diff options
Diffstat (limited to 'src/xtensa')
| -rw-r--r-- | src/xtensa/ffi.c | 298 | ||||
| -rw-r--r-- | src/xtensa/ffitarget.h | 53 | ||||
| -rw-r--r-- | src/xtensa/sysv.S | 253 |
3 files changed, 604 insertions, 0 deletions
diff --git a/src/xtensa/ffi.c b/src/xtensa/ffi.c new file mode 100644 index 0000000..fd94daf --- /dev/null +++ b/src/xtensa/ffi.c @@ -0,0 +1,298 @@ +/* ----------------------------------------------------------------------- + ffi.c - Copyright (c) 2013 Tensilica, Inc. + + XTENSA 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> + +/* + |----------------------------------------| + | | + on entry to ffi_call ----> |----------------------------------------| + | caller stack frame for registers a0-a3 | + |----------------------------------------| + | | + | additional arguments | + entry of the function ---> |----------------------------------------| + | copy of function arguments a2-a7 | + | - - - - - - - - - - - - - | + | | + + The area below the entry line becomes the new stack frame for the function. + +*/ + + +#define FFI_TYPE_STRUCT_REGS FFI_TYPE_LAST + + +extern void ffi_call_SYSV(void *rvalue, unsigned rsize, unsigned flags, + void(*fn)(void), unsigned nbytes, extended_cif*); +extern void ffi_closure_SYSV(void) FFI_HIDDEN; + +ffi_status ffi_prep_cif_machdep(ffi_cif *cif) +{ + switch(cif->rtype->type) { + case FFI_TYPE_SINT8: + case FFI_TYPE_UINT8: + case FFI_TYPE_SINT16: + case FFI_TYPE_UINT16: + cif->flags = cif->rtype->type; + break; + case FFI_TYPE_VOID: + case FFI_TYPE_FLOAT: + cif->flags = FFI_TYPE_UINT32; + break; + case FFI_TYPE_DOUBLE: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + cif->flags = FFI_TYPE_UINT64; // cif->rtype->type; + break; + case FFI_TYPE_STRUCT: + cif->flags = FFI_TYPE_STRUCT; //_REGS; + /* Up to 16 bytes are returned in registers */ + if (cif->rtype->size > 4 * 4) { + /* returned structure is referenced by a register; use 8 bytes + (including 4 bytes for potential additional alignment) */ + cif->flags = FFI_TYPE_STRUCT; + cif->bytes += 8; + } + break; + + default: + cif->flags = FFI_TYPE_UINT32; + break; + } + + /* Round the stack up to a full 4 register frame, just in case + (we use this size in movsp). This way, it's also a multiple of + 8 bytes for 64-bit arguments. */ + cif->bytes = ALIGN(cif->bytes, 16); + + return FFI_OK; +} + +void ffi_prep_args(extended_cif *ecif, unsigned char* stack) +{ + unsigned int i; + unsigned long *addr; + ffi_type **ptr; + + union { + void **v; + char **c; + signed char **sc; + unsigned char **uc; + signed short **ss; + unsigned short **us; + unsigned int **i; + long long **ll; + float **f; + double **d; + } p_argv; + + /* Verify that everything is aligned up properly */ + FFI_ASSERT (((unsigned long) stack & 0x7) == 0); + + p_argv.v = ecif->avalue; + addr = (unsigned long*)stack; + + /* structures with a size greater than 16 bytes are passed in memory */ + if (ecif->cif->rtype->type == FFI_TYPE_STRUCT && ecif->cif->rtype->size > 16) + { + *addr++ = (unsigned long)ecif->rvalue; + } + + for (i = ecif->cif->nargs, ptr = ecif->cif->arg_types; + i > 0; + i--, ptr++, p_argv.v++) + { + switch ((*ptr)->type) + { + case FFI_TYPE_SINT8: + *addr++ = **p_argv.sc; + break; + case FFI_TYPE_UINT8: + *addr++ = **p_argv.uc; + break; + case FFI_TYPE_SINT16: + *addr++ = **p_argv.ss; + break; + case FFI_TYPE_UINT16: + *addr++ = **p_argv.us; + break; + case FFI_TYPE_FLOAT: + case FFI_TYPE_INT: + case FFI_TYPE_UINT32: + case FFI_TYPE_SINT32: + case FFI_TYPE_POINTER: + *addr++ = **p_argv.i; + break; + case FFI_TYPE_DOUBLE: + case FFI_TYPE_UINT64: + case FFI_TYPE_SINT64: + if (((unsigned long)addr & 4) != 0) + addr++; + *(unsigned long long*)addr = **p_argv.ll; + addr += sizeof(unsigned long long) / sizeof (addr); + break; + + case FFI_TYPE_STRUCT: + { + unsigned long offs; + unsigned long size; + + if (((unsigned long)addr & 4) != 0 && (*ptr)->alignment > 4) + addr++; + + offs = (unsigned long) addr - (unsigned long) stack; + size = (*ptr)->size; + + /* Entire structure must fit the argument registers or referenced */ + if (offs < FFI_REGISTER_NARGS * 4 + && offs + size > FFI_REGISTER_NARGS * 4) + addr = (unsigned long*) (stack + FFI_REGISTER_NARGS * 4); + + memcpy((char*) addr, *p_argv.c, size); + addr += (size + 3) / 4; + break; + } + + default: + FFI_ASSERT(0); + } + } +} + + +void ffi_call(ffi_cif* cif, void(*fn)(void), void *rvalue, void **avalue) +{ + extended_cif ecif; + unsigned long rsize = cif->rtype->size; + int flags = cif->flags; + void *alloc = NULL; + + ecif.cif = cif; + ecif.avalue = avalue; + + /* Note that for structures that are returned in registers (size <= 16 bytes) + we allocate a temporary buffer and use memcpy to copy it to the final + destination. The reason is that the target address might be misaligned or + the length not a multiple of 4 bytes. Handling all those cases would be + very complex. */ + + if (flags == FFI_TYPE_STRUCT && (rsize <= 16 || rvalue == NULL)) + { + alloc = alloca(ALIGN(rsize, 4)); + ecif.rvalue = alloc; + } + else + { + ecif.rvalue = rvalue; + } + + if (cif->abi != FFI_SYSV) + FFI_ASSERT(0); + + ffi_call_SYSV (ecif.rvalue, rsize, cif->flags, fn, cif->bytes, &ecif); + + if (alloc != NULL && rvalue != NULL) + memcpy(rvalue, alloc, rsize); +} + +extern void ffi_trampoline(); +extern void ffi_cacheflush(void* start, void* end); + +ffi_status +ffi_prep_closure_loc (ffi_closure* closure, + ffi_cif* cif, + void (*fun)(ffi_cif*, void*, void**, void*), + void *user_data, + void *codeloc) +{ + /* copye trampoline to stack and patch 'ffi_closure_SYSV' pointer */ + memcpy(closure->tramp, ffi_trampoline, FFI_TRAMPOLINE_SIZE); + *(unsigned int*)(&closure->tramp[8]) = (unsigned int)ffi_closure_SYSV; + + // Do we have this function? + // __builtin___clear_cache(closer->tramp, closer->tramp + FFI_TRAMPOLINE_SIZE) + ffi_cacheflush(closure->tramp, closure->tramp + FFI_TRAMPOLINE_SIZE); + + closure->cif = cif; + closure->fun = fun; + closure->user_data = user_data; + return