diff options
Diffstat (limited to 'libffi/src')
| -rw-r--r-- | libffi/src/aarch64/ffi.c | 1076 | ||||
| -rw-r--r-- | libffi/src/aarch64/ffitarget.h | 59 | ||||
| -rw-r--r-- | libffi/src/aarch64/sysv.S | 307 | ||||
| -rw-r--r-- | libffi/src/arm/ffi.c | 4 | ||||
| -rw-r--r-- | libffi/src/arm/sysv.S | 13 | ||||
| -rw-r--r-- | libffi/src/arm/trampoline.S | 2 | ||||
| -rw-r--r-- | libffi/src/bfin/ffi.c | 195 | ||||
| -rw-r--r-- | libffi/src/bfin/ffitarget.h | 43 | ||||
| -rw-r--r-- | libffi/src/bfin/sysv.S | 177 | ||||
| -rw-r--r-- | libffi/src/closures.c | 27 | ||||
| -rw-r--r-- | libffi/src/m68k/ffi.c | 83 | ||||
| -rw-r--r-- | libffi/src/m68k/sysv.S | 53 | ||||
| -rw-r--r-- | libffi/src/powerpc/ffi_darwin.c | 6 | ||||
| -rw-r--r-- | libffi/src/prep_cif.c | 13 | ||||
| -rw-r--r-- | libffi/src/tile/ffi.c | 355 | ||||
| -rw-r--r-- | libffi/src/tile/ffitarget.h | 65 | ||||
| -rw-r--r-- | libffi/src/tile/tile.S | 360 | ||||
| -rw-r--r-- | libffi/src/x86/ffi.c | 29 | ||||
| -rw-r--r-- | libffi/src/x86/ffi64.c | 8 | ||||
| -rw-r--r-- | libffi/src/x86/ffitarget.h | 7 | ||||
| -rw-r--r-- | libffi/src/x86/win32.S | 168 |
21 files changed, 2947 insertions, 103 deletions
diff --git a/libffi/src/aarch64/ffi.c b/libffi/src/aarch64/ffi.c new file mode 100644 index 0000000..1405665 --- /dev/null +++ b/libffi/src/aarch64/ffi.c @@ -0,0 +1,1076 @@ +/* Copyright (c) 2009, 2010, 2011, 2012 ARM Ltd. + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +``Software''), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice shall be +included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY +CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE +SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include <stdio.h> + +#include <ffi.h> +#include <ffi_common.h> + +#include <stdlib.h> + +/* Stack alignment requirement in bytes */ +#define AARCH64_STACK_ALIGN 16 + +#define N_X_ARG_REG 8 +#define N_V_ARG_REG 8 + +#define AARCH64_FFI_WITH_V (1 << AARCH64_FFI_WITH_V_BIT) + +union _d +{ + UINT64 d; + UINT32 s[2]; +}; + +struct call_context +{ + UINT64 x [AARCH64_N_XREG]; + struct + { + union _d d[2]; + } v [AARCH64_N_VREG]; +}; + +static void * +get_x_addr (struct call_context *context, unsigned n) +{ + return &context->x[n]; +} + +static void * +get_s_addr (struct call_context *context, unsigned n) +{ +#if defined __AARCH64EB__ + return &context->v[n].d[1].s[1]; +#else + return &context->v[n].d[0].s[0]; +#endif +} + +static void * +get_d_addr (struct call_context *context, unsigned n) +{ +#if defined __AARCH64EB__ + return &context->v[n].d[1]; +#else + return &context->v[n].d[0]; +#endif +} + +static void * +get_v_addr (struct call_context *context, unsigned n) +{ + return &context->v[n]; +} + +/* Return the memory location at which a basic type would reside + were it to have been stored in register n. */ + +static void * +get_basic_type_addr (unsigned short type, struct call_context *context, + unsigned n) +{ + switch (type) + { + case FFI_TYPE_FLOAT: + return get_s_addr (context, n); + case FFI_TYPE_DOUBLE: + return get_d_addr (context, n); + case FFI_TYPE_LONGDOUBLE: + return get_v_addr (context, n); + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_INT: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + return get_x_addr (context, n); + default: + FFI_ASSERT (0); + return NULL; + } +} + +/* Return the alignment width for each of the basic types. */ + +static size_t +get_basic_type_alignment (unsigned short type) +{ + switch (type) + { + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + return sizeof (UINT64); + case FFI_TYPE_LONGDOUBLE: + return sizeof (long double); + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_INT: + case FFI_TYPE_SINT32: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + return sizeof (UINT64); + + default: + FFI_ASSERT (0); + return 0; + } +} + +/* Return the size in bytes for each of the basic types. */ + +static size_t +get_basic_type_size (unsigned short type) +{ + switch (type) + { + case FFI_TYPE_FLOAT: + return sizeof (UINT32); + case FFI_TYPE_DOUBLE: + return sizeof (UINT64); + case FFI_TYPE_LONGDOUBLE: + return sizeof (long double); + case FFI_TYPE_UINT8: + return sizeof (UINT8); + case FFI_TYPE_SINT8: + return sizeof (SINT8); + case FFI_TYPE_UINT16: + return sizeof (UINT16); + case FFI_TYPE_SINT16: + return sizeof (SINT16); + case FFI_TYPE_UINT32: + return sizeof (UINT32); + case FFI_TYPE_INT: + case FFI_TYPE_SINT32: + return sizeof (SINT32); + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + return sizeof (UINT64); + case FFI_TYPE_SINT64: + return sizeof (SINT64); + + default: + FFI_ASSERT (0); + return 0; + } +} + +extern void +ffi_call_SYSV (unsigned (*)(struct call_context *context, unsigned char *, + extended_cif *), + struct call_context *context, + extended_cif *, + unsigned, + void (*fn)(void)); + +extern void +ffi_closure_SYSV (ffi_closure *); + +/* Test for an FFI floating point representation. */ + +static unsigned +is_floating_type (unsigned short type) +{ + return (type == FFI_TYPE_FLOAT || type == FFI_TYPE_DOUBLE + || type == FFI_TYPE_LONGDOUBLE); +} + +/* Test for a homogeneous structure. */ + +static unsigned short +get_homogeneous_type (ffi_type *ty) +{ + if (ty->type == FFI_TYPE_STRUCT && ty->elements) + { + unsigned i; + unsigned short candidate_type + = get_homogeneous_type (ty->elements[0]); + for (i =1; ty->elements[i]; i++) + { + unsigned short iteration_type = 0; + /* If we have a nested struct, we must find its homogeneous type. + If that fits with our candidate type, we are still + homogeneous. */ + if (ty->elements[i]->type == FFI_TYPE_STRUCT + && ty->elements[i]->elements) + { + iteration_type = get_homogeneous_type (ty->elements[i]); + } + else + { + iteration_type = ty->elements[i]->type; + } + + /* If we are not homogeneous, return FFI_TYPE_STRUCT. */ + if (candidate_type != iteration_type) + return FFI_TYPE_STRUCT; + } + return candidate_type; + } + + /* Base case, we have no more levels of nesting, so we + are a basic type, and so, trivially homogeneous in that type. */ + return ty->type; +} + +/* Determine the number of elements within a STRUCT. + + Note, we must handle nested structs. + + If ty is not a STRUCT this function will return 0. */ + +static unsigned +element_count (ffi_type *ty) +{ + if (ty->type == FFI_TYPE_STRUCT && ty->elements) + { + unsigned n; + unsigned elems = 0; + for (n = 0; ty->elements[n]; n++) + { + if (ty->elements[n]->type == FFI_TYPE_STRUCT + && ty->elements[n]->elements) + elems += element_count (ty->elements[n]); + else + elems++; + } + return elems; + } + return 0; +} + +/* Test for a homogeneous floating point aggregate. + + A homogeneous floating point aggregate is a homogeneous aggregate of + a half- single- or double- precision floating point type with one + to four elements. Note that this includes nested structs of the + basic type. */ + +static int +is_hfa (ffi_type *ty) +{ + if (ty->type == FFI_TYPE_STRUCT + && ty->elements[0] + && is_floating_type (get_homogeneous_type (ty))) + { + unsigned n = element_count (ty); + return n >= 1 && n <= 4; + } + return 0; +} + +/* Test if an ffi_type is a candidate for passing in a register. + + This test does not check that sufficient registers of the + appropriate class are actually available, merely that IFF + sufficient registers are available then the argument will be passed + in register(s). + + Note that an ffi_type that is deemed to be a register candidate + will always be returned in registers. + + Returns 1 if a register candidate else 0. */ + +static int +is_register_candidate (ffi_type *ty) +{ + switch (ty->type) + { + case FFI_TYPE_VOID: + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + case FFI_TYPE_LONGDOUBLE: + case FFI_TYPE_UINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_UINT64: + case FFI_TYPE_POINTER: + case FFI_TYPE_SINT8: + case FFI_TYPE_SINT16: + case FFI_TYPE_SINT32: + case FFI_TYPE_INT: + case FFI_TYPE_SINT64: + return 1; + + case FFI_TYPE_STRUCT: + if (is_hfa (ty)) + { + return 1; + } + else if (ty->size > 16) + { + /* Too large. Will be replaced with a pointer to memory. The + pointer MAY be passed in a register, but the value will + not. This test specifically fails since the argument will + never be passed by value in registers. */ + return 0; + } + else + { + /* Might be passed in registers depending on the number of + registers required. */ + return (ty->size + 7) / 8 < N_X_ARG_REG; + } + break; + + default: + FFI_ASSERT (0); + break; + } + + return 0; +} + +/* Test if an ffi_type argument or result is a candidate for a vector + register. */ + +static int +is_v_register_candidate (ffi_type *ty) +{ + return is_floating_type (ty->type) + || (ty->type == FFI_TYPE_STRUCT && is_hfa (ty)); +} + +/* Representation of the procedure call argument marshalling + state. + + The terse state variable names match the names used in the AARCH64 + PCS. */ + +struct arg_state +{ + unsigned ngrn; /* Next general-purpose register number. */ + unsigned nsrn; /* Next vector register number. */ + unsigned nsaa; /* Next stack offset. */ +}; + +/* Initialize a procedure call argument marshalling state. */ +static void +arg_init (struct arg_state *state, unsigned call_frame_size) +{ + state->ngrn = 0; + state->nsrn = 0; + state->nsaa = 0; +} + +/* Return the number of available consecutive core argument + registers. */ + +static unsigned +available_x (struct arg_state *state) +{ + return N_X_ARG_REG - state->ngrn; +} + +/* Return the number of available consecutive vector argument + registers. */ + +static unsigned +available_v (struct arg_state *state) +{ + return N_V_ARG_REG - state->nsrn; +} + +static void * +allocate_to_x (struct call_context *context, struct arg_state *state) +{ + FFI_ASSERT (state->ngrn < N_X_ARG_REG) + return get_x_addr (context, (state->ngrn)++); +} + +static void * +allocate_to_s (struct call_context *context, struct arg_state *state) +{ + FFI_ASSERT (state->nsrn < N_V_ARG_REG) + return get_s_addr (context, (state->nsrn)++); +} + +static void * +allocate_to_d (struct call_context *context, struct arg_state *state) +{ + FFI_ASSERT (state->nsrn < N_V_ARG_REG) + return get_d_addr (context, (state->nsrn)++); +} + +static void * +allocate_to_v (struct call_context *context, struct arg_state *state) +{ + FFI_ASSERT (state->nsrn < N_V_ARG_REG) + return get_v_addr (context, (state->nsrn)++); +} + +/* Allocate an aligned slot on the stack and return a pointer to it. */ +static void * +allocate_to_stack (struct arg_state *state, void *stack, unsigned alignment, + unsigned size) +{ + void *allocation; + + /* Round up the NSAA to the larger of 8 or the natural + alignment of the argument's type. */ + state->nsaa = ALIGN (state->nsaa, alignment); + state->nsaa = ALIGN (state->nsaa, alignment); + state->nsaa = ALIGN (state->nsaa, 8); + + allocation = stack + state->nsaa; + + state->nsaa += size; + return allocation; +} + +static void +copy_basic_type (void *dest, void *source, unsigned short type) +{ + /* This is neccessary to ensure