rax, P_ARG0(P) # save retval
movl $HIPE_MODE_SWITCH_RES_RETURN, %eax
/* FALLTHROUGH to .flush_exit
*
* Return to the calling C function with result token in %eax.
*
* .nosave_exit saves no state
* .flush_exit saves cached P state
* .suspend_exit also saves RA
*/
.suspend_exit:
/* save RA, no-op on x86 */
.flush_exit:
/* flush cached P state */
SAVE_CACHED_STATE
.nosave_exit:
/* switch to C stack */
SWITCH_ERLANG_TO_C_QUICK
/* restore C callee-save registers, drop frame, return */
movq (%rsp), %rbp # kills P
movq 8(%rsp), %rbx
movq 16(%rsp), %r12
movq 24(%rsp), %r13
movq 32(%rsp), %r14
movq 40(%rsp), %r15 # kills HP
addq $(7*8), %rsp
ret
/*
* Native code calls emulated code via a linker-generated
* stub (hipe_x86_loader.erl) which should look as follows:
*
* stub for f/N:
* movq $<f's BEAM code address>, P_BEAM_IP(P)
* movb $<N>, P_ARITY(P)
* jmp nbif_callemu
*
* XXX: Different stubs for different number of register parameters?
*/
.align 4
GLOBAL(ASYM(nbif_callemu))
ASYM(nbif_callemu):
STORE_ARG_REGS
movl $HIPE_MODE_SWITCH_RES_CALL, %eax
jmp .suspend_exit
/*
* nbif_apply
*/
.align 4
GLOBAL(ASYM(nbif_apply))
ASYM(nbif_apply):
STORE_ARG_REGS
movl $HIPE_MODE_SWITCH_RES_APPLY, %eax
jmp .suspend_exit
/*
* Native code calls an emulated-mode closure via a stub defined below.
*
* The closure is appended as the last actual parameter, and parameters
* beyond the first few passed in registers are pushed onto the stack in
* left-to-right order.
* Hence, the location of the closure parameter only depends on the number
* of parameters in registers, not the total number of parameters.
*/
#if NR_ARG_REGS >= 6
.align 4
GLOBAL(ASYM(nbif_ccallemu6))
ASYM(nbif_ccallemu6):
movq ARG5, P_ARG5(P)
#if NR_ARG_REGS > 6
movq ARG6, ARG5
#else
movq 8(NSP), ARG5
#endif
/*FALLTHROUGH*/
#endif
#if NR_ARG_REGS >= 5
.align 4
GLOBAL(ASYM(nbif_ccallemu5))
ASYM(nbif_ccallemu5):
movq ARG4, P_ARG4(P)
#if NR_ARG_REGS > 5
movq ARG5, ARG4
#else
movq 8(NSP), ARG4
#endif
/*FALLTHROUGH*/
#endif
#if NR_ARG_REGS >= 4
.align 4
GLOBAL(ASYM(nbif_ccallemu4))
ASYM(nbif_ccallemu4):
movq ARG3, P_ARG3(P)
#if NR_ARG_REGS > 4
movq ARG4, ARG3
#else
movq 8(NSP), ARG3
#endif
/*FALLTHROUGH*/
#endif
#if NR_ARG_REGS >= 3
.align 4
GLOBAL(ASYM(nbif_ccallemu3))
ASYM(nbif_ccallemu3):
movq ARG2, P_ARG2(P)
#if NR_ARG_REGS > 3
movq ARG3, ARG2
#else
movq 8(NSP), ARG2
#endif
/*FALLTHROUGH*/
#endif
#if NR_ARG_REGS >= 2
.align 4
GLOBAL(ASYM(nbif_ccallemu2))
ASYM(nbif_ccallemu2):
movq ARG1, P_ARG1(P)
#if NR_ARG_REGS > 2
movq ARG2, ARG1
#else
movq 8(NSP), ARG1
#endif
/*FALLTHROUGH*/
#endif
#if NR_ARG_REGS >= 1
.align 4
GLOBAL(ASYM(nbif_ccallemu1))
ASYM(nbif_ccallemu1):
movq ARG0, P_ARG0(P)
#if NR_ARG_REGS > 1
movq ARG1, ARG0
#else
movq 8(NSP), ARG0
#endif
/*FALLTHROUGH*/
#endif
.align 4
GLOBAL(ASYM(nbif_ccallemu0))
ASYM(nbif_ccallemu0):
/* We use %rsi not ARG0 here because ARG0 is not
defined when NR_ARG_REGS == 0. */
#if NR_ARG_REGS == 0
movq 8(NSP), %rsi
#endif
movq %rsi, P_CLOSURE(P)
movl $HIPE_MODE_SWITCH_RES_CALL_CLOSURE, %eax
jmp .suspend_exit
/*
* This is where native code suspends.
