static void memcpy_endian(void *dst, void *src, size_t n)
{
union { uint8_t b; uint32_t u; } host_endian;
host_endian.u = 1;
if (host_endian.b == LJ_ENDIAN_SELECT(1, 0)) {
memcpy(dst, src, n);
} else {
size_t i;
for (i = 0; i < n; i++)
((uint8_t *)dst)[i] = ((uint8_t *)src)[n-i-1];
}
}
static void libdef_fixupbc(uint8_t *p)
{
uint32_t i, sizebc;
p += 4;
p = libdef_uleb128(p, &sizebc);
p = libdef_uleb128(p, &sizebc);
p = libdef_uleb128(p, &sizebc);
for (i = 0; i < sizebc; i++, p += 4) {
uint8_t op = p[libbc_endian ? 3 : 0];
uint8_t ra = p[libbc_endian ? 2 : 1];
uint8_t rc = p[libbc_endian ? 1 : 2];
uint8_t rb = p[libbc_endian ? 0 : 3];
if (!LJ_DUALNUM && op == BC_ISTYPE && rc == ~LJ_TNUMX+1) {
op = BC_ISNUM; rc++;
}
p[LJ_ENDIAN_SELECT(0, 3)] = op;
p[LJ_ENDIAN_SELECT(1, 2)] = ra;
p[LJ_ENDIAN_SELECT(2, 1)] = rc;
p[LJ_ENDIAN_SELECT(3, 0)] = rb;
}
}
/* Emit Windows PE object file. */
void emit_peobj(BuildCtx *ctx)
{
PEheader pehdr;
PEsection pesect[PEOBJ_NSECTIONS];
uint32_t sofs;
int i, nrsym;
union { uint8_t b; uint32_t u; } host_endian;
host_endian.u = 1;
if (host_endian.b != LJ_ENDIAN_SELECT(1, 0)) {
fprintf(stderr, "Error: different byte order for host and target\n");
exit(1);
}
sofs = sizeof(PEheader) + PEOBJ_NSECTIONS*sizeof(PEsection);
/* Fill in PE sections. */
memset(&pesect, 0, PEOBJ_NSECTIONS*sizeof(PEsection));
memcpy(pesect[PEOBJ_SECT_TEXT].name, ".text", sizeof(".text")-1);
pesect[PEOBJ_SECT_TEXT].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_TEXT].size = (uint32_t)ctx->codesz);
pesect[PEOBJ_SECT_TEXT].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_TEXT].nreloc = (uint16_t)ctx->nreloc) * PEOBJ_RELOC_SIZE;
/* Flags: 60 = read+execute, 50 = align16, 20 = code. */
pesect[PEOBJ_SECT_TEXT].flags = 0x60500020;
#if LJ_TARGET_X64
memcpy(pesect[PEOBJ_SECT_PDATA].name, ".pdata", sizeof(".pdata")-1);
pesect[PEOBJ_SECT_PDATA].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_PDATA].size = 6*4);
pesect[PEOBJ_SECT_PDATA].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_PDATA].nreloc = 6) * PEOBJ_RELOC_SIZE;
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_PDATA].flags = 0x40300040;
memcpy(pesect[PEOBJ_SECT_XDATA].name, ".xdata", sizeof(".xdata")-1);
pesect[PEOBJ_SECT_XDATA].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_XDATA].size = 8*2+4+6*2); /* See below. */
pesect[PEOBJ_SECT_XDATA].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_XDATA].nreloc = 1) * PEOBJ_RELOC_SIZE;
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_XDATA].flags = 0x40300040;
#endif
memcpy(pesect[PEOBJ_SECT_RDATA_Z].name, ".rdata$Z", sizeof(".rdata$Z")-1);
pesect[PEOBJ_SECT_RDATA_Z].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_RDATA_Z].size = (uint32_t)strlen(ctx->dasm_ident)+1);
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_RDATA_Z].flags = 0x40300040;
/* Fill in PE header. */
pehdr.arch = PEOBJ_ARCH_TARGET;
pehdr.nsects = PEOBJ_NSECTIONS;
pehdr.time = 0; /* Timestamp is optional. */
pehdr.symtabofs = sofs;
pehdr.opthdrsz = 0;
pehdr.flags = 0;
/* Compute the size of the symbol table:
** @feat.00 + nsections*2
** + asm_start + nsym
** + nrsym
*/
nrsym = ctx->nrelocsym;
pehdr.nsyms = 1+PEOBJ_NSECTIONS*2 + 1+ctx->nsym + nrsym;
#if LJ_TARGET_X64
pehdr.nsyms += 1; /* Symbol for lj_err_unwind_win64. */