that basic types are copied + sign extended to 64-bits as libffi expects. */ + switch (type) + { + case FFI_TYPE_FLOAT: + *(float *) dest = *(float *) source; + break; + case FFI_TYPE_DOUBLE: + *(double *) dest = *(double *) source; + break; + case FFI_TYPE_LONGDOUBLE: + *(long double *) dest = *(long double *) source; + break; + case FFI_TYPE_UINT8: + *(ffi_arg *) dest = *(UINT8 *) source; + break; + case FFI_TYPE_SINT8: + *(ffi_sarg *) dest = *(SINT8 *) source; + break; + case FFI_TYPE_UINT16: + *(ffi_arg *) dest = *(UINT16 *) source; + break; + case FFI_TYPE_SINT16: + *(ffi_sarg *) dest = *(SINT16 *) source; + break; + case FFI_TYPE_UINT32: + *(ffi_arg *) dest = *(UINT32 *) source; + break; + case FFI_TYPE_INT: + case FFI_TYPE_SINT32: + *(ffi_sarg *) dest = *(SINT32 *) source; + break; + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + *(ffi_arg *) dest = *(UINT64 *) source; + break; + case FFI_TYPE_SINT64: + *(ffi_sarg *) dest = *(SINT64 *) source; + break; + + default: + FFI_ASSERT (0); + } +} + +static void +copy_hfa_to_reg_or_stack (void *memory, + ffi_type *ty, + struct call_context *context, + unsigned char *stack, + struct arg_state *state) +{ + unsigned elems = element_count (ty); + if (available_v (state) < elems) + { + /* There are insufficient V registers. Further V register allocations + are prevented, the NSAA is adjusted (by allocate_to_stack ()) + and the argument is copied to memory at the adjusted NSAA. */ + state->nsrn = N_V_ARG_REG; + memcpy (allocate_to_stack (state, stack, ty->alignment, ty->size), + memory, + ty->size); + } + else + { + int i; + unsigned short type = get_homogeneous_type (ty); + unsigned elems = element_count (ty); + for (i = 0; i < elems; i++) + { + void *reg = allocate_to_v (context, state); + copy_basic_type (reg, memory, type); + memory += get_basic_type_size (type); + } + } +} + +/* Either allocate an appropriate register for the argument type, or if + none are available, allocate a stack slot and return a pointer + to the allocated space. */ + +static void * +allocate_to_register_or_stack (struct call_context *context, + unsigned char *stack, + struct arg_state *state, + unsigned short type) +{ + size_t alignment = get_basic_type_alignment (type); + size_t size = alignment; + switch (type) + { + case FFI_TYPE_FLOAT: + /* This is the only case for which the allocated stack size + should not match the alignment of the type. */ + size = sizeof (UINT32); + /* Fall through. */ + case FFI_TYPE_DOUBLE: + if (state->nsrn < N_V_ARG_REG) + return allocate_to_d (context, state); + state->nsrn = N_V_ARG_REG; + break; + case FFI_TYPE_LONGDOUBLE: + if (state->nsrn < N_V_ARG_REG) + return allocate_to_v (context, state); + state->nsrn = N_V_ARG_REG; + break; + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_INT: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + if (state->ngrn < N_X_ARG_REG) + return allocate_to_x (context, state); + state->ngrn = N_X_ARG_REG; + break; + default: + FFI_ASSERT (0); + } + + return allocate_to_stack (state, stack, alignment, size); +} + +/* Copy a value to an appropriate register, or if none are + available, to the stack. */ + +static void +copy_to_register_or_stack (struct call_context *context, + unsigned char *stack, + struct arg_state *state, + void *value, + unsigned short type) +{ + copy_basic_type ( + allocate_to_register_or_stack (context, stack, state, type), + value, + type); +} + +/* Marshall the arguments from FFI representation to procedure call + context and stack. */ + +static unsigned +aarch64_prep_args (struct call_context *context, unsigned char *stack, + extended_cif *ecif) +{ + int i; + struct arg_state state; + + arg_init (&state, ALIGN(ecif->cif->bytes, 16)); + + for (i = 0; i < ecif->cif->nargs; i++) + { + ffi_type *ty = ecif->cif->arg_types[i]; + switch (ty->type) + { + case FFI_TYPE_VOID: + FFI_ASSERT (0); + break; + + /* If the argument is a basic type the argument is allocated to an + appropriate register, or if none are available, to the stack. */ + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + case FFI_TYPE_LONGDOUBLE: + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_INT: + case FFI_TYPE_SINT32: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + copy_to_register_or_stack (context, stack, &state, + ecif->avalue[i], ty->type); + break; + + case FFI_TYPE_STRUCT: + if (is_hfa (ty)) + { + copy_hfa_to_reg_or_stack (ecif->avalue[i], ty, context, + stack, &state); + } + else if (ty->size > 16) + { + /* If the argument is a composite type that is larger than 16 + bytes, then the argument has been copied to memory, and + the argument is replaced by a pointer to the copy. */ + + copy_to_register_or_stack (context, stack, &state, + &(ecif->avalue[i]), FFI_TYPE_POINTER); + } + else if (available_x (&state) >= (ty->size + 7) / 8) + { + /* If the argument is a composite type and the size in + double-words is not more than the number of available + X registers, then the argument is copied into consecutive + X registers. */ + int j; + for (j = 0; j < (ty->size + 7) / 8; j++) + { + memcpy (allocate_to_x (context, &state), + &(((UINT64 *) ecif->avalue[i])[j]), + sizeof (UINT64)); + } + } + else + { + /* Otherwise, there are insufficient X registers. Further X + register allocations are prevented, the NSAA is adjusted + (by allocate_to_stack ()) and the argument is copied to + memory at the adjusted NSAA. */ + state.ngrn = N_X_ARG_REG; + + memcpy (allocate_to_stack (&state, stack, ty->alignment, + ty->size), ecif->avalue + i, ty->size); + } + break; + + default: + FFI_ASSERT (0); + break; + } + } + + return ecif->cif->aarch64_flags; +} + +ffi_status +ffi_prep_cif_machdep (ffi_cif *cif) +{ + /* Round the stack up to a multiple of the stack alignment requirement. */ + cif->bytes = + (cif->bytes + (AARCH64_STACK_ALIGN - 1)) & ~ (AARCH64_STACK_ALIGN - 1); + + /* Initialize our flags. We are interested if this CIF will touch a + vector register, if so we will enable context save and load to + those registers, otherwise not. This is intended to be friendly + to lazy float context switching in the kernel. */ + cif->aarch64_flags = 0; + + if (is_v_register_candidate (cif->rtype)) + { + cif->aarch64_flags |= AARCH64_FFI_WITH_V; + } + else + { + int i; + for (i = 0; i < cif->nargs; i++) + if (is_v_register_candidate (cif->arg_types[i])) + { + cif->aarch64_flags |= AARCH64_FFI_WITH_V; + break; + } + } + + return FFI_OK; +} + +/* Call a function with the provided arguments and capture the return + value. */ +void +ffi_call (ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue) +{ + extended_cif ecif; + + ecif.cif = cif; + ecif.avalue = avalue; + ecif.rvalue = rvalue; + + switch (cif->abi) + { + case FFI_SYSV: + { + struct call_context context; + unsigned stack_bytes; + + /* Figure out the total amount of stack space we need, the + above call frame space needs to be 16 bytes aligned to + ensure correct alignment of the first object inserted in + that space hence the ALIGN applied to cif->bytes.*/ + stack_bytes = ALIGN(cif->bytes, 16); + + memset (&context, 0, sizeof (context)); + if (is_register_candidate (cif->rtype)) + { + ffi_call_SYSV (aarch64_prep_args, &context, &ecif, stack_bytes, fn); + switch (cif->rtype->type) + { + case FFI_TYPE_VOID: + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + case FFI_TYPE_LONGDOUBLE: + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_INT: + case FFI_TYPE_SINT64: + { + void *addr = get_basic_type_addr (cif->rtype->type, + &context, 0); + copy_basic_type (rvalue, addr, cif->rtype->type); + break; + } + + case FFI_TYPE_STRUCT: + if (is_hfa (cif->rtype)) + { + int j; + unsigned short type = get_homogeneous_type (cif->rtype); + unsigned elems = element_count (cif->rtype); + for (j = 0; j < elems; j++) + { + void *reg = get_basic_type_addr (type, &context, j); + copy_basic_type (rvalue, reg, type); + rvalue += get_basic_type_size (type); + } + } + else if ((cif->rtype->size + 7) / 8 < N_X_ARG_REG) + { + unsigned size = ALIGN (cif->rtype->size, sizeof (UINT64)); + memcpy (rvalue, get_x_addr (&context, 0), size); + } + else + { + FFI_ASSERT (0); + } + break; + + default: + FFI_ASSERT (0); + break; + } + } + else + { + memcpy (get_x_addr (&context, 8), &rvalue, sizeof (UINT64)); + ffi_call_SYSV (aarch64_prep_args, &context, &ecif, + stack_bytes, fn); + } + break; + } + + default: + FFI_ASSERT (0); + break; + } +} + +static unsigned char trampoline [] = +{ 0x70, 0x00, 0x00, 0x58, /* ldr x16, 1f */ + 0x91, 0x00, 0x00, 0x10, /* adr x17, 2f */ + 0x00, 0x02, 0x1f, 0xd6 /* br x16 */ +}; + +/* Build a trampoline. */ + +#define FFI_INIT_TRAMPOLINE(TRAMP,FUN,CTX,FLAGS) \ + ({unsigned char *__tramp = (unsigned char*)(TRAMP); \ + UINT64 __fun = (UINT64)(FUN); \ + UINT64 __ctx = (UINT64)(CTX); \ + UINT64 __flags = (UINT64)(FLAGS); \ + memcpy (__tramp, trampoline, sizeof (trampoline)); \ + memcpy (__tramp + 12, &__fun, sizeof (__fun)); \ + memcpy (__tramp + 20, &__ctx, sizeof (__ctx)); \ + memcpy (__tramp + 28, &__flags, sizeof (__flags)); \ + __clear_cache(__tramp, __tramp + FFI_TRAMPOLINE_SIZE); \ + }) + +ffi_status +ffi_prep_closure_loc (ffi_closure* closure, + ffi_cif* cif, + void (*fun)(ffi_cif*,void*,void**,void*), + void *user_data, + void *codeloc) +{ + if (cif->abi != FFI_SYSV) + return FFI_BAD_ABI; + + FFI_INIT_TRAMPOLINE (&closure->tramp[0], &ffi_closure_SYSV, codeloc, + cif->aarch64_flags); + + closure->cif = cif; + closure->user_data = user_data; + closure->fun = fun; + + return FFI_OK; +} + +/* Primary handler to setup and invoke a function within a closure. + + A closure when invoked enters via the assembler wrapper + ffi_closure_SYSV(). The wrapper allocates a call context on the + stack, saves the interesting registers (from the perspective of + the calling convention) into the context then passes control to + ffi_closure_SYSV_inner() passing the saved context and a pointer to + the stack at the point ffi_closure_SYSV() was invoked. + + On the return path the assembler wrapper will reload call context + regsiters. + + ffi_closure_SYSV_inner() marshalls the call context into ffi value + desriptors, invokes the wrapped function, then marshalls the return + value back into the call context. */ + +void +ffi_closure_SYSV_inner (ffi_closure *closure, struct call_context *context, + void *stack) +{ + ffi_cif *cif = closure->cif; + void **avalue = (void**) alloca (cif->nargs * sizeof (void*)); + void *rvalue = NULL; + int i; + struct arg_state state; + + arg_init (&state, ALIGN(cif->bytes, 16)); + + for (i = 0; i < cif->nargs; i++) + { + ffi_type *ty = cif->arg_types[i]; + + switch (ty->type) + { + case FFI_TYPE_VOID: + FFI_ASSERT (0); + break; + + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_INT: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + case FFI_TYPE_LONGDOUBLE: + avalue[i] = allocate_to_register_or_stack (context, stack, + &state, ty->type); + break; + + case FFI_TYPE_STRUCT: + if (is_hfa (ty)) + { + unsigned n = element_count (ty); + if (available_v (&state) < n) + { + state.nsrn = N_V_ARG_REG; + avalue[i] = allocate_to_stack (&state, stack, ty->alignment, + ty->size); + } + else + { + switch (get_homogeneous_type (ty)) + { + case FFI_TYPE_FLOAT: + { + /* Eeek! We need a pointer to the structure, + however the homogeneous float elements are + being passed in individual S registers, + therefore the structure is not represented as + a contiguous sequence of bytes in our saved + register context. We need to fake up a copy + of the structure layed out in memory + correctly. The fake can be tossed once the + closure function has returned hence alloca() + is sufficient. */ + int j; + UINT32 *p = avalue[i] = alloca (ty->size); + for (j = 0; j < element_count (ty); j++) + memcpy (&p[j], + allocate_to_s (context, &state), + sizeof (*p)); + break; + } + + case FFI_TYPE_DOUBLE: + { + /* Eeek! We need a pointer to the structure, + however the homogeneous float elements are + being passed in individual S registers, + therefore the structure is not represented as + a contiguous sequence of bytes in our saved + register context. We need to fake up a copy + of the structure layed out in memory + correctly. The fake can be tossed once the + closure function has returned hence alloca() + is sufficient. */ + int j; + UINT64 *p = avalue[i] = alloca (ty->size); + for (j = 0; j < element_count (ty); j++) + memcpy (&p[j], + allocate_to_d (context, &state), + sizeof (*p)); + break; + } + + case FFI_TYPE_LONGDOUBLE: + memcpy (&avalue[i], + allocate_to_v (context, &state), + sizeof (*avalue)); + break; + + default: + FFI_ASSERT (0); + break; + } + } + } + else if (ty->size > 16) + { + /* Replace Composite type of size greater than 16 with a + pointer. */ + memcpy (&avalue[i], + allocate_to_register_or_stack (context, stack, + &state, FFI_TYPE_POINTER), + sizeof (avalue[i])); + } + else if (available_x (&state) >= (ty->size + 7) / 8) + { + avalue[i] = get_x_addr (context, state.ngrn); + state.ngrn += (ty->size + 7) / 8; + } + else + { + state.ngrn = N_X_ARG_REG; + + avalue[i] = allocate_to_stack (&state, stack, ty->alignment, + ty->size); + } + break; + + default: + FFI_ASSERT (0); + break; + } + } + + /* Figure out where the return value will be passed, either in + registers or in a memory block allocated by the caller and passed + in x8. */ + + if (is_register_candidate (cif->rtype)) + { + /* Register candidates are *always* returned in registers. */ + + /* Allocate a scratchpad for the return value, we will let the + callee scrible the result into the scratch pad then move the + contents into the appropriate return value location for the + call convention. */ + rvalue = alloca (cif->rtype->size); + (closure->fun) (cif, rvalue, avalue, closure->user_data); + + /* Copy the return value into the call context so that it is returned + as expected to our caller. */ + switch (cif->rtype->type) + { + case FFI_TYPE_VOID: + break; + + case FFI_TYPE_UINT8: + case FFI_TYPE_UINT16: + case FFI_TYPE_UINT32: + case FFI_TYPE_POINTER: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT8: + case FFI_TYPE_SINT16: + case FFI_TYPE_INT: + case FFI_TYPE_SINT32: + case FFI_TYPE_SINT64: + case FFI_TYPE_FLOAT: + case FFI_TYPE_DOUBLE: + case FFI_TYPE_LONGDOUBLE: + { + void *addr = get_basic_type_addr (cif->rtype->type, context, 0); + copy_basic_type (addr, rvalue, cif->rtype->type); + break; + } + case FFI_TYPE_STRUCT: + if (is_hfa (cif->rtype)) + { + int i; + unsigned short type = get_homogeneous_type (cif->rtype); + unsigned elems = element_count (cif->rtype); + for (i = 0; i < elems; i++) + { + void *reg = get_basic_type_addr (type, context, i); + copy_basic_type (reg, rvalue, type); + rvalue += get_basic_type_size (type); + } + } + else if ((cif->rtype->size + 7) / 8 < N_X_ARG_REG) + { + unsigned size = ALIGN (cif->rtype->size, sizeof (UINT64)) ; + memcpy (get_x_addr (context, 0), rvalue, size); + } + else + { + FFI_ASSERT (0); + } + break; + default: + FFI_ASSERT (0); + break; + } + } + else + { + memcpy (&rvalue, get_x_addr (context, 8), sizeof (UINT64)); + (closure->fun) (cif, rvalue, avalue, closure->user_data); + } +} + diff --git a/libffi/src/aarch64/ffitarget.h b/libffi/src/aarch64/ffitarget.h new file mode 100644 index 0000000..6f1a348 --- /dev/null +++ b/libffi/src/aarch64/ffitarget.h @@ -0,0 +1,59 @@ +/* Copyright (c) 2009, 2010, 2011, 2012 ARM Ltd. + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +``Software''), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice shall be +included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY +CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE +SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef LIBFFI_TARGET_H +#define LIBFFI_TARGET_H + +#ifndef LIBFFI_H +#error "Please do not include ffitarget.h directly into your source. Use ffi.h instead." +#endif + +#ifndef LIBFFI_ASM +typedef unsigned long ffi_arg; +typedef signed long ffi_sarg; + +typedef enum ffi_abi + { + FFI_FIRST_ABI = 0, + FFI_SYSV, + FFI_LAST_ABI, + FFI_DEFAULT_ABI = FFI_SYSV + } ffi_abi; +#endif + +/* ---- Definitions for closures ----------------------------------------- */ + +#define FFI_CLOSURES 1 +#define FFI_TRAMPOLINE_SIZE 36 +#define FFI_NATIVE_RAW_API 0 + +/* ---- Internal ---- */ + + +#define FFI_EXTRA_CIF_FIELDS unsigned aarch64_flags + +#define AARCH64_FFI_WITH_V_BIT 0 + +#define AARCH64_N_XREG 32 +#define AARCH64_N_VREG 32 +#define AARCH64_CALL_CONTEXT_SIZE (AARCH64_N_XREG * 8 + AARCH64_N_VREG * 16) + +#endif diff --git a/libffi/src/aarch64/sysv.S b/libffi/src/aarch64/sysv.S new file mode 100644 index 0000000..b8cd421 --- /dev/null +++ b/libffi/src/aarch64/sysv.S @@ -0,0 +1,307 @@ +/* Copyright (c) 2009, 2010, 2011, 2012 ARM Ltd. + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +``Software''), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice shall be +included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY +CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE +SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#define LIBFFI_ASM +#include <fficonfig.h> +#include <ffi.h> + +#define cfi_adjust_cfa_offset(off) .cfi_adjust_cfa_offset off +#define cfi_rel_offset(reg, off) .cfi_rel_offset reg, off +#define cfi_restore(reg) .cfi_restore reg +#define cfi_def_cfa_register(reg) .cfi_def_cfa_register reg + + .text + .globl ffi_call_SYSV + .type ffi_call_SYSV, #function + +/* ffi_call_SYSV() + + Create a stack frame, setup an argument context, call the callee + and extract the result. + + The maximum required argument stack size is provided, + ffi_call_SYSV() allocates that stack space then calls the + prepare_fn to populate register context and stack. The + argument passing registers are loaded from the register + context and the callee called, on return the register passing + register are saved back to the context. Our caller will + extract the return value from the final state of the saved + register context. + + Prototype: + + extern unsigned + ffi_call_SYSV (void (*)(struct call_context *context, unsigned char *, + extended_cif *), + struct call_context *context, + extended_cif *, + unsigned required_stack_size, + void (*fn)(void)); + + Therefore on entry we have: + + x0 prepare_fn + x1 &context + x2 &ecif + x3 bytes + x4 fn + + This function uses the following stack frame layout: + + == + saved x30(lr) + x29(fp)-> saved x29(fp) + saved x24 + saved x23 + saved x22 + sp' -> saved x21 + ... + sp -> (constructed callee stack arguments) + == + + Voila! */ + +#define ffi_call_SYSV_FS (8 * 4) + + .cfi_startproc +ffi_call_SYSV: + stp x29, x30, [sp, #-16]! + cfi_adjust_cfa_offset (16) + cfi_rel_offset (x29, 0) + cfi_rel_offset (x30, 8) + + mov x29, sp + cfi_def_cfa_register (x29) + sub sp, sp, #ffi_call_SYSV_FS + + stp x21, x22, [sp, 0] + cfi_rel_offset (x21, 0 - ffi_call_SYSV_FS) + cfi_rel_offset (x22, 8 - ffi_call_SYSV_FS) + + stp x23, x24, [sp, 16] + cfi_rel_offset (x23, 16 - ffi_call_SYSV_FS) + cfi_rel_offset (x24, 24 - ffi_call_SYSV_FS) + + mov x21, x1 + mov x22, x2 + mov x24, x4 + + /* Allocate the stack space for the actual arguments, many + arguments will be passed in registers, but we assume + worst case and allocate sufficient stack for ALL of + the arguments. */ + sub sp, sp, x3 + + /* unsigned (*prepare_fn) (struct call_context *context, + unsigned char *stack, extended_cif *ecif); + */ + mov x23, x0 + mov x0, x1 + mov x1, sp + /* x2 already in place */ + blr x23 + + /* Preserve the flags returned. */ + mov x23, x0 + + /* Figure out if we should touch the vector registers. */ + tbz x23, #AARCH64_FFI_WITH_V_BIT, 1f + + /* Load the vector argument passing registers. */ + ldp q0, q1, [x21, #8*32 + 0] + ldp q2, q3, [x21, #8*32 + 32] + ldp q4, q5, [x21, #8*32 + 64] + ldp q6, q7, [x21, #8*32 + 96] +1: + /* Load the core argument passing registers. */ + ldp x0, x1, [x21, #0] + ldp x2, x3, [x21, #16] + ldp x4, x5, [x21, #32] + ldp x6, x7, [x21, #48] + + /* Don't forget x8 which may be holding the address of a return buffer. + */ + ldr x8, [x21, #8*8] + + blr x24 + + /* Save the core argument passing registers. */ + stp x0, x1, [x21, #0] + stp x2, x3, [x21, #16] + stp x4, x5, [x21, #32] + stp x6, x7, [x21, #48] + + /* Note nothing useful ever comes back in x8! */ + + /* Figure out if we should touch the vector registers. */ + tbz x23, #AARCH64_FFI_WITH_V_BIT, 1f + + /* Save the vector argument passing registers. */ + stp q0, q1, [x21, #8*32 + 0] + stp q2, q3, [x21, #8*32 + 32] + stp q4, q5, [x21, #8*32 + 64] + stp q6, q7, [x21, #8*32 + 96] +1: + /* All done, unwind our stack frame. */ + ldp x21, x22, [x29, # - ffi_call_SYSV_FS] + cfi_restore (x21) + cfi_restore (x22) + + ldp x23, x24, [x29, # - ffi_call_SYSV_FS + 16] + cfi_restore (x23) + cfi_restore (x24) + + mov sp, x29 + cfi_def_cfa_register (sp) + + ldp x29, x30, [sp], #16 + cfi_adjust_cfa_offset (-16) + cfi_restore (x29) + cfi_restore (x30) + + ret + + .cfi_endproc + .size ffi_call_SYSV, .