xScopedStackFrame::xScopedStackFrame(bool base_frame, bool save_base_pointer, int offset)
{
m_base_frame = base_frame;
m_save_base_pointer = save_base_pointer;
m_offset = offset;
#ifdef __x86_64__
m_offset += 8; // Call stores the return address (4 bytes)
// Note rbp can surely be optimized in 64 bits
if (m_base_frame) {
xPUSH( rbp );
xMOV( rbp, rsp );
m_offset += 8;
} else if (m_save_base_pointer) {
xPUSH( rbp );
m_offset += 8;
}
xPUSH( rbx );
xPUSH( r12 );
xPUSH( r13 );
xPUSH( r14 );
xPUSH( r15 );
m_offset += 40;
#else
m_offset += 4; // Call stores the return address (4 bytes)
// Create a new frame
if (m_base_frame) {
xPUSH( ebp );
xMOV( ebp, esp );
m_offset += 4;
} else if (m_save_base_pointer) {
xPUSH( ebp );
m_offset += 4;
}
// Save the register context
xPUSH( edi );
xPUSH( esi );
xPUSH( ebx );
m_offset += 12;
#endif
ALIGN_STACK(-(16 - m_offset % 16));
}
xScopedStackFrame::~xScopedStackFrame()
{
ALIGN_STACK(16 - m_offset % 16);
#ifdef __x86_64__
// Restore the register context
xPOP( r15 );
xPOP( r14 );
xPOP( r13 );
xPOP( r12 );
xPOP( rbx );
// Destroy the frame
if (m_base_frame) {
xLEAVE();
} else if (m_save_base_pointer) {
xPOP( rbp );
}
#else
// Restore the register context
xPOP( ebx );
xPOP( esi );
xPOP( edi );
// Destroy the frame
if (m_base_frame) {
xLEAVE();
} else if (m_save_base_pointer) {
xPOP( ebp );
}
#endif
}
static void ffi_call_x86_64(ktap_state_t *ks, csymbol_func *csf, void *rvalue)
{
int i;
int gpr_nr;
int arg_bytes; /* total bytes needed for exceeded args in stack */
int mem_bytes; /* total bytes needed for memory storage */
char *stack, *stack_p, *gpr_p, *arg_p, *mem_p, *tmp_p;
int arg_nr;
csymbol *rsym;
ffi_type rtype;
size_t rsize;
bool ret_in_memory;
/* New stack to call C function */
char space[NEWSTACK_SIZE];
arg_nr = kp_arg_nr(ks);
rsym = csymf_ret(ks, csf);
rtype = csym_type(rsym);
rsize = csym_size(ks, rsym);
ret_in_memory = false;
if (rtype == FFI_STRUCT || rtype == FFI_UNION) {
if (rsize > 16) {
rvalue = kp_malloc(ks, rsize);
rtype = FFI_VOID;
ret_in_memory = true;
} else {
/* much easier to always copy 16 bytes from registers */
rvalue = kp_malloc(ks, 16);
}
}
gpr_nr = 0;
arg_bytes = mem_bytes = 0;
if (ret_in_memory)
gpr_nr++;
/* calculate bytes needed for stack */
for (i = 0; i < arg_nr; i++) {
csymbol *cs = ffi_get_arg_csym(ks, csf, i);
size_t size = csym_size(ks, cs);
size_t align = csym_align(ks, cs);
enum arg_status st = IN_REGISTER;
int n_gpr_nr = 0;
if (size > 32) {
st = IN_MEMORY;
n_gpr_nr = 1;
} else if (size > 16)
st = IN_STACK;
else
n_gpr_nr = ALIGN(size, GPR_SIZE) / GPR_SIZE;
if (gpr_nr + n_gpr_nr > MAX_GPR) {
if (st == IN_MEMORY)
arg_bytes += GPR_SIZE;
else
st = IN_STACK;
} else