#endif
/* Write PE object header and all sections. */
owrite(ctx, &pehdr, sizeof(PEheader));
owrite(ctx, &pesect, sizeof(PEsection)*PEOBJ_NSECTIONS);
/* Write .text section. */
owrite(ctx, ctx->code, ctx->codesz);
for (i = 0; i < ctx->nreloc; i++) {
PEreloc reloc;
reloc.vaddr = (uint32_t)ctx->reloc[i].ofs;
reloc.symidx = 1+2+ctx->reloc[i].sym; /* Reloc syms are after .text sym. */
reloc.type = ctx->reloc[i].type ? PEOBJ_RELOC_REL32 : PEOBJ_RELOC_DIR32;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
}
#if LJ_TARGET_X64
{ /* Write .pdata section. */
uint32_t fcofs = (uint32_t)ctx->sym[ctx->nsym-1].ofs;
uint32_t pdata[3]; /* Start of .text, end of .text and .xdata. */
PEreloc reloc;
pdata[0] = 0; pdata[1] = fcofs; pdata[2] = 0;
owrite(ctx, &pdata, sizeof(pdata));
pdata[0] = fcofs; pdata[1] = (uint32_t)ctx->codesz; pdata[2] = 20;
owrite(ctx, &pdata, sizeof(pdata));
reloc.vaddr = 0; reloc.symidx = 1+2+nrsym+2+2+1;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
reloc.vaddr = 4; reloc.symidx = 1+2+nrsym+2+2+1;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
reloc.vaddr = 8; reloc.symidx = 1+2+nrsym+2;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
reloc.vaddr = 12; reloc.symidx = 1+2+nrsym+2+2+1;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
reloc.vaddr = 16; reloc.symidx = 1+2+nrsym+2+2+1;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
reloc.vaddr = 20; reloc.symidx = 1+2+nrsym+2;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
}
{ /* Write .xdata section. */
uint16_t xdata[8+2+6];
PEreloc reloc;
xdata[0] = 0x01|0x08|0x10; /* Ver. 1, uhandler/ehandler, prolog size 0. */
xdata[1] = 0x0005; /* Number of unwind codes, no frame pointer. */
xdata[2] = 0x4200; /* Stack offset 4*8+8 = aword*5. */
xdata[3] = 0x3000; /* Push rbx. */
xdata[4] = 0x6000; /* Push rsi. */
xdata[5] = 0x7000; /* Push rdi. */
xdata[6] = 0x5000; /* Push rbp. */
xdata[7] = 0; /* Alignment. */
xdata[8] = xdata[9] = 0; /* Relocated address of exception handler. */
xdata[10] = 0x01; /* Ver. 1, no handler, prolog size 0. */
xdata[11] = 0x1504; /* Number of unwind codes, fp = rbp, fpofs = 16. */
xdata[12] = 0x0300; /* set_fpreg. */
xdata[13] = 0x0200; /* stack offset 0*8+8 = aword*1. */
xdata[14] = 0x3000; /* Push rbx. */
xdata[15] = 0x5000; /* Push rbp. */
owrite(ctx, &xdata, sizeof(xdata));
reloc.vaddr = 2*8; reloc.symidx = 1+2+nrsym+2+2;
reloc.type = PEOBJ_RELOC_ADDR32NB;
owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
}
#endif
/* Write .rdata$Z section. */
owrite(ctx, ctx->dasm_ident, strlen(ctx->dasm_ident)+1);
/* Write symbol table. */
strtab = NULL; /* 1st pass: collect string sizes. */
for (;;) {
strtabofs = 4;
/* Mark as SafeSEH compliant. */
emit_peobj_sym(ctx, "@feat.00", 1,
PEOBJ_SECT_ABS, PEOBJ_TYPE_NULL, PEOBJ_SCL_STATIC);
emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_TEXT);
for (i = 0; i < nrsym; i++)
emit_peobj_sym(ctx, ctx->relocsym[i], 0,
PEOBJ_SECT_UNDEF, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
#if LJ_TARGET_X64
emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_PDATA);
emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_XDATA);
emit_peobj_sym(ctx, "lj_err_unwind_win64", 0,
PEOBJ_SECT_UNDEF, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
#endif