-ffi_call_SYSV + +#define ffi_closure_SYSV_FS (8 * 2 + AARCH64_CALL_CONTEXT_SIZE) + +/* ffi_closure_SYSV + + Closure invocation glue. This is the low level code invoked directly by + the closure trampoline to setup and call a closure. + + On entry x17 points to a struct trampoline_data, x16 has been clobbered + all other registers are preserved. + + We allocate a call context and save the argument passing registers, + then invoked the generic C ffi_closure_SYSV_inner() function to do all + the real work, on return we load the result passing registers back from + the call context. + + On entry + + extern void + ffi_closure_SYSV (struct trampoline_data *); + + struct trampoline_data + { + UINT64 *ffi_closure; + UINT64 flags; + }; + + This function uses the following stack frame layout: + + == + saved x30(lr) + x29(fp)-> saved x29(fp) + saved x22 + saved x21 + ... + sp -> call_context + == + + Voila! */ + + .text + .globl ffi_closure_SYSV + .cfi_startproc +ffi_closure_SYSV: + stp x29, x30, [sp, #-16]! + cfi_adjust_cfa_offset (16) + cfi_rel_offset (x29, 0) + cfi_rel_offset (x30, 8) + + mov x29, sp + + sub sp, sp, #ffi_closure_SYSV_FS + cfi_adjust_cfa_offset (ffi_closure_SYSV_FS) + + stp x21, x22, [x29, #-16] + cfi_rel_offset (x21, 0) + cfi_rel_offset (x22, 8) + + /* Load x21 with &call_context. */ + mov x21, sp + /* Preserve our struct trampoline_data * */ + mov x22, x17 + + /* Save the rest of the argument passing registers. */ + stp x0, x1, [x21, #0] + stp x2, x3, [x21, #16] + stp x4, x5, [x21, #32] + stp x6, x7, [x21, #48] + /* Don't forget we may have been given a result scratch pad address. + */ + str x8, [x21, #64] + + /* Figure out if we should touch the vector registers. */ + ldr x0, [x22, #8] + tbz x0, #AARCH64_FFI_WITH_V_BIT, 1f + + /* Save the argument passing vector registers. */ + stp q0, q1, [x21, #8*32 + 0] + stp q2, q3, [x21, #8*32 + 32] + stp q4, q5, [x21, #8*32 + 64] + stp q6, q7, [x21, #8*32 + 96] +1: + /* Load &ffi_closure.. */ + ldr x0, [x22, #0] + mov x1, x21 + /* Compute the location of the stack at the point that the + trampoline was called. */ + add x2, x29, #16 + + bl ffi_closure_SYSV_inner + + /* Figure out if we should touch the vector registers. */ + ldr x0, [x22, #8] + tbz x0, #AARCH64_FFI_WITH_V_BIT, 1f + + /* Load the result passing vector registers. */ + ldp q0, q1, [x21, #8*32 + 0] + ldp q2, q3, [x21, #8*32 + 32] + ldp q4, q5, [x21, #8*32 + 64] + ldp q6, q7, [x21, #8*32 + 96] +1: + /* Load the result passing core registers. */ + ldp x0, x1, [x21, #0] + ldp x2, x3, [x21, #16] + ldp x4, x5, [x21, #32] + ldp x6, x7, [x21, #48] + /* Note nothing usefull is returned in x8. */ + + /* We are done, unwind our frame. */ + ldp x21, x22, [x29, #-16] + cfi_restore (x21) + cfi_restore (x22) + + mov sp, x29 + cfi_adjust_cfa_offset (-ffi_closure_SYSV_FS) + + ldp x29, x30, [sp], #16 + cfi_adjust_cfa_offset (-16) + cfi_restore (x29) + cfi_restore (x30) + + ret + .cfi_endproc + .size ffi_closure_SYSV, .-ffi_closure_SYSV diff --git a/libffi/src/arm/ffi.c b/libffi/src/arm/ffi.c index 1f8597d..3ccceb9 100644 --- a/libffi/src/arm/ffi.c +++ b/libffi/src/arm/ffi.c @@ -251,8 +251,10 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue) break; case FFI_VFP: +#ifdef __ARM_EABI__ ffi_call_VFP (fn, &ecif, cif->bytes, cif->flags, ecif.rvalue); break; +#endif default: FFI_ASSERT(0); @@ -609,8 +611,10 @@ ffi_prep_closure_loc (ffi_closure* closure, if (cif->abi == FFI_SYSV) closure_func = &ffi_closure_SYSV; +#ifdef __ARM_EABI__ else if (cif->abi == FFI_VFP) closure_func = &ffi_closure_VFP; +#endif else return FFI_BAD_ABI; diff --git a/libffi/src/arm/sysv.S b/libffi/src/arm/sysv.S index 60e2ae3..fb38cd6 100644 --- a/libffi/src/arm/sysv.S +++ b/libffi/src/arm/sysv.S @@ -41,7 +41,7 @@ #define CNAME(x) x #endif #ifdef __APPLE__ -#define ENTRY(x) .globl CNAME(x); CNAME(x): +#define ENTRY(x) .globl _##x; _##x: #else #define ENTRY(x) .globl CNAME(x); .type CNAME(x),%function; CNAME(x): #endif /* __APPLE__ */ @@ -187,7 +187,7 @@ ARM_FUNC_START ffi_call_SYSV @ r1 already set @ Call ffi_prep_args(stack, &ecif) - bl ffi_prep_args + bl CNAME(ffi_prep_args) @ move first 4 parameters in registers ldmia sp, {r0-r3} @@ -334,7 +334,9 @@ ARM_FUNC_START ffi_closure_SYSV /* Below are VFP hard-float ABI call and closure implementations. - Add VFP FPU directive here. */ + Add VFP FPU directive here. This is only compiled into the library + under EABI. */ +#ifdef __ARM_EABI__ .fpu vfp @ r0: fn @@ -362,7 +364,7 @@ ARM_FUNC_START ffi_call_VFP sub r2, fp, #64 @ VFP scratch space @ Call ffi_prep_args(stack, &ecif, vfp_space) - bl ffi_prep_args + bl CNAME(ffi_prep_args) @ Load VFP register args if needed cmp r0, #0 @@ -444,7 +446,7 @@ ARM_FUNC_START ffi_closure_VFP sub sp, sp, #72 str sp, [sp, #64] add r1, sp, #64 - bl ffi_closure_SYSV_inner + bl CNAME(ffi_closure_SYSV_inner) cmp r0, #FFI_TYPE_INT beq .Lretint_vfp @@ -491,6 +493,7 @@ ARM_FUNC_START ffi_closure_VFP .ffi_closure_VFP_end: UNWIND .fnend .size CNAME(ffi_closure_VFP),.ffi_closure_VFP_end-CNAME(ffi_closure_VFP) +#endif ENTRY(ffi_arm_trampoline) stmfd sp!, {r0-r3} diff --git a/libffi/src/arm/trampoline.S b/libffi/src/arm/trampoline.S index 7b47429..935e8de 100644 --- a/libffi/src/arm/trampoline.S +++ b/libffi/src/arm/trampoline.S @@ -1,5 +1,5 @@ # GENERATED CODE - DO NOT EDIT -# This file was generated by ./gentramp.sh +# This file was generated by src/arm/gentramp.sh # Copyright (c) 2010, Plausible Labs Cooperative, Inc. # diff --git a/libffi/src/bfin/ffi.c b/libffi/src/bfin/ffi.c new file mode 100644 index 0000000..0beccc1 --- /dev/null +++ b/libffi/src/bfin/ffi.c @@ -0,0 +1,195 @@ +/* ----------------------------------------------------------------------- + ffi.c - Copyright (c) 2012 Alexandre K. I. de Mendonca <alexandre.keunecke@gmail.com> + + Blackfin Foreign Function Interface + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ +#include <ffi.h> +#include <ffi_common.h> + +#include <stdlib.h> +#include <stdio.h> + +/* Maximum number of GPRs available for argument passing. */ +#define MAX_GPRARGS 3 + +/* + * Return types + */ +#define FFIBFIN_RET_VOID 0 +#define FFIBFIN_RET_BYTE 1 +#define FFIBFIN_RET_HALFWORD 2 +#define FFIBFIN_RET_INT64 3 +#define FFIBFIN_RET_INT32 4 + +/*====================================================================*/ +/* PROTOTYPE * + /*====================================================================*/ +void ffi_prep_args(unsigned char *, extended_cif *); + +/*====================================================================*/ +/* Externals */ +/* (Assembly) */ +/*====================================================================*/ + +extern void ffi_call_SYSV(unsigned, extended_cif *, void(*)(unsigned char *, extended_cif *), unsigned, void *, void(*fn)(void)); + +/*====================================================================*/ +/* Implementation */ +/* */ +/*====================================================================*/ + + +/* + * This function calculates the return type (size) based on type. + */ + +ffi_status ffi_prep_cif_machdep(ffi_cif *cif) +{ + /* --------------------------------------* + * Return handling * + * --------------------------------------*/ + switch (cif->rtype->type) { + case FFI_TYPE_VOID: + cif->flags = FFIBFIN_RET_VOID; + break; + case FFI_TYPE_UINT16: + case FFI_TYPE_SINT16: + cif->flags = FFIBFIN_RET_HALFWORD; + break; + case FFI_TYPE_UINT8: + cif->flags = FFIBFIN_RET_BYTE; + break; + case FFI_TYPE_INT: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_FLOAT: + case FFI_TYPE_POINTER: + case FFI_TYPE_SINT8: + cif->flags = FFIBFIN_RET_INT32; + break; + case FFI_TYPE_SINT64: + case FFI_TYPE_UINT64: + case FFI_TYPE_DOUBLE: + cif->flags = FFIBFIN_RET_INT64; + break; + case FFI_TYPE_STRUCT: + if (cif->rtype->size <= 4){ + cif->flags = FFIBFIN_RET_INT32; + }else if (cif->rtype->size == 8){ + cif->flags = FFIBFIN_RET_INT64; + }else{ + //it will return via a hidden pointer in P0 + cif->flags = FFIBFIN_RET_VOID; + } + break; + default: + FFI_ASSERT(0); + break; + } + return FFI_OK; +} + +/* + * This will prepare the arguments and will call the assembly routine + * cif = the call interface + * fn = the function to be called + * rvalue = the return value + * avalue = the arguments + */ +void ffi_call(ffi_cif *cif, void(*fn)(void), void *rvalue, void **avalue) +{ + int ret_type = cif->flags; + extended_cif ecif; + ecif.cif = cif; + ecif.avalue = avalue; + ecif.rvalue = rvalue; + + switch (cif->abi) { + case FFI_SYSV: + ffi_call_SYSV(cif->bytes, &ecif, ffi_prep_args, ret_type, ecif.rvalue, fn); + break; + default: + FFI_ASSERT(0); + break; + } +} + + +/* +* This function prepares the parameters (copies them from the ecif to the stack) +* to call the function (ffi_prep_args is called by the assembly routine in file +* sysv.S, which also calls the actual function) +*/ +void ffi_prep_args(unsigned char *stack, extended_cif *ecif) +{ + register unsigned int i = 0; + void **p_argv; + unsigned char *argp; + ffi_type **p_arg; + argp = stack; + p_argv = ecif->avalue; + for (i = ecif->cif->nargs, p_arg = ecif->cif->arg_types; + (i != 0); + i--, p_arg++) { + size_t z; + z = (*p_arg)->size; + if (z < sizeof(int)) { + z = sizeof(int); + switch ((*p_arg)->type) { + case FFI_TYPE_SINT8: { + signed char v = *(SINT8 *)(* p_argv); + signed int t = v; + *(signed int *) argp = t; + } + break; + case FFI_TYPE_UINT8: { + unsigned char v = *(UINT8 *)(* p_argv); + unsigned int t = v; + *(unsigned int *) argp = t; + } + break; + case FFI_TYPE_SINT16: + *(signed int *) argp = (signed int) * (SINT16 *)(* p_argv); + break; + case FFI_TYPE_UINT16: + *(unsigned int *) argp = (unsigned int) * (UINT16 *)(* p_argv); + break; + case FFI_TYPE_STRUCT: + memcpy(argp, *p_argv, (*p_arg)->size); + break; + default: + FFI_ASSERT(0); + break; + } + } else if (z == sizeof(int)) { + *(unsigned int *) argp = (unsigned int) * (UINT32 *)(* p_argv); + } else { + memcpy(argp, *p_argv, z); + } + p_argv++; + argp += z; + } +} + + + diff --git a/libffi/src/bfin/ffitarget.h b/libffi/src/bfin/ffitarget.h new file mode 100644 index 0000000..2175c01 --- /dev/null +++ b/libffi/src/bfin/ffitarget.h @@ -0,0 +1,43 @@ +/* ----------------------------------------------------------------------- + ffitarget.h - Copyright (c) 2012 Alexandre K. I. de Mendonca <alexandre.keunecke@gmail.com> + + Blackfin Foreign Function Interface + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ + +#ifndef LIBFFI_TARGET_H +#define LIBFFI_TARGET_H + +#ifndef LIBFFI_ASM +typedef unsigned long ffi_arg; +typedef signed long ffi_sarg; + +typedef enum ffi_abi { + FFI_FIRST_ABI = 0, + FFI_SYSV, + FFI_LAST_ABI, + FFI_DEFAULT_ABI = FFI_SYSV +} ffi_abi; +#endif + +#endif + diff --git a/libffi/src/bfin/sysv.S b/libffi/src/bfin/sysv.S new file mode 100644 index 0000000..ae7a152 --- /dev/null +++ b/libffi/src/bfin/sysv.S @@ -0,0 +1,177 @@ +/* ----------------------------------------------------------------------- + sysv.S - Copyright (c) 2012 Alexandre K. I. de Mendonca <alexandre.keunecke@gmail.com> + + Blackfin Foreign Function Interface + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ + +#define LIBFFI_ASM +#include <fficonfig.h> +#include <ffi.h> + +.text +.align 4 + + /* + There is a "feature" in the bfin toolchain that it puts a _ before funcion names + that's why the function here it's called _ffi_call_SYSV and not ffi_call_SYSV + */ + .global _ffi_call_SYSV; + .type _ffi_call_SYSV, STT_FUNC; + .func ffi_call_SYSV + + /* + cif->bytes = R0 (fp+8) + &ecif = R1 (fp+12) + ffi_prep_args = R2 (fp+16) + ret_type = stack (fp+20) + ecif.rvalue = stack (fp+24) + fn = stack (fp+28) + got (fp+32) + There is room for improvement here (we can use temporary registers + instead of saving the values in the memory) + REGS: + P5 => Stack pointer (function arguments) + R5 => cif->bytes + R4 => ret->type + + FP-20 = P3 + FP-16 = SP (parameters area) + FP-12 = SP (temp) + FP-08 = function return part 1 [R0] + FP-04 = function return part 2 [R1] + */ + +_ffi_call_SYSV: +.prologue: + LINK 20; + [FP-20] = P3; + [FP+8] = R0; + [FP+12] = R1; + [FP+16] = R2; + +.allocate_stack: + //alocate cif->bytes into the stack + R1 = [FP+8]; + R0 = SP; + R0 = R0 - R1; + R1 = 4; + R0 = R0 - R1; + [FP-12] = SP; + SP = R0; + [FP-16] = SP; + +.call_prep_args: + //get the addr of prep_args + P0 = [P3 + _ffi_prep_args@FUNCDESC_GOT17M4]; + P1 = [P0]; + P3 = [P0+4]; + R0 = [FP-16];//SP (parameter area) + R1 = [FP+12];//ecif + call (P1); + +.call_user_function: + //ajust SP so as to allow the user function access the parameters on the stack + SP = [FP-16]; //point to function parameters + R0 = [SP]; + R1 = [SP+4]; + R2 = [SP+8]; + //load user function address + P0 = FP; + P0 +=28; + P1 = [P0]; + P1 = [P1]; + P3 = [P0+4]; + /* + For functions returning aggregate values (struct) occupying more than 8 bytes, + the caller allocates the return value object on the stack and the address + of this object is passed to the callee as a hidden argument in register P0. + */ + P0 = [FP+24]; + + call (P1); + SP = [FP-12]; +.compute_return: + P2 = [FP-20]; + [FP-8] = R0; + [FP-4] = R1; + + R0 = [FP+20]; + R1 = R0 << 2; + + R0 = [P2+.rettable@GOT17M4]; + R0 = R1 + R0; + P2 = R0; + R1 = [P2]; + + P2 = [FP+-20]; + R0 = [P2+.rettable@GOT17M4]; + R0 = R1 + R0; + P2 = R0; + R0 = [FP-8]; + R1 = [FP-4]; + jump (P2); + +/* +#define FFIBFIN_RET_VOID 0 +#define FFIBFIN_RET_BYTE 1 +#define FFIBFIN_RET_HALFWORD 2 +#define FFIBFIN_RET_INT64 3 +#define FFIBFIN_RET_INT32 4 +*/ +.align 4 +.align 4 +.rettable: + .dd .epilogue - .rettable + .dd .rbyte - .rettable; + .dd .rhalfword - .rettable; + .dd .rint64 - .rettable; + .dd .rint32 - .rettable; + +.rbyte: + P0 = [FP+24]; + R0 = R0.B (Z); + [P0] = R0; + JUMP .epilogue +.rhalfword: + P0 = [FP+24]; + R0 = R0.L; + [P0] = R0; + JUMP .epilogue +.rint64: + P0 = [FP+24];// &rvalue + [P0] = R0; + [P0+4] = R1; + JUMP .epilogue +.rint32: + P0 = [FP+24]; + [P0] = R0; +.epilogue: + R0 = [FP+8]; + R1 = [FP+12]; + R2 = [FP+16]; + P3 = [FP-20]; + UNLINK; + RTS; + +.size _ffi_call_SYSV,.-_ffi_call_SYSV; +.endfunc diff --git a/libffi/src/closures.c b/libffi/src/closures.c index 1b37827..fecbc4a 100644 --- a/libffi/src/closures.c +++ b/libffi/src/closures.c @@ -172,6 +172,27 @@ selinux_enabled_check (void) #endif /* !FFI_MMAP_EXEC_SELINUX */ +/* On PaX enable kernels that have MPROTECT enable we can't use PROT_EXEC. */ +#ifdef FFI_MMAP_EXEC_EMUTRAMP_PAX +#include <stdlib.h> + +static int emutramp_enabled = -1; + +static int +emutramp_enabled_check (void) +{ + if (getenv ("FFI_DISABLE_EMUTRAMP") == NULL) + return 1; + else + return 0; +} + +#define is_emutramp_enabled() (emutramp_enabled >= 0 ? emutramp_enabled \ + : (emutramp_enabled = emutramp_enabled_check ())) +#else +#define is_emutramp_enabled() 0 +#endif /* FFI_MMAP_EXEC_EMUTRAMP_PAX */ + #elif defined (__CYGWIN__) || defined(__INTERIX) #include <sys/mman.h> @@ -458,6 +479,12 @@ dlmmap (void *start, size_t length, int prot, printf ("mapping in %zi\n", length); #endif + if (execfd == -1 && is_emutramp_enabled ()) + { + ptr = mmap (start, length, prot & ~PROT_EXEC, flags, fd, offset); + return ptr; + } + if (execfd == -1 && !is_selinux_enabled ()) { ptr = mmap (start, length, prot | PROT_EXEC, flags, fd, offset); diff --git a/libffi/src/m68k/ffi.c b/libffi/src/m68k/ffi.c index d95c72b..37a0784 100644 --- a/libffi/src/m68k/ffi.c +++ b/libffi/src/m68k/ffi.c @@ -1,7 +1,7 @@ /* ----------------------------------------------------------------------- ffi.c - - m68k Foreign Function Interface + + m68k Foreign Function Interface ----------------------------------------------------------------------- */ #include <ffi.h> @@ -13,8 +13,13 @@ void rtems_cache_flush_multiple_data_lines( const void *, size_t ); #else #include <sys/syscall.h> +#ifdef __MINT__ +#include <mint/mintbind.h> +#include <mint/ssystem.h> +#else #include <asm/cachectl.h> #endif +#endif void ffi_call_SYSV (extended_cif *, unsigned, unsigned, @@ -39,8 +44,12 @@ ffi_prep_args (void *stack, extended_cif *ecif) argp = stack; - if (ecif->cif->rtype->type == FFI_TYPE_STRUCT - && !ecif->cif->flags) + if ( +#ifdef __MINT__ + (ecif->cif->rtype->type == FFI_TYPE_LONGDOUBLE) || +#endif + (((ecif->cif->rtype->type == FFI_TYPE_STRUCT) + && !ecif->cif->flags))) struct_value_ptr = ecif->rvalue; else struct_value_ptr = NULL; @@ -51,12 +60,12 @@ ffi_prep_args (void *stack, extended_cif *ecif) i != 0; i--, p_arg++) { - size_t z; + size_t z = (*p_arg)->size; + int type = (*p_arg)->type; - z = (*p_arg)->size; if (z < sizeof (int)) { - switch ((*p_arg)->type) + switch (type) { case FFI_TYPE_SINT8: *(signed int *) argp = (signed int) *(SINT8 *) *p_argv; @@ -75,7 +84,14 @@ ffi_prep_args (void *stack, extended_cif *ecif) break; case FFI_TYPE_STRUCT: +#ifdef __MINT__ + if (z == 1 || z == 2) + memcpy (argp + 2, *p_argv, z); + else + memcpy (argp, *p_argv, z); +#else memcpy (argp + sizeof (int) - z, *p_argv, z); +#endif break; default: @@ -120,17 +136,34 @@ ffi_prep_cif_machdep (ffi_cif *cif) break; case FFI_TYPE_STRUCT: + if (cif->rtype->elements[0]->type == FFI_TYPE_STRUCT && + cif->rtype->elements[1]) + { + cif->flags = 0; + break; + } + switch (cif->rtype->size) { case 1: +#ifdef __MINT__ + cif->flags = CIF_FLAGS_STRUCT2; +#else cif->flags = CIF_FLAGS_STRUCT1; +#endif break; case 2: cif->flags = CIF_FLAGS_STRUCT2; break; +#ifdef __MINT__ + case 3: +#endif case 4: cif->flags = CIF_FLAGS_INT; break; +#ifdef __MINT__ + case 7: +#endif case 8: cif->flags = CIF_FLAGS_DINT; break; @@ -150,7 +183,11 @@ ffi_prep_cif_machdep (ffi_cif *cif) #if (FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE) case FFI_TYPE_LONGDOUBLE: +#ifdef __MINT__ + cif->flags = 0; +#else cif->flags = CIF_FLAGS_LDOUBLE; +#endif break; #endif @@ -218,6 +255,26 @@ ffi_prep_incoming_args_SYSV (char *stack, void **avalue, ffi_cif *cif) size_t z; z = (*p_arg)->size; +#ifdef __MINT__ + if (cif->flags && + cif->rtype->type == FFI_TYPE_STRUCT && + (z == 1 || z == 2)) + { + *p_argv = (void *) (argp + 2); + + z = 4; + } + else + if (cif->flags && + cif->rtype->type == FFI_TYPE_STRUCT && + (z == 3 || z == 4)) + { + *p_argv = (void *) (argp); + + z = 4; + } + else +#endif if (z <= 4) { *p_argv = (void *) (argp + 4 - z); @@ -267,14 +324,21 @@ ffi_prep_closure_loc (ffi_closure* closure, *(unsigned short *)closure->tramp = 0x207c; *(void **)(closure->tramp + 2) = codeloc; *(unsigned short *)(closure->tramp + 6) = 0x4ef9; - if (cif->rtype->type == FFI_TYPE_STRUCT - && !cif->flags) + + if ( +#ifdef __MINT__ + (cif->rtype->type == FFI_TYPE_LONGDOUBLE) || +#endif + (((cif->rtype->type == FFI_TYPE_STRUCT) + && !cif->flags))) *(void **)(closure->tramp + 8) = ffi_closure_struct_SYSV; else *(void **)(closure->tramp + 8) = ffi_closure_SYSV; #ifdef __rtems__ rtems_cache_flush_multiple_data_lines( codeloc, FFI_TRAMPOLINE_SIZE ); +#elif defined(__MINT__) + Ssystem(S_FLUSHCACHE, codeloc, FFI_TRAMPOLINE_SIZE); #else syscall(SYS_cacheflush, codeloc, FLUSH_SCOPE_LINE, FLUSH_CACHE_BOTH, FFI_TRAMPOLINE_SIZE); @@ -286,4 +350,3 @@ ffi_prep_closure_loc (ffi_closure* closure, return FFI_OK; } - diff --git a/libffi/src/m68k/sysv.S b/libffi/src/m68k/sysv.S index dfdd864..f6f4ef9 100644 --- a/libffi/src/m68k/sysv.S +++ b/libffi/src/m68k/sysv.S @@ -1,6 +1,7 @@ /* ----------------------------------------------------------------------- - sysv.S - Copyright (c) 1998, 2012 Andreas Schwab + sysv.S - Copyright (c) 2012 Alan Hourihane + Copyright (c) 1998, 2012 Andreas Schwab Copyright (c) 2008 Red Hat, Inc. m68k Foreign Function Interface @@ -42,13 +43,19 @@ #define CFI_ENDPROC() #endif +#ifdef __MINT__ +#define CALLFUNC(funcname) _ ## funcname +#else +#define CALLFUNC(funcname) funcname +#endif + .text - .globl ffi_call_SYSV - .type ffi_call_SYSV,@function + .globl CALLFUNC(ffi_call_SYSV) + .type CALLFUNC(ffi_call_SYSV),@function .align 4 -ffi_call_SYSV: +CALLFUNC(ffi_call_SYSV): CFI_STARTPROC() link %fp,#0 CFI_OFFSET(14,-8) @@ -63,14 +70,18 @@ ffi_call_SYSV: move.l 8(%fp),-(%sp) pea 4(%sp) #if !defined __PIC__ - jsr ffi_prep_args + jsr CALLFUNC(ffi_prep_args) #else - bsr.l ffi_prep_args@PLTPC + bsr.l CALLFUNC(ffi_prep_args@PLTPC) #endif addq.l #8,%sp | Pass pointer to struct value, if any +#ifdef __MINT__ + move.l %d0,%a1 +#else move.l %a0,%a1 +#endif | Call the function move.l 24(%fp),%a0 @@ -142,7 +153,11 @@ retlongdouble: retpointer: btst #5,%d2 jbeq retstruct1 +#ifdef __MINT__ + move.l %d0,(%a1) +#else move.l %a0,(%a1) +#endif jbra epilogue retstruct1: @@ -162,13 +177,13 @@ epilogue: unlk %fp rts CFI_ENDPROC() - .size ffi_call_SYSV,.-ffi_call_SYSV + .size CALLFUNC(ffi_call_SYSV),.-CALLFUNC(ffi_call_SYSV) - .globl ffi_closure_SYSV - .type ffi_closure_SYSV, @function + .globl CALLFUNC(ffi_closure_SYSV) + .type CALLFUNC(ffi_closure_SYSV), @function .align 4 -ffi_closure_SYSV: +CALLFUNC(ffi_closure_SYSV): CFI_STARTPROC() link %fp,#-12 CFI_OFFSET(14,-8) @@ -178,9 +193,9 @@ ffi_closure_SYSV: pea -12(%fp) move.l %a0,-(%sp) #if !defined __PIC__ - jsr ffi_closure_SYSV_inner + jsr CALLFUNC(ffi_closure_SYSV_inner) #else - bsr.l ffi_closure_SYSV_inner@PLTPC + bsr.l CALLFUNC(ffi_closure_SYSV_inner@PLTPC) #endif lsr.l #1,%d0 @@ -240,13 +255,13 @@ ffi_closure_SYSV: jra .Lcls_epilogue CFI_ENDPROC() - .size ffi_closure_SYSV,.-ffi_closure_SYSV + .size CALLFUNC(ffi_closure_SYSV),.-CALLFUNC(ffi_closure_SYSV) - .globl ffi_closure_struct_SYSV - .type ffi_closure_struct_SYSV, @function + .globl CALLFUNC(ffi_closure_struct_SYSV) + .type CALLFUNC(ffi_closure_struct_SYSV), @function .align 4 -ffi_closure_struct_SYSV: +CALLFUNC(ffi_closure_struct_SYSV): CFI_STARTPROC() link %fp,#0 CFI_OFFSET(14,-8) @@ -256,14 +271,14 @@ ffi_closure_struct_SYSV: move.l %a1,-(%sp) move.l %a0,-(%sp) #if !defined __PIC__ - jsr ffi_closure_SYSV_inner + jsr CALLFUNC(ffi_closure_SYSV_inner) #else - bsr.l ffi_closure_SYSV_inner@PLTPC + bsr.l CALLFUNC(ffi_closure_SYSV_inner@PLTPC) #endif unlk %fp rts CFI_ENDPROC() - .size ffi_closure_struct_SYSV,.-ffi_closure_struct_SYSV + .size CALLFUNC(ffi_closure_struct_SYSV),.