gpr_nr += n_gpr_nr;
if (st == IN_STACK) {
arg_bytes = ALIGN(arg_bytes, align);
arg_bytes += size;
arg_bytes = ALIGN(arg_bytes, STACK_ALIGNMENT);
}
if (st == IN_MEMORY) {
mem_bytes = ALIGN(mem_bytes, align);
mem_bytes += size;
mem_bytes = ALIGN(mem_bytes, STACK_ALIGNMENT);
}
}
/* apply space to fake stack for C function call */
if (16 + REDZONE_SIZE + MAX_GPR_SIZE + arg_bytes +
mem_bytes + 6 * 8 >= NEWSTACK_SIZE) {
kp_error(ks, "Unable to handle that many arguments by now\n");
return;
}
stack = space;
/* 128 bytes below %rsp is red zone */
/* stack should be 16-bytes aligned */
stack_p = ALIGN_STACK(stack + REDZONE_SIZE, 16);
/* save general purpose registers here */
gpr_p = stack_p;
memset(gpr_p, 0, MAX_GPR_SIZE);
/* save arguments in stack here */
arg_p = gpr_p + MAX_GPR_SIZE;
/* save arguments in memory here */
mem_p = arg_p + arg_bytes;
/* set additional space as temporary space */
tmp_p = mem_p + mem_bytes;
/* copy arguments here */
gpr_nr = 0;
if (ret_in_memory) {
memcpy(gpr_p, &rvalue, GPR_SIZE);
gpr_p += GPR_SIZE;
gpr_nr++;
}
for (i = 0; i < arg_nr; i++) {
csymbol *cs = ffi_get_arg_csym(ks, csf, i);
size_t size = csym_size(ks, cs);
size_t align = csym_align(ks, cs);
enum arg_status st = IN_REGISTER;
int n_gpr_nr = 0;
if (size > 32) {
st = IN_MEMORY;
n_gpr_nr = 1;
} else if (size > 16)
st = IN_STACK;
else
n_gpr_nr = ALIGN(size, GPR_SIZE) / GPR_SIZE;
if (st == IN_MEMORY)
mem_p = ALIGN_STACK(mem_p, align);
/* Tricky way about storing it above mem_p. It won't overflow
* because temp region can be temporarily used if necesseary. */
ffi_unpack(ks, mem_p, csf, i, GPR_SIZE);
if (gpr_nr + n_gpr_nr > MAX_GPR) {
if (st == IN_MEMORY) {
memcpy(arg_p, &mem_p, GPR_SIZE);
arg_p += GPR_SIZE;
} else
st = IN_STACK;
} else {
memcpy(gpr_p, mem_p, n_gpr_nr * GPR_SIZE);
gpr_p += n_gpr_nr * GPR_SIZE;
gpr_nr += n_gpr_nr;
}
if (st == IN_STACK) {
arg_p = ALIGN_STACK(arg_p, align);
memcpy(arg_p, mem_p, size);
arg_p += size;
arg_p = ALIGN_STACK(arg_p, STACK_ALIGNMENT);
}
if (st == IN_MEMORY) {
mem_p += size;
mem_p = ALIGN_STACK(mem_p, STACK_ALIGNMENT);
}
}
kp_verbose_printf(ks, "Stack location: %p -redzone- %p -general purpose "
"register used- %p -zero- %p -stack for argument- %p"
" -memory for argument- %p -temp stack-\n",
stack, stack_p, gpr_p, stack_p + MAX_GPR_SIZE,
arg_p, mem_p);
kp_verbose_printf(ks, "GPR number: %d; arg in stack: %d; "
"arg in mem: %d\n",
gpr_nr, arg_bytes, mem_bytes);
kp_verbose_printf(ks, "Return: address %p type %d\n", rvalue, rtype);
kp_verbose_printf(ks, "Number of register used: %d\n", gpr_nr);
kp_verbose_printf(ks, "Start FFI call on %p\n", csf->addr);
ffi_call_assem_x86_64(stack_p, tmp_p, csf->addr, rvalue, rtype);
}