*/
.align 4
GLOBAL(ASYM(nbif_suspend_0))
ASYM(nbif_suspend_0):
movl $HIPE_MODE_SWITCH_RES_SUSPEND, %eax
jmp .suspend_exit
/*
* Suspend from a receive (waiting for a message)
*/
.align 4
GLOBAL(ASYM(nbif_suspend_msg))
ASYM(nbif_suspend_msg):
movl $HIPE_MODE_SWITCH_RES_WAIT, %eax
jmp .suspend_exit
/*
* Suspend from a receive with a timeout (waiting for a message)
* if (!(p->flags & F_TIMO)) { suspend }
* else { return 0; }
*/
.align 4
GLOBAL(ASYM(nbif_suspend_msg_timeout))
ASYM(nbif_suspend_msg_timeout):
movq P_FLAGS(P), %rax
/* this relies on F_TIMO (1<<2) fitting in a byte */
testb $F_TIMO, %al # F_TIMO set?
jz .no_timeout # if not set, suspend
/* timeout has occurred */
xorl %eax, %eax # return 0 to signal timeout
NSP_RET0
.no_timeout:
movl $HIPE_MODE_SWITCH_RES_WAIT_TIMEOUT, %eax
jmp .suspend_exit
/*
* int x86_return_to_native(Process *p);
* Emulated code returns to its native code caller.
*/
.align 4
GLOBAL(CSYM(x86_return_to_native))
CSYM(x86_return_to_native):
ENTER_FROM_C
/* get return value */
movq P_ARG0(P), %rax
/*
* Return using the stacked return address.
* The parameters were popped at the original native-to-emulated
* call (hipe_call_from_native_is_recursive), so a plain ret suffices.
*/
NSP_RET0
/*
* int x86_tailcall_to_native(Process *p);
* Emulated code tailcalls native code.
*/
.align 4
GLOBAL(CSYM(x86_tailcall_to_native))
CSYM(x86_tailcall_to_native):
ENTER_FROM_C
/* get argument registers */
LOAD_ARG_REGS
/* jump to the target label */
jmp *P_NCALLEE(P)
/*
* int x86_throw_to_native(Process *p);
* Emulated code throws an exception to its native code caller.
*/
.align 4
GLOBAL(CSYM(x86_throw_to_native))
CSYM(x86_throw_to_native):
ENTER_FROM_C
/* invoke the handler */
jmp *P_NCALLEE(P) # set by hipe_find_handler()
/*
* This is the default exception handler for native code.