-CALLFUNC(ffi_closure_struct_SYSV) #if defined __ELF__ && defined __linux__ .section .note.GNU-stack,"",@progbits diff --git a/libffi/src/powerpc/ffi_darwin.c b/libffi/src/powerpc/ffi_darwin.c index ee03dab..dd897f4 100644 --- a/libffi/src/powerpc/ffi_darwin.c +++ b/libffi/src/powerpc/ffi_darwin.c @@ -1065,10 +1065,10 @@ ffi_prep_closure_loc (ffi_closure* closure, closure->cif = cif; closure->fun = fun; closure->user_data = user_data; + break; default: - - FFI_ASSERT(0); + return FFI_BAD_ABI; break; } return FFI_OK; @@ -1235,7 +1235,7 @@ ffi_closure_helper_DARWIN (ffi_closure *closure, void *rvalue, if (arg_types[i]->elements[0]->type == FFI_TYPE_DOUBLE) size_al = ALIGN(arg_types[i]->size, 8); # if defined(POWERPC64) - FFI_ASSERT (cif->abi != FFI_DARWIN) + FFI_ASSERT (cif->abi != FFI_DARWIN); avalue[i] = pgr; pgr += (size_al + 7) / 8; # else diff --git a/libffi/src/prep_cif.c b/libffi/src/prep_cif.c index eb68341..5d1924b 100644 --- a/libffi/src/prep_cif.c +++ b/libffi/src/prep_cif.c @@ -140,6 +140,9 @@ ffi_status FFI_HIDDEN ffi_prep_cif_core(ffi_cif *cif, ffi_abi abi, #ifdef SPARC && (cif->abi != FFI_V9 || cif->rtype->size > 32) #endif +#ifdef TILE + && (cif->rtype->size > 10 * FFI_SIZEOF_ARG) +#endif ) bytes = STACK_ARG_SIZE(sizeof(void*)); #endif @@ -169,6 +172,16 @@ ffi_status FFI_HIDDEN ffi_prep_cif_core(ffi_cif *cif, ffi_abi abi, if (((*ptr)->alignment - 1) & bytes) bytes = ALIGN(bytes, (*ptr)->alignment); +#ifdef TILE + if (bytes < 10 * FFI_SIZEOF_ARG && + bytes + STACK_ARG_SIZE((*ptr)->size) > 10 * FFI_SIZEOF_ARG) + { + /* An argument is never split between the 10 parameter + registers and the stack. */ + bytes = 10 * FFI_SIZEOF_ARG; + } +#endif + bytes += STACK_ARG_SIZE((*ptr)->size); } #endif diff --git a/libffi/src/tile/ffi.c b/libffi/src/tile/ffi.c new file mode 100644 index 0000000..3a94469 --- /dev/null +++ b/libffi/src/tile/ffi.c @@ -0,0 +1,355 @@ +/* ----------------------------------------------------------------------- + ffi.c - Copyright (c) 2012 Tilera Corp. + + TILE Foreign Function Interface + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ + +#include <ffi.h> +#include <ffi_common.h> +#include <stdlib.h> +#include <stdint.h> +#include <unistd.h> +#include <arch/abi.h> +#include <arch/icache.h> +#include <arch/opcode.h> + + +/* The first 10 registers are used to pass arguments and return values. */ +#define NUM_ARG_REGS 10 + +/* Performs a raw function call with the given NUM_ARG_REGS register arguments + and the specified additional stack arguments (if any). */ +extern void ffi_call_tile(ffi_sarg reg_args[NUM_ARG_REGS], + const ffi_sarg *stack_args, + size_t stack_args_bytes, + void (*fnaddr)(void)) + FFI_HIDDEN; + +/* This handles the raw call from the closure stub, cleaning up the + parameters and delegating to ffi_closure_tile_inner. */ +extern void ffi_closure_tile(void) FFI_HIDDEN; + + +ffi_status +ffi_prep_cif_machdep(ffi_cif *cif) +{ + /* We always allocate room for all registers. Even if we don't + use them as parameters, they get returned in the same array + as struct return values so we need to make room. */ + if (cif->bytes < NUM_ARG_REGS * FFI_SIZEOF_ARG) + cif->bytes = NUM_ARG_REGS * FFI_SIZEOF_ARG; + + if (cif->rtype->size > NUM_ARG_REGS * FFI_SIZEOF_ARG) + cif->flags = FFI_TYPE_STRUCT; + else + cif->flags = FFI_TYPE_INT; + + /* Nothing to do. */ + return FFI_OK; +} + + +static long +assign_to_ffi_arg(ffi_sarg *out, void *in, const ffi_type *type, + int write_to_reg) +{ + switch (type->type) + { + case FFI_TYPE_SINT8: + *out = *(SINT8 *)in; + return 1; + + case FFI_TYPE_UINT8: + *out = *(UINT8 *)in; + return 1; + + case FFI_TYPE_SINT16: + *out = *(SINT16 *)in; + return 1; + + case FFI_TYPE_UINT16: + *out = *(UINT16 *)in; + return 1; + + case FFI_TYPE_SINT32: + case FFI_TYPE_UINT32: +#ifndef __LP64__ + case FFI_TYPE_POINTER: +#endif + /* Note that even unsigned 32-bit quantities are sign extended + on tilegx when stored in a register. */ + *out = *(SINT32 *)in; + return 1; + + case FFI_TYPE_FLOAT: +#ifdef __tilegx__ + if (write_to_reg) + { + /* Properly sign extend the value. */ + union { float f; SINT32 s32; } val; + val.f = *(float *)in; + *out = val.s32; + } + else +#endif + { + *(float *)out = *(float *)in; + } + return 1; + + case FFI_TYPE_SINT64: + case FFI_TYPE_UINT64: + case FFI_TYPE_DOUBLE: +#ifdef __LP64__ + case FFI_TYPE_POINTER: +#endif + *(UINT64 *)out = *(UINT64 *)in; + return sizeof(UINT64) / FFI_SIZEOF_ARG; + + case FFI_TYPE_STRUCT: + memcpy(out, in, type->size); + return (type->size + FFI_SIZEOF_ARG - 1) / FFI_SIZEOF_ARG; + + case FFI_TYPE_VOID: + /* Must be a return type. Nothing to do. */ + return 0; + + default: + FFI_ASSERT(0); + return -1; + } +} + + +void +ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue) +{ + ffi_sarg * const arg_mem = alloca(cif->bytes); + ffi_sarg * const reg_args = arg_mem; + ffi_sarg * const stack_args = ®_args[NUM_ARG_REGS]; + ffi_sarg *argp = arg_mem; + ffi_type ** const arg_types = cif->arg_types; + const long num_args = cif->nargs; + long i; + + if (cif->flags == FFI_TYPE_STRUCT) + { + /* Pass a hidden pointer to the return value. We make sure there + is scratch space for the callee to store the return value even if + our caller doesn't care about it. */ + *argp++ = (intptr_t)(rvalue ? rvalue : alloca(cif->rtype->size)); + + /* No more work needed to return anything. */ + rvalue = NULL; + } + + for (i = 0; i < num_args; i++) + { + ffi_type *type = arg_types[i]; + void * const arg_in = avalue[i]; + ptrdiff_t arg_word = argp - arg_mem; + +#ifndef __tilegx__ + /* Doubleword-aligned values are always in an even-number register + pair, or doubleword-aligned stack slot if out of registers. */ + long align = arg_word & (type->alignment > FFI_SIZEOF_ARG); + argp += align; + arg_word += align; +#endif + + if (type->type == FFI_TYPE_STRUCT) + { + const size_t arg_size_in_words = + (type->size + FFI_SIZEOF_ARG - 1) / FFI_SIZEOF_ARG; + + if (arg_word < NUM_ARG_REGS && + arg_word + arg_size_in_words > NUM_ARG_REGS) + { + /* Args are not allowed to span registers and the stack. */ + argp = stack_args; + } + + memcpy(argp, arg_in, type->size); + argp += arg_size_in_words; + } + else + { + argp += assign_to_ffi_arg(argp, arg_in, arg_types[i], 1); + } + } + + /* Actually do the call. */ + ffi_call_tile(reg_args, stack_args, + cif->bytes - (NUM_ARG_REGS * FFI_SIZEOF_ARG), fn); + + if (rvalue != NULL) + assign_to_ffi_arg(rvalue, reg_args, cif->rtype, 0); +} + + +/* Template code for closure. */ +extern const UINT64 ffi_template_tramp_tile[] FFI_HIDDEN; + + +ffi_status +ffi_prep_closure_loc (ffi_closure *closure, + ffi_cif *cif, + void (*fun)(ffi_cif*, void*, void**, void*), + void *user_data, + void *codeloc) +{ +#ifdef __tilegx__ + /* TILE-Gx */ + SINT64 c; + SINT64 h; + int s; + UINT64 *out; + + if (cif->abi != FFI_UNIX) + return FFI_BAD_ABI; + + out = (UINT64 *)closure->tramp; + + c = (intptr_t)closure; + h = (intptr_t)ffi_closure_tile; + s = 0; + + /* Find the smallest shift count that doesn't lose information + (i.e. no need to explicitly insert high bits of the address that + are just the sign extension of the low bits). */ + while ((c >> s) != (SINT16)(c >> s) || (h >> s) != (SINT16)(h >> s)) + s += 16; + +#define OPS(a, b, shift) \ + (create_Imm16_X0((a) >> (shift)) | create_Imm16_X1((b) >> (shift))) + + /* Emit the moveli. */ + *out++ = ffi_template_tramp_tile[0] | OPS(c, h, s); + for (s -= 16; s >= 0; s -= 16) + *out++ = ffi_template_tramp_tile[1] | OPS(c, h, s); + +#undef OPS + + *out++ = ffi_template_tramp_tile[2]; + +#else + /* TILEPro */ + UINT64 *out; + intptr_t delta; + + if (cif->abi != FFI_UNIX) + return FFI_BAD_ABI; + + out = (UINT64 *)closure->tramp; + delta = (intptr_t)ffi_closure_tile - (intptr_t)codeloc; + + *out++ = ffi_template_tramp_tile[0] | create_JOffLong_X1(delta >> 3); +#endif + + closure->cif = cif; + closure->fun = fun; + closure->user_data = user_data; + + invalidate_icache(closure->tramp, (char *)out - closure->tramp, + getpagesize()); + + return FFI_OK; +} + + +/* This is called by the assembly wrapper for closures. This does + all of the work. On entry reg_args[0] holds the values the registers + had when the closure was invoked. On return reg_args[1] holds the register + values to be returned to the caller (many of which may be garbage). */ +void FFI_HIDDEN +ffi_closure_tile_inner(ffi_closure *closure, + ffi_sarg reg_args[2][NUM_ARG_REGS], + ffi_sarg *stack_args) +{ + ffi_cif * const cif = closure->cif; + void ** const avalue = alloca(cif->nargs * sizeof(void *)); + void *rvalue; + ffi_type ** const arg_types = cif->arg_types; + ffi_sarg * const reg_args_in = reg_args[0]; + ffi_sarg * const reg_args_out = reg_args[1]; + ffi_sarg * argp; + long i, arg_word, nargs = cif->nargs; + /* Use a union to guarantee proper alignment for double. */ + union { ffi_sarg arg[NUM_ARG_REGS]; double d; UINT64 u64; } closure_ret; + + /* Start out reading register arguments. */ + argp = reg_args_in; + + /* Copy the caller's structure return address to that the closure + returns the data directly to the caller. */ + if (cif->flags == FFI_TYPE_STRUCT) + { + /* Return by reference via hidden pointer. */ + rvalue = (void *)(intptr_t)*argp++; + arg_word = 1; + } + else + { + /* Return the value in registers. */ + rvalue = &closure_ret; + arg_word = 0; + } + + /* Grab the addresses of the arguments. */ + for (i = 0; i < nargs; i++) + { + ffi_type * const type = arg_types[i]; + const size_t arg_size_in_words = + (type->size + FFI_SIZEOF_ARG - 1) / FFI_SIZEOF_ARG; + +#ifndef __tilegx__ + /* Doubleword-aligned values are always in an even-number register + pair, or doubleword-aligned stack slot if out of registers. */ + long align = arg_word & (type->alignment > FFI_SIZEOF_ARG); + argp += align; + arg_word += align; +#endif + + if (arg_word == NUM_ARG_REGS || + (arg_word < NUM_ARG_REGS && + arg_word + arg_size_in_words > NUM_ARG_REGS)) + { + /* Switch to reading arguments from the stack. */ + argp = stack_args; + arg_word = NUM_ARG_REGS; + } + + avalue[i] = argp; + argp += arg_size_in_words; + arg_word += arg_size_in_words; + } + + /* Invoke the closure. */ + closure->fun(cif, rvalue, avalue, closure->user_data); + + if (cif->flags != FFI_TYPE_STRUCT) + { + /* Canonicalize for register representation. */ + assign_to_ffi_arg(reg_args_out, &closure_ret, cif->rtype, 1); + } +} diff --git a/libffi/src/tile/ffitarget.h b/libffi/src/tile/ffitarget.h new file mode 100644 index 0000000..679fb5d --- /dev/null +++ b/libffi/src/tile/ffitarget.h @@ -0,0 +1,65 @@ +/* -----------------------------------------------------------------*-C-*- + ffitarget.h - Copyright (c) 2012 Tilera Corp. + Target configuration macros for TILE. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ + +#ifndef LIBFFI_TARGET_H +#define LIBFFI_TARGET_H + +#ifndef LIBFFI_H +#error "Please do not include ffitarget.h directly into your source. Use ffi.h instead." +#endif + +#ifndef LIBFFI_ASM + +#include <arch/abi.h> + +typedef uint_reg_t ffi_arg; +typedef int_reg_t ffi_sarg; + +typedef enum ffi_abi { + FFI_FIRST_ABI = 0, + FFI_UNIX, + FFI_LAST_ABI, + FFI_DEFAULT_ABI = FFI_UNIX +} ffi_abi; +#endif + +/* ---- Definitions for closures ----------------------------------------- */ +#define FFI_CLOSURES 1 + +#ifdef __tilegx__ +/* We always pass 8-byte values, even in -m32 mode. */ +# define FFI_SIZEOF_ARG 8 +# ifdef __LP64__ +# define FFI_TRAMPOLINE_SIZE (8 * 5) /* 5 bundles */ +# else +# define FFI_TRAMPOLINE_SIZE (8 * 3) /* 3 bundles */ +# endif +#else +# define FFI_SIZEOF_ARG 4 +# define FFI_TRAMPOLINE_SIZE 8 /* 1 bundle */ +#endif +#define FFI_NATIVE_RAW_API 0 + +#endif diff --git a/libffi/src/tile/tile.S b/libffi/src/tile/tile.S new file mode 100644 index 0000000..a186e1f --- /dev/null +++ b/libffi/src/tile/tile.S @@ -0,0 +1,360 @@ +/* ----------------------------------------------------------------------- + tile.S - Copyright (c) 2011 Tilera Corp. + + Tilera TILEPro and TILE-Gx Foreign Function Interface + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + ``Software''), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice shall be included + in all copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT + HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, + WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + DEALINGS IN THE SOFTWARE. + ----------------------------------------------------------------------- */ + +#define LIBFFI_ASM +#include <fficonfig.h> +#include <ffi.h> + +/* Number of bytes in a register. */ +#define REG_SIZE FFI_SIZEOF_ARG + +/* Number of bytes in stack linkage area for backtracing. + + A note about the ABI: on entry to a procedure, sp points to a stack + slot where it must spill the return address if it's not a leaf. + REG_SIZE bytes beyond that is a slot owned by the caller which + contains the sp value that the caller had when it was originally + entered (i.e. the caller's frame pointer). */ +#define LINKAGE_SIZE (2 * REG_SIZE) + +/* The first 10 registers are used to pass arguments and return values. */ +#define NUM_ARG_REGS 10 + +#ifdef __tilegx__ +#define SW st +#define LW ld +#define BGZT bgtzt +#else +#define SW sw +#define LW lw +#define BGZT bgzt +#endif + + +/* void ffi_call_tile (int_reg_t reg_args[NUM_ARG_REGS], + const int_reg_t *stack_args, + unsigned long stack_args_bytes, + void (*fnaddr)(void)); + + On entry, REG_ARGS contain the outgoing register values, + and STACK_ARGS containts STACK_ARG_BYTES of additional values + to be passed on the stack. If STACK_ARG_BYTES is zero, then + STACK_ARGS is ignored. + + When the invoked function returns, the values of r0-r9 are + blindly stored back into REG_ARGS for the caller to examine. */ + + .section .text.ffi_call_tile, "ax", @progbits + .align 8 + .globl ffi_call_tile + FFI_HIDDEN(ffi_call_tile) +ffi_call_tile: + +/* Incoming arguments. */ +#define REG_ARGS r0 +#define INCOMING_STACK_ARGS r1 +#define STACK_ARG_BYTES r2 +#define ORIG_FNADDR r3 + +/* Temporary values. */ +#define FRAME_SIZE r10 +#define TMP r11 +#define TMP2 r12 +#define OUTGOING_STACK_ARGS r13 +#define REG_ADDR_PTR r14 +#define RETURN_REG_ADDR r15 +#define FNADDR r16 + + .cfi_startproc + { + /* Save return address. */ + SW sp, lr + .cfi_offset lr, 0 + /* Prepare to spill incoming r52. */ + addi TMP, sp, -REG_SIZE + /* Increase frame size to have room to spill r52 and REG_ARGS. + The +7 is to round up mod 8. */ + addi FRAME_SIZE, STACK_ARG_BYTES, \ + REG_SIZE + REG_SIZE + LINKAGE_SIZE + 7 + } + { + /* Round stack frame size to a multiple of 8 to satisfy ABI. */ + andi FRAME_SIZE, FRAME_SIZE, -8 + /* Compute where to spill REG_ARGS value. */ + addi TMP2, sp, -(REG_SIZE * 2) + } + { + /* Spill incoming r52. */ + SW TMP, r52 + .cfi_offset r52, -REG_SIZE + /* Set up our frame pointer. */ + move r52, sp + .cfi_def_cfa_register r52 + /* Push stack frame. */ + sub sp, sp, FRAME_SIZE + } + { + /* Prepare to set up stack linkage. */ + addi TMP, sp, REG_SIZE + /* Prepare to memcpy stack args. */ + addi OUTGOING_STACK_ARGS, sp, LINKAGE_SIZE + /* Save REG_ARGS which we will need after we call the subroutine. */ + SW TMP2, REG_ARGS + } + { + /* Set up linkage info to hold incoming stack pointer. */ + SW TMP, r52 + } + { + /* Skip stack args memcpy if we don't have any stack args (common). */ + blezt STACK_ARG_BYTES, .Ldone_stack_args_memcpy + } + +.Lmemcpy_stack_args: + { + /* Load incoming argument from stack_args. */ + LW TMP, INCOMING_STACK_ARGS + addi INCOMING_STACK_ARGS, INCOMING_STACK_ARGS, REG_SIZE + } + { + /* Store stack argument into outgoing stack argument area. */ + SW OUTGOING_STACK_ARGS, TMP + addi OUTGOING_STACK_ARGS, OUTGOING_STACK_ARGS, REG_SIZE + addi STACK_ARG_BYTES, STACK_ARG_BYTES, -REG_SIZE + } + { + BGZT STACK_ARG_BYTES, .Lmemcpy_stack_args + } +.Ldone_stack_args_memcpy: + + { + /* Copy aside ORIG_FNADDR so we can overwrite its register. */ + move FNADDR, ORIG_FNADDR + /* Prepare to load argument registers. */ + addi REG_ADDR_PTR, r0, REG_SIZE + /* Load outgoing r0. */ + LW r0, r0 + } + + /* Load up argument registers from the REG_ARGS array. */ +#define LOAD_REG(REG, PTR) \ + { \ + LW REG, PTR ; \ + addi PTR, PTR, REG_SIZE \ + } + + LOAD_REG(r1, REG_ADDR_PTR) + LOAD_REG(r2, REG_ADDR_PTR) + LOAD_REG(r3, REG_ADDR_PTR) + LOAD_REG(r4, REG_ADDR_PTR) + LOAD_REG(r5, REG_ADDR_PTR) + LOAD_REG(r6, REG_ADDR_PTR) + LOAD_REG(r7, REG_ADDR_PTR) + LOAD_REG(r8, REG_ADDR_PTR) + LOAD_REG(r9, REG_ADDR_PTR) + + { + /* Call the subroutine. */ + jalr FNADDR + } + + { + /* Restore original lr. */ + LW lr, r52 + /* Prepare to recover ARGS, which we spilled earlier. */ + addi TMP, r52, -(2 * REG_SIZE) + } + { + /* Restore ARGS, so we can fill it in with the return regs r0-r9. */ + LW RETURN_REG_ADDR, TMP + /* Prepare to restore original r52. */ + addi TMP, r52, -REG_SIZE + } + + { + /* Pop stack frame. */ + move sp, r52 + /* Restore original r52. */ + LW r52, TMP + } + +#define STORE_REG(REG, PTR) \ + { \ + SW PTR, REG ; \ + addi PTR, PTR, REG_SIZE \ + } + + /* Return all register values by reference. */ + STORE_REG(r0, RETURN_REG_ADDR) + STORE_REG(r1, RETURN_REG_ADDR) + STORE_REG(r2, RETURN_REG_ADDR) + STORE_REG(r3, RETURN_REG_ADDR) + STORE_REG(r4, RETURN_REG_ADDR) + STORE_REG(r5, RETURN_REG_ADDR) + STORE_REG(r6, RETURN_REG_ADDR) + STORE_REG(r7, RETURN_REG_ADDR) + STORE_REG(r8, RETURN_REG_ADDR) + STORE_REG(r9, RETURN_REG_ADDR) + + { + jrp lr + } + + .cfi_endproc + .size ffi_call_tile, .-ffi_call_tile + +/* ffi_closure_tile(...) + + On entry, lr points to the closure plus 8 bytes, and r10 + contains the actual return address. + + This function simply dumps all register parameters into a stack array + and passes the closure, the registers array, and the stack arguments + to C code that does all of the actual closure processing. */ + + .section .text.ffi_closure_tile, "ax", @progbits + .align 8 + .globl ffi_closure_tile + FFI_HIDDEN(ffi_closure_tile) + + .cfi_startproc +/* Room to spill all NUM_ARG_REGS incoming registers, plus frame linkage. */ +#define CLOSURE_FRAME_SIZE (((NUM_ARG_REGS * REG_SIZE * 2 + LINKAGE_SIZE) + 7) & -8) +ffi_closure_tile: + { +#ifdef __tilegx__ + st sp, lr + .cfi_offset lr, 0 +#else + /* Save return address (in r10 due to closure stub wrapper). */ + SW sp, r10 + .cfi_return_column r10 + .cfi_offset r10, 0 +#endif + /* Compute address for stack frame linkage. */ + addli r10, sp, -(CLOSURE_FRAME_SIZE - REG_SIZE) + } + { + /* Save incoming stack pointer in linkage area. */ + SW r10, sp + .cfi_offset sp, -(CLOSURE_FRAME_SIZE - REG_SIZE) + /* Push a new stack frame. */ + addli sp, sp, -CLOSURE_FRAME_SIZE + .cfi_adjust_cfa_offset CLOSURE_FRAME_SIZE + } + + { + /* Create pointer to where to start spilling registers. */ + addi r10, sp, LINKAGE_SIZE + } + + /* Spill all the incoming registers. */ + STORE_REG(r0, r10) + STORE_REG(r1, r10) + STORE_REG(r2, r10) + STORE_REG(r3, r10) + STORE_REG(r4, r10) + STORE_REG(r5, r10) + STORE_REG(r6, r10) + STORE_REG(r7, r10) + STORE_REG(r8, r10) + { + /* Save r9. */ + SW r10, r9 +#ifdef __tilegx__ + /* Pointer to closure is passed in r11. */ + move r0, r11 +#else + /* Compute pointer to the closure object. Because the closure + starts with a "jal ffi_closure_tile", we can just take the + value of lr (a phony return address pointing into the closure) + and subtract 8. */ + addi r0, lr, -8 +#endif + /* Compute a pointer to the register arguments we just spilled. */ + addi r1, sp, LINKAGE_SIZE + } + { + /* Compute a pointer to the extra stack arguments (if any). */ + addli r2, sp, CLOSURE_FRAME_SIZE + LINKAGE_SIZE + /* Call C code to deal with all of the grotty details. */ + jal ffi_closure_tile_inner + } + { + addli r10, sp, CLOSURE_FRAME_SIZE + } + { + /* Restore the return address. */ + LW lr, r10 + /* Compute pointer to registers array. */ + addli r10, sp, LINKAGE_SIZE + (NUM_ARG_REGS * REG_SIZE) + } + /* Return all the register values, which C code may have set. */ + LOAD_REG(r0, r10) + LOAD_REG(r1, r10) + LOAD_REG(r2, r10) + LOAD_REG(r3, r10) + LOAD_REG(r4, r10) + LOAD_REG(r5, r10) + LOAD_REG(r6, r10) + LOAD_REG(r7, r10) + LOAD_REG(r8, r10) + LOAD_REG(r9, r10) + { + /* Pop the frame. */ + addli sp, sp, CLOSURE_FRAME_SIZE + jrp lr + } + + .cfi_endproc + .size ffi_closure_tile, . - ffi_closure_tile + + +/* What follows are code template instructions that get copied to the + closure trampoline by ffi_prep_closure_loc. The zeroed operands + get replaced by their proper values at runtime. */ + + .section .text.ffi_template_tramp_tile, "ax", @progbits + .align 8 + .globl ffi_template_tramp_tile + FFI_HIDDEN(ffi_template_tramp_tile) +ffi_template_tramp_tile: +#ifdef __tilegx__ + { + moveli r11, 0 /* backpatched to address of containing closure. */ + moveli r10, 0 /* backpatched to ffi_closure_tile. */ + } + /* Note: the following bundle gets generated multiple times + depending on the pointer value (esp. useful for -m32 mode). */ + { shl16insli r11, r11, 0 ; shl16insli r10, r10, 0 } + { info 2+8 /* for backtracer: -> pc in lr, frame size 0 */ ; jr r10 } +#else + /* 'jal .' yields a PC-relative offset of zero so we can OR in the + right offset at runtime. */ + { move r10, lr ; jal . /* ffi_closure_tile */ } +#endif + + .size ffi_template_tramp_tile, . - ffi_template_tramp_tile diff --git a/libffi/src/x86/ffi.c b/libffi/src/x86/ffi.c index f643b34..611e221 100644 --- a/libffi/src/x86/ffi.c +++ b/libffi/src/x86/ffi.c @@ -58,7 +58,8 @@ void ffi_prep_args(char *stack, extended_cif *ecif) argp = stack; - if (ecif->cif->flags == FFI_TYPE_STRUCT + if ((ecif->cif->flags == FFI_TYPE_STRUCT + || ecif->cif->flags == FFI_TYPE_MS_STRUCT) #ifdef X86_WIN64 && (ecif->cif->rtype->size != 1 && ecif->cif->rtype->size != 2 && ecif->cif->rtype->size != 4 && ecif->cif->rtype->size != 8) @@ -279,7 +280,12 @@ ffi_status ffi_prep_cif_machdep(ffi_cif *cif) else #endif { - cif->flags = FFI_TYPE_STRUCT; +#ifdef X86_WIN32 + if (cif->abi == FFI_MS_CDECL) + cif->flags = FFI_TYPE_MS_STRUCT; + else +#endif + cif->flags = FFI_TYPE_STRUCT; /* allocate space for return value pointer */ cif->bytes += ALIGN(sizeof(void*), FFI_SIZEOF_ARG); } @@ -349,7 +355,8 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue) } #else if (rvalue == NULL - && cif->flags == FFI_TYPE_STRUCT) + && (cif->flags == FFI_TYPE_STRUCT + || cif->flags == FFI_TYPE_MS_STRUCT)) { ecif.rvalue = alloca(cif->rtype->size); } @@ -368,6 +375,7 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue) #elif defined(X86_WIN32) case FFI_SYSV: case FFI_STDCALL: + case FFI_MS_CDECL: ffi_call_win32(ffi_prep_args, &ecif, cif->abi, cif->bytes, cif->flags, ecif.rvalue, fn); break; @@ -513,7 +521,8 @@ ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, void **avalue, argp += sizeof(void *); } #else - if ( cif->flags == FFI_TYPE_STRUCT ) { + if ( cif->flags == FFI_TYPE_STRUCT + || cif->flags == FFI_TYPE_MS_STRUCT ) { *rvalue = *(void **) argp; argp += sizeof(void *); } @@ -673,6 +682,12 @@ ffi_prep_closure_loc (ffi_closure* closure, &ffi_closure_STDCALL, (void*)codeloc, cif->bytes); } + else if (cif->abi == FFI_MS_CDECL) + { + FFI_INIT_TRAMPOLINE (&closure->tramp[0], + &ffi_closure_SYSV, + (void*)codeloc); + } #endif /* X86_WIN32 */ #endif /* !X86_WIN64 */ else @@ -762,8 +777,9 @@ ffi_raw_call(ffi_cif *cif, void (*fn)(void), void *rvalue, ffi_raw *fake_avalue) /* If the return value is a struct and we don't have a return */ /* value address then we need to make one */ - if ((rvalue == NULL) && - (cif->rtype->type == FFI_TYPE_STRUCT)) + if (rvalue == NULL + && (cif->flags == FFI_TYPE_STRUCT + || cif->flags == FFI_TYPE_MS_STRUCT)) { ecif.rvalue = alloca(cif->rtype->size); } @@ -776,6 +792,7 @@ ffi_raw_call(ffi_cif *cif, void (*fn)(void), void *rvalue, ffi_raw *fake_avalue) #ifdef X86_WIN32 case FFI_SYSV: case FFI_STDCALL: + case FFI_MS_CDECL: ffi_call_win32(ffi_prep_args_raw, &ecif, cif->abi, cif->bytes, cif->flags, ecif.rvalue, fn); break; diff --git a/libffi/src/x86/ffi64.c b/libffi/src/x86/ffi64.c index defd774..1daa1c0 100644 --- a/libffi/src/x86/ffi64.c +++ b/libffi/src/x86/ffi64.c @@ -37,11 +37,17 @@ #define MAX_GPR_REGS 6 #define MAX_SSE_REGS 8 +#ifdef __INTEL_COMPILER +#define UINT128 __m128 +#else +#define UINT128 __int128_t +#endif + struct register_args { /* Registers for argument passing. */ UINT64 gpr[MAX_GPR_REGS]; - __int128_t sse[MAX_SSE_REGS]; + UINT128 sse[MAX_SSE_REGS]; }; extern void ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags, diff --git a/libffi/src/x86/ffitarget.h b/libffi/src/x86/ffitarget.h index fc01541..46f294c 100644 --- a/libffi/src/x86/ffitarget.h +++ b/libffi/src/x86/ffitarget.h @@ -81,9 +81,13 @@ typedef enum ffi_abi { FFI_STDCALL, FFI_THISCALL, FFI_FASTCALL, + FFI_MS_CDECL, FFI_LAST_ABI, - /* TODO: Add fastcall support for the sake of completeness */ +#ifdef _MSC_VER + FFI_DEFAULT_ABI = FFI_MS_CDECL +#else FFI_DEFAULT_ABI = FFI_SYSV +#endif #elif defined(X86_WIN64) FFI_WIN64, @@ -110,6 +114,7 @@ typedef enum ffi_abi { #define FFI_TYPE_SMALL_STRUCT_1B (FFI_TYPE_LAST + 1) #define FFI_TYPE_SMALL_STRUCT_2B (FFI_TYPE_LAST + 2) #define FFI_TYPE_SMALL_STRUCT_4B (FFI_TYPE_LAST + 3) +#define FFI_TYPE_MS_STRUCT (FFI_TYPE_LAST + 4) #if defined (X86_64) || (defined (__x86_64__) && defined (X86_DARWIN)) #define FFI_TRAMPOLINE_SIZE 24 diff --git a/libffi/src/x86/win32.S b/libffi/src/x86/win32.S index deb4a03..24b7bbd 100644 --- a/libffi/src/x86/win32.S +++ b/libffi/src/x86/win32.S @@ -108,31 +108,37 @@ ca_jumpdata: dd offset ca_retfloat ;; FFI_TYPE_FLOAT dd offset ca_retdouble ;; FFI_TYPE_DOUBLE dd offset ca_retlongdouble ;; FFI_TYPE_LONGDOUBLE - dd offset ca_retint8 ;; FFI_TYPE_UINT8 - dd offset ca_retint8 ;; FFI_TYPE_SINT8 - dd offset ca_retint16 ;; FFI_TYPE_UINT16 - dd offset ca_retint16 ;; FFI_TYPE_SINT16 + dd offset ca_retuint8 ;; FFI_TYPE_UINT8 + dd offset ca_retsint8 ;; FFI_TYPE_SINT8 + dd offset ca_retuint16 ;; FFI_TYPE_UINT16 + dd offset ca_retsint16 ;; FFI_TYPE_SINT16 dd offset ca_retint ;; FFI_TYPE_UINT32 dd offset ca_retint ;; FFI_TYPE_SINT32 dd offset ca_retint64 ;; FFI_TYPE_UINT64 dd offset ca_retint64 ;; FFI_TYPE_SINT64 dd offset ca_epilogue ;; FFI_TYPE_STRUCT dd offset ca_retint ;; FFI_TYPE_POINTER - dd offset ca_retint8 ;; FFI_TYPE_SMALL_STRUCT_1B - dd offset ca_retint16 ;; FFI_TYPE_SMALL_STRUCT_2B + dd offset ca_retstruct1b ;; FFI_TYPE_SMALL_STRUCT_1B + dd offset ca_retstruct2b ;; FFI_TYPE_SMALL_STRUCT_2B dd offset ca_retint ;; FFI_TYPE_SMALL_STRUCT_4B + dd offset ca_epilogue ;; FFI_TYPE_MS_STRUCT -ca_retint8: - ;; Load %ecx with the pointer to storage for the return value - mov ecx, rvalue - mov [ecx + 0], al - jmp ca_epilogue + /* Sign/zero extend as appropriate. */ +ca_retuint8: + movzx eax, al + jmp ca_retint -ca_retint16: - ;; Load %ecx with the pointer to storage for the return value - mov ecx, rvalue - mov [ecx + 0], ax - jmp ca_epilogue +ca_retsint8: + movsx eax, al + jmp ca_retint + +ca_retuint16: + movzx eax, ax + jmp ca_retint + +ca_retsint16: + movsx eax, ax + jmp ca_retint ca_retint: ;; Load %ecx with the pointer to storage for the return value @@ -165,14 +171,24 @@ ca_retlongdouble: fstp TBYTE PTR [ecx] jmp ca_epilogue +ca_retstruct1b: + ;; Load %ecx with the pointer to storage for the return value + mov ecx, rvalue + mov [ecx + 0], al + jmp ca_epilogue + +ca_retstruct2b: + ;; Load %ecx with the pointer to storage for the return value + mov ecx, rvalue + mov [ecx + 0], ax + jmp ca_epilogue + ca_epilogue: ;; Epilogue code is autogenerated. ret ffi_call_win32 ENDP ffi_closure_THISCALL PROC NEAR FORCEFRAME - push ebp - mov ebp, esp sub esp, 40 lea edx, [ebp -24] mov [ebp - 12], edx /* resp */ @@ -187,7 +203,7 @@ ffi_closure_SYSV PROC NEAR FORCEFRAME lea edx, [ebp - 24] mov [ebp - 12], edx ;; resp lea edx, [ebp + 8] -stub: +stub:: mov [esp + 8], edx ;; args lea edx, [ebp - 12] mov [esp + 4], edx ;; &resp @@ -204,26 +220,35 @@ cs_jumpdata: dd offset cs_retfloat ;; FFI_TYPE_FLOAT dd offset cs_retdouble ;; FFI_TYPE_DOUBLE dd offset cs_retlongdouble ;; FFI_TYPE_LONGDOUBLE - dd offset cs_retint8 ;; FFI_TYPE_UINT8 - dd offset cs_retint8 ;; FFI_TYPE_SINT8 - dd offset cs_retint16 ;; FFI_TYPE_UINT16 - dd offset cs_retint16 ;; FFI_TYPE_SINT16 + dd offset cs_retuint8 ;; FFI_TYPE_UINT8 + dd offset cs_retsint8 ;; FFI_TYPE_SINT8 + dd offset cs_retuint16 ;; FFI_TYPE_UINT16 + dd offset cs_retsint16 ;; FFI_TYPE_SINT16 dd offset cs_retint ;; FFI_TYPE_UINT32 dd offset cs_retint ;; FFI_TYPE_SINT32 dd offset cs_retint64 ;; FFI_TYPE_UINT64 dd offset cs_retint64 ;; FFI_TYPE_SINT64 dd offset cs_retstruct ;; FFI_TYPE_STRUCT dd offset cs_retint ;; FFI_TYPE_POINTER - dd offset cs_retint8 ;; FFI_TYPE_SMALL_STRUCT_1B - dd offset cs_retint16 ;; FFI_TYPE_SMALL_STRUCT_2B + dd offset cs_retsint8 ;; FFI_TYPE_SMALL_STRUCT_1B + dd offset cs_retsint16 ;; FFI_TYPE_SMALL_STRUCT_2B dd offset cs_retint ;; FFI_TYPE_SMALL_STRUCT_4B + dd offset cs_retmsstruct ;; FFI_TYPE_MS_STRUCT + +cs_retuint8: + movzx eax, BYTE PTR [ecx] + jmp cs_epilogue + +cs_retsint8: + movsx eax, BYTE PTR [ecx] + jmp cs_epilogue -cs_retint8: - mov al, [ecx] +cs_retuint16: + movzx eax, WORD PTR [ecx] jmp cs_epilogue -cs_retint16: - mov ax, [ecx] +cs_retsint16: + movsx eax, WORD PTR [ecx] jmp cs_epilogue cs_retint: @@ -252,6 +277,12 @@ cs_retstruct: ;; Epilogue code is autogenerated. ret 4 +cs_retmsstruct: + ;; Caller expects us to return a pointer to the real return value. + mov eax, ecx + ;; Caller doesn't expects us to pop struct return value pointer hidden arg. + jmp cs_epilogue + cs_epilogue: ;; Epilogue code is autogenerated. ret @@ -264,19 +295,16 @@ ffi_closure_SYSV ENDP #define RAW_CLOSURE_USER_DATA_OFFSET (RAW_CLOSURE_FUN_OFFSET + 4) #define CIF_FLAGS_OFFSET 20 -ffi_closure_raw_THISCALL PROC NEAR - push ebp - mov ebp, esp - push esi +ffi_closure_raw_THISCALL PROC NEAR USES esi FORCEFRAME sub esp, 36 mov esi, [eax + RAW_CLOSURE_CIF_OFFSET] ;; closure->cif mov edx, [eax + RAW_CLOSURE_USER_DATA_OFFSET] ;; closure->user_data mov [esp + 12], edx - lea edx, [ebp + 12], edx + lea edx, [ebp + 12] jmp stubraw -ffi_closure_raw_SYSV ENDP +ffi_closure_raw_THISCALL ENDP -ffi_closure_raw_SYSV PROC NEAR USES esi +ffi_closure_raw_SYSV PROC NEAR USES esi FORCEFRAME ;; the ffi_closure ctx is passed in eax by the trampoline. sub esp, 40 @@ -284,7 +312,7 @@ ffi_closure_raw_SYSV PROC NEAR USES esi mov edx, [eax + RAW_CLOSURE_USER_DATA_OFFSET] ;; closure->user_data mov [esp + 12], edx ;; user_data lea edx, [ebp + 8] -stubraw: +stubraw:: mov [esp + 8], edx ;; raw_args lea edx, [ebp - 24] mov [esp + 4], edx ;; &res @@ -302,26 +330,35 @@ cr_jumpdata: dd offset cr_retfloat ;; FFI_TYPE_FLOAT dd offset cr_retdouble ;; FFI_TYPE_DOUBLE dd offset cr_retlongdouble ;; FFI_TYPE_LONGDOUBLE - dd offset cr_retint8 ;; FFI_TYPE_UINT8 - dd offset cr_retint8 ;; FFI_TYPE_SINT8 - dd offset cr_retint16 ;; FFI_TYPE_UINT16 - dd offset cr_retint16 ;; FFI_TYPE_SINT16 + dd offset cr_retuint8 ;; FFI_TYPE_UINT8 + dd offset cr_retsint8 ;; FFI_TYPE_SINT8 + dd offset cr_retuint16 ;; FFI_TYPE_UINT16 + dd offset cr_retsint16 ;; FFI_TYPE_SINT16 dd offset cr_retint ;; FFI_TYPE_UINT32 dd offset cr_retint ;; FFI_TYPE_SINT32 dd offset cr_retint64 ;; FFI_TYPE_UINT64 dd offset cr_retint64 ;; FFI_TYPE_SINT64 dd offset cr_epilogue ;; FFI_TYPE_STRUCT dd offset cr_retint ;; FFI_TYPE_POINTER - dd offset cr_retint8 ;; FFI_TYPE_SMALL_STRUCT_1B - dd offset cr_retint16 ;; FFI_TYPE_SMALL_STRUCT_2B + dd offset cr_retsint8 ;; FFI_TYPE_SMALL_STRUCT_1B + dd offset cr_retsint16 ;; FFI_TYPE_SMALL_STRUCT_2B dd offset cr_retint ;; FFI_TYPE_SMALL_STRUCT_4B + dd offset cr_epilogue ;; FFI_TYPE_MS_STRUCT + +cr_retuint8: + movzx eax, BYTE PTR [ecx] + jmp cr_epilogue + +cr_retsint8: + movsx eax, BYTE PTR [ecx] + jmp cr_epilogue -cr_retint8: - mov al, [ecx] +cr_retuint16: + movzx eax, WORD PTR [ecx] jmp cr_epilogue -cr_retint16: - mov ax, [ecx] +cr_retsint16: + movsx eax, WORD PTR [ecx] jmp cr_epilogue cr_retint: @@ -375,26 +412,34 @@ cd_jumpdata: dd offset cd_retfloat ;; FFI_TYPE_FLOAT dd offset cd_retdouble ;; FFI_TYPE_DOUBLE dd offset cd_retlongdouble ;; FFI_TYPE_LONGDOUBLE - dd offset cd_retint8 ;; FFI_TYPE_UINT8 - dd offset cd_retint8 ;; FFI_TYPE_SINT8 - dd offset cd_retint16 ;; FFI_TYPE_UINT16 - dd offset cd_retint16 ;; FFI_TYPE_SINT16 + dd offset cd_retuint8 ;; FFI_TYPE_UINT8 + dd offset cd_retsint8 ;; FFI_TYPE_SINT8 + dd offset cd_retuint16 ;; FFI_TYPE_UINT16 + dd offset cd_retsint16 ;; FFI_TYPE_SINT16 dd offset cd_retint ;; FFI_TYPE_UINT32 dd offset cd_retint ;; FFI_TYPE_SINT32 dd offset cd_retint64 ;; FFI_TYPE_UINT64 dd offset cd_retint64 ;; FFI_TYPE_SINT64 dd offset cd_epilogue ;; FFI_TYPE_STRUCT dd offset cd_retint ;; FFI_TYPE_POINTER - dd offset cd_retint8 ;; FFI_TYPE_SMALL_STRUCT_1B - dd offset cd_retint16 ;; FFI_TYPE_SMALL_STRUCT_2B + dd offset cd_retsint8 ;; FFI_TYPE_SMALL_STRUCT_1B + dd offset cd_retsint16 ;; FFI_TYPE_SMALL_STRUCT_2B dd offset cd_retint ;; FFI_TYPE_SMALL_STRUCT_4B -cd_retint8: - mov al, [ecx] +cd_retuint8: + movzx eax, BYTE PTR [ecx] + jmp cd_epilogue + +cd_retsint8: + movsx eax, BYTE PTR [ecx] jmp cd_epilogue -cd_retint16: - mov ax, [ecx] +cd_retuint16: + movzx eax, WORD PTR [ecx] + jmp cd_epilogue + +cd_retsint16: + movsx eax, WORD PTR [ecx] jmp cd_epilogue cd_retint: @@ -515,6 +560,7 @@ _ffi_call_win32: .long .Lretstruct1b /* FFI_TYPE_SMALL_STRUCT_1B */ .long .Lretstruct2b /* FFI_TYPE_SMALL_STRUCT_2B */ .long .Lretstruct4b /* FFI_TYPE_SMALL_STRUCT_4B */ + .long .Lretstruct /* FFI_TYPE_MS_STRUCT */ 1: add %ecx, %ecx add %ecx, %ecx @@ -657,6 +703,7 @@ _ffi_closure_SYSV: .long .Lcls_retstruct1 /* FFI_TYPE_SMALL_STRUCT_1B */ .long .Lcls_retstruct2 /* FFI_TYPE_SMALL_STRUCT_2B */ .long .Lcls_retstruct4 /* FFI_TYPE_SMALL_STRUCT_4B */ + .long .Lcls_retmsstruct /* FFI_TYPE_MS_STRUCT */ 1: add %eax, %eax @@ -721,6 +768,12 @@ _ffi_closure_SYSV: popl %ebp ret $0x4 +.Lcls_retmsstruct: + # Caller expects us to return a pointer to the real return value. + mov %ecx, %eax + # Caller doesn't expects us to pop struct return value pointer hidden arg. + jmp .Lcls_epilogue + .Lcls_noretval: .Lcls_epilogue: movl %ebp, %esp @@ -798,6 +851,7 @@ _ffi_closure_raw_SYSV: .long .Lrcls_retstruct1 /* FFI_TYPE_SMALL_STRUCT_1B */ .long .Lrcls_retstruct2 /* FFI_TYPE_SMALL_STRUCT_2B */ .long .Lrcls_retstruct4 /* FFI_TYPE_SMALL_STRUCT_4B */ + .long .Lrcls_retstruct /* FFI_TYPE_MS_STRUCT */ 1: add %eax, %eax add %eax, %eax |