FFI_OK; +} + + +long FFI_HIDDEN +ffi_closure_SYSV_inner(ffi_closure *closure, void **values, void *rvalue) +{ + ffi_cif *cif; + ffi_type **arg_types; + void **avalue; + int i, areg; + + cif = closure->cif; + if (cif->abi != FFI_SYSV) + return FFI_BAD_ABI; + + areg = 0; + + int rtype = cif->rtype->type; + if (rtype == FFI_TYPE_STRUCT && cif->rtype->size > 4 * 4) + { + rvalue = *values; + areg++; + } + + cif = closure->cif; + arg_types = cif->arg_types; + avalue = alloca(cif->nargs * sizeof(void *)); + + for (i = 0; i < cif->nargs; i++) + { + if (arg_types[i]->alignment == 8 && (areg & 1) != 0) + areg++; + + // skip the entry 16,a1 framework, add 16 bytes (4 registers) + if (areg == FFI_REGISTER_NARGS) + areg += 4; + + if (arg_types[i]->type == FFI_TYPE_STRUCT) + { + int numregs = ((arg_types[i]->size + 3) & ~3) / 4; + if (areg < FFI_REGISTER_NARGS && areg + numregs > FFI_REGISTER_NARGS) + areg = FFI_REGISTER_NARGS + 4; + } + + avalue[i] = &values[areg]; + areg += (arg_types[i]->size + 3) / 4; + } + + (closure->fun)(cif, rvalue, avalue, closure->user_data); + + return rtype; +} diff --git a/src/xtensa/ffitarget.h b/src/xtensa/ffitarget.h new file mode 100644 index 0000000..0ba728b --- /dev/null +++ b/src/xtensa/ffitarget.h @@ -0,0 +1,53 @@ +/* -----------------------------------------------------------------*-C-*- + ffitarget.h - Copyright (c) 2013 Tensilica, Inc. + Target configuration macros for XTENSA. + + 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 + +#define FFI_REGISTER_NARGS 6 + +/* ---- Definitions for closures ----------------------------------------- */ + +#define FFI_CLOSURES 1 +#define FFI_NATIVE_RAW_API 0 +#define FFI_TRAMPOLINE_SIZE 24 + +#endif diff --git a/src/xtensa/sysv.S b/src/xtensa/sysv.S new file mode 100644 index 0000000..64e6a09 --- /dev/null +++ b/src/xtensa/sysv.S @@ -0,0 +1,253 @@ +/* ----------------------------------------------------------------------- + sysv.S - Copyright (c) 2013 Tensilica, Inc. + + XTENSA 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> + +#define ENTRY(name) .text; .globl name; .type name,@function; .align 4; name: +#define END(name) .size name , . - name + +/* Assert that the table below is in sync with ffi.h. */ + +#if FFI_TYPE_UINT8 != 5 \ + || FFI_TYPE_SINT8 != 6 \ + || FFI_TYPE_UINT16 != 7 \ + || FFI_TYPE_SINT16 != 8 \ + || FFI_TYPE_UINT32 != 9 \ + || FFI_TYPE_SINT32 != 10 \ + || FFI_TYPE_UINT64 != 11 +#error "xtensa/sysv.S out of sync with ffi.h" +#endif + + +/* ffi_call_SYSV (rvalue, rbytes, flags, (*fnaddr)(), bytes, ecif) + void *rvalue; a2 + unsigned long rbytes; a3 + unsigned flags; a4 + void (*fnaddr)(); a5 + unsigned long bytes; a6 + extended_cif* ecif) a7 +*/ + +ENTRY(ffi_call_SYSV) + + entry a1, 32 # 32 byte frame for using call8 below + + mov a10, a7 # a10(->arg0): ecif + sub a11, a1, a6 # a11(->arg1): stack pointer + mov a7, a1 # fp + movsp a1, a11 # set new sp = old_sp - bytes + + movi a8, ffi_prep_args + callx8 a8 # ffi_prep_args(ecif, stack) + + # prepare