emit_peobj_sym(ctx, ctx->beginsym, 0,
PEOBJ_SECT_TEXT, PEOBJ_TYPE_NULL, PEOBJ_SCL_EXTERN);
for (i = 0; i < ctx->nsym; i++)
emit_peobj_sym(ctx, ctx->sym[i].name, (uint32_t)ctx->sym[i].ofs,
PEOBJ_SECT_TEXT, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_RDATA_Z);
if (strtab)
break;
/* 2nd pass: alloc strtab, write syms and copy strings. */
strtab = (char *)malloc(strtabofs);
*(uint32_t *)strtab = (uint32_t)strtabofs;
}
/* Write string table. */
owrite(ctx, strtab, strtabofs);
}
/* Emit Windows PE object file. */
void emit_peobj(BuildCtx *ctx)
{
PEheader pehdr;
PEsection pesect[PEOBJ_NSECTIONS];
uint32_t sofs;
int i, nrsym;
union { uint8_t b; uint32_t u; } host_endian;
sofs = sizeof(PEheader) + PEOBJ_NSECTIONS*sizeof(PEsection);
/* Fill in PE sections. */
memset(&pesect, 0, PEOBJ_NSECTIONS*sizeof(PEsection));
memcpy(pesect[PEOBJ_SECT_TEXT].name, ".text", sizeof(".text")-1);
pesect[PEOBJ_SECT_TEXT].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_TEXT].size = (uint32_t)ctx->codesz);
pesect[PEOBJ_SECT_TEXT].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_TEXT].nreloc = (uint16_t)ctx->nreloc) * PEOBJ_RELOC_SIZE;
/* Flags: 60 = read+execute, 50 = align16, 20 = code. */
pesect[PEOBJ_SECT_TEXT].flags = PEOBJ_TEXT_FLAGS;
#if LJ_TARGET_X64
memcpy(pesect[PEOBJ_SECT_PDATA].name, ".pdata", sizeof(".pdata")-1);
pesect[PEOBJ_SECT_PDATA].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_PDATA].size = 6*4);
pesect[PEOBJ_SECT_PDATA].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_PDATA].nreloc = 6) * PEOBJ_RELOC_SIZE;
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_PDATA].flags = 0x40300040;
memcpy(pesect[PEOBJ_SECT_XDATA].name, ".xdata", sizeof(".xdata")-1);
pesect[PEOBJ_SECT_XDATA].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_XDATA].size = 8*2+4+6*2); /* See below. */
pesect[PEOBJ_SECT_XDATA].relocofs = sofs;
sofs += (pesect[PEOBJ_SECT_XDATA].nreloc = 1) * PEOBJ_RELOC_SIZE;
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_XDATA].flags = 0x40300040;
#endif
memcpy(pesect[PEOBJ_SECT_RDATA_Z].name, ".rdata$Z", sizeof(".rdata$Z")-1);
pesect[PEOBJ_SECT_RDATA_Z].ofs = sofs;
sofs += (pesect[PEOBJ_SECT_RDATA_Z].size = (uint32_t)strlen(ctx->dasm_ident)+1);
/* Flags: 40 = read, 30 = align4, 40 = initialized data. */
pesect[PEOBJ_SECT_RDATA_Z].flags = 0x40300040;
/* Fill in PE header. */
pehdr.arch = PEOBJ_ARCH_TARGET;
pehdr.nsects = PEOBJ_NSECTIONS;
pehdr.time = 0; /* Timestamp is optional. */
pehdr.symtabofs = sofs;
pehdr.opthdrsz = 0;
pehdr.flags = 0;
/* Compute the size of the symbol table:
** @feat.00 + nsections*2
** + asm_start + nsym
** + nrsym
*/
nrsym = ctx->nrelocsym;
pehdr.nsyms = 1+PEOBJ_NSECTIONS*2 + 1+ctx->nsym + nrsym;
#if LJ_TARGET_X64
pehdr.nsyms += 1; /* Symbol for lj_err_unwind_win64. */
#endif
/* Write PE object header and all sections. */
owrite(ctx, &pehdr, sizeof(PEheader));
owrite(ctx, &pesect, sizeof(PEsection)*PEOBJ_NSECTIONS);
/* Write .text section. */
host_endian.u = 1;
if (host_endian.b != LJ_ENDIAN_SELECT(1, 0)) {
#if LJ_TARGET_PPC
uint32_t *p = (uint32_t *)ctx->code;
int n = (int)(ctx->codesz >> 2);
for (i = 0; i < n; i++, p++)
*p = lj_bswap(*p); /* Byteswap .text section. */
#else
fprintf(stderr, "Error: different byte order for host and target\n");
exit(1);
#endif
}