*/
.align 4
GLOBAL(ASYM(nbif_fail))
ASYM(nbif_fail):
movl $HIPE_MODE_SWITCH_RES_THROW, %eax
jmp .flush_exit
GLOBAL(nbif_0_gc_after_bif)
GLOBAL(nbif_1_gc_after_bif)
GLOBAL(nbif_2_gc_after_bif)
GLOBAL(nbif_3_gc_after_bif)
.align 4
nbif_0_gc_after_bif:
xorl %edx, %edx
jmp .gc_after_bif
.align 4
nbif_1_gc_after_bif:
movl $1, %edx
jmp .gc_after_bif
.align 4
nbif_2_gc_after_bif:
movl $2, %edx
jmp .gc_after_bif
.align 4
nbif_3_gc_after_bif:
movl $3, %edx
/*FALLTHROUGH*/
.align 4
.gc_after_bif:
movl %edx, P_NARITY(P) # Note: narity is a 32-bit field
subq $(16-8), %rsp
movq P, %rdi
movq %rax, %rsi
xorl %edx, %edx # Pass NULL in regs
xorl %ecx, %ecx # Pass 0 in arity
call CSYM(erts_gc_after_bif_call)
addq $(16-8), %rsp
movl $0, P_NARITY(P) # Note: narity is a 32-bit field
ret
/*
* We end up here when a BIF called from native signals an
* exceptional condition.
* The stack/heap registers were just read from P.
*/
GLOBAL(nbif_0_simple_exception)
GLOBAL(nbif_1_simple_exception)
GLOBAL(nbif_2_simple_exception)
GLOBAL(nbif_3_simple_exception)
.align 4
nbif_0_simple_exception:
xorl %eax, %eax
jmp .nbif_simple_exception
.align 4
nbif_1_simple_exception:
movl $1, %eax
jmp .nbif_simple_exception
.align 4
nbif_2_simple_exception:
movl $2, %eax
jmp .nbif_simple_exception
.align 4
nbif_3_simple_exception:
movl $3, %eax
/*FALLTHROUGH*/
.align 4
.nbif_simple_exception:
cmpq $FREASON_TRAP, P_FREASON(P)
je .handle_trap
/*
* Find and invoke catch handler (it must exist).
* The stack/heap registers were just read from P.
* - %eax should contain the current call's arity
*/
movl %eax, P_NARITY(P) # Note: narity is a 32-bit field
/* find and prepare to invoke the handler */
SWITCH_ERLANG_TO_C_QUICK # The cached state is clean and need not be saved.
movq P, %rdi
call CSYM(hipe_handle_exception) # Note: hipe_handle_exception() conses
SWITCH_C_TO_ERLANG # %rsp updated by hipe_find_handler()
/* now invoke the handler */
jmp *P_NCALLEE(P) # set by hipe_find_handler()
/*
* A BIF failed with freason TRAP:
* - the BIF's arity is in %rax
* - the native heap/stack/reds registers are saved in P
*/
.handle_trap:
movl %eax, P_NARITY(P) # Note: narity is a 32-bit field
movl $HIPE_MODE_SWITCH_RES_TRAP, %eax
jmp .nosave_exit
/*
* nbif_stack_trap_ra: trap return address for maintaining
* the gray/white stack boundary
*/
GLOBAL(ASYM(nbif_stack_trap_ra))
.align 4
ASYM(nbif_stack_trap_ra): # a return address, not a function
# This only handles a single return value.
# If we have more, we need to save them in the PCB.
movq %rax, TEMP_RV # save retval
SWITCH_ERLANG_TO_C_QUICK
movq P, %rdi
call CSYM(hipe_handle_stack_trap) # must not cons; preserves TEMP_RV
movq %rax, %rdx # original RA
SWITCH_C_TO_ERLANG_QUICK
movq TEMP_RV, %rax # restore retval
jmp *%rdx # resume at original RA
/*
* nbif_inc_stack_0
*/
GLOBAL(ASYM(nbif_inc_stack_0))
.align 4
ASYM(nbif_inc_stack_0):
SWITCH_ERLANG_TO_C_QUICK
STORE_ARG_REGS
movq P, %rdi
# hipe_inc_nstack reads and writes NSP and NSP_LIMIT,
# but does not access HP or FCALLS (or the non-amd64 NRA).
call CSYM(hipe_inc_nstack)
LOAD_ARG_REGS
SWITCH_C_TO_ERLANG_QUICK
NSP_RET0
#if defined(__linux__) && defined(__ELF__)
.section .note.GNU-stack,"",%progbits
#endif