to move stack pointer back up to 6 arguments + # note that 'bytes' is already aligned + + movi a10, 6*4 + sub a11, a6, a10 + movgez a6, a10, a11 + add a6, a1, a6 + + + # we can pass up to 6 arguments in registers + # for simplicity, just load 6 arguments + # (the stack size is at least 32 bytes, so no risk to cross boundaries) + + l32i a10, a1, 0 + l32i a11, a1, 4 + l32i a12, a1, 8 + l32i a13, a1, 12 + l32i a14, a1, 16 + l32i a15, a1, 20 + + # move stack pointer + + movsp a1, a6 + + callx8 a5 # (*fn)(args...) + + # Handle return value(s) + + beqz a2, .Lexit + + movi a5, FFI_TYPE_STRUCT + bne a4, a5, .Lstore + movi a5, 16 + blt a5, a3, .Lexit + + s32i a10, a2, 0 + blti a3, 5, .Lexit + addi a3, a3, -1 + s32i a11, a2, 4 + blti a3, 8, .Lexit + s32i a12, a2, 8 + blti a3, 12, .Lexit + s32i a13, a2, 12 + +.Lexit: retw + +.Lstore: + addi a4, a4, -FFI_TYPE_UINT8 + bgei a4, 7, .Lexit # should never happen + movi a6, store_calls + add a4, a4, a4 + addx4 a6, a4, a6 # store_table + idx * 8 + jx a6 + + .align 8 +store_calls: + # UINT8 + s8i a10, a2, 0 + retw + + # SINT8 + .align 8 + s8i a10, a2, 0 + retw + + # UINT16 + .align 8 + s16i a10, a2, 0 + retw + + # SINT16 + .align 8 + s16i a10, a2, 0 + retw + + # UINT32 + .align 8 + s32i a10, a2, 0 + retw + + # SINT32 + .align 8 + s32i a10, a2, 0 + retw + + # UINT64 + .align 8 + s32i a10, a2, 0 + s32i a11, a2, 4 + retw + +END(ffi_call_SYSV) + + +/* + * void ffi_cacheflush (unsigned long start, unsigned long end) + */ + +#define EXTRA_ARGS_SIZE 24 + +ENTRY(ffi_cacheflush) + + entry a1, 16 + +1: dhwbi a2, 0 + ihi a2, 0 + addi a2, a2, 4 + blt a2, a3, 1b + + retw + +END(ffi_cacheflush) + +/* ffi_trampoline is copied to the stack */ + +ENTRY(ffi_trampoline) + + entry a1, 16 + (FFI_REGISTER_NARGS * 4) + (4 * 4) # [ 0] + j 2f # [ 3] + .align 4 # [ 6] +1: .long 0 # [ 8] +2: l32r a15, 1b # [12] + _mov a14, a0 # [15] + callx0 a15 # [18] + # [21] +END(ffi_trampoline) + +/* + * ffi_closure() + * + * a0: closure + 21 + * a14: return address (a0) + */ + +ENTRY(ffi_closure_SYSV) + + /* intentionally omitting entry here */ + + # restore return address (a0) and move pointer to closure to a10 + addi a10, a0, -21 + mov a0, a14 + + # allow up to 4 arguments as return values + addi a11, a1, 4 * 4 + + # save up to 6 arguments to stack (allocated by entry below) + s32i a2, a11, 0 + s32i a3, a11, 4 + s32i a4, a11, 8 + s32i a5, a11, 12 + s32i a6, a11, 16 + s32i a7, a11, 20 + + movi a8, ffi_closure_SYSV_inner + mov a12, a1 + callx8 a8 # .._inner(*closure, **avalue, *rvalue) + + # load up to four return arguments + l32i a2, a1, 0 + l32i a3, a1, 4 + l32i a4, a1, 8 + l32i a5, a1, 12 + + # (sign-)extend return value + movi a11, FFI_TYPE_UINT8 + bne a10, a11, 1f + extui a2, a2, 0, 8 + retw + +1: movi a11, FFI_TYPE_SINT8 + bne a10, a11, 1f + sext a2, a2, 7 + retw + +1: movi a11, FFI_TYPE_UINT16 + bne a10, a11, 1f + extui a2, a2, 0, 16 + retw + +1: movi a11, FFI_TYPE_SINT16 + bne a10, a11, 1f + sext a2, a2, 15 + +1: retw + +END(ffi_closure_SYSV) |