/* wind-back a syscall instruction */
void
pal_syscall_restart(thread_t thread __unused, x86_saved_state_t *state)
{
/* work out which flavour thread it is */
if( is_saved_state32(state) )
{
x86_saved_state32_t *regs32;
regs32 = saved_state32(state);
if (regs32->cs == SYSENTER_CS || regs32->cs == SYSENTER_TF_CS)
regs32->eip -= 5;
else
regs32->eip -= 2;
}
else
{
x86_saved_state64_t *regs64;
assert( is_saved_state64(state) );
regs64 = saved_state64(state);
/* Only one instruction for 64-bit threads */
regs64->isf.rip -= 2;
}
}
/*
* Copy values from saved_state to ts32.
*/
void
saved_state_to_thread_state32(const arm_saved_state_t *saved_state, arm_thread_state32_t *ts32)
{
uint32_t i;
assert(is_saved_state32(saved_state));
ts32->lr = (uint32_t)get_saved_state_lr(saved_state);
ts32->sp = (uint32_t)get_saved_state_sp(saved_state);
ts32->pc = (uint32_t)get_saved_state_pc(saved_state);
ts32->cpsr = get_saved_state_cpsr(saved_state);
for (i = 0; i < 13; i++)
ts32->r[i] = (uint32_t)get_saved_state_reg(saved_state, i);
}
/*
* Copy values from ts32 to saved_state.
*/
void
thread_state32_to_saved_state(const arm_thread_state32_t *ts32, arm_saved_state_t *saved_state)
{
uint32_t i;
assert(is_saved_state32(saved_state));
set_saved_state_lr(saved_state, ts32->lr);
set_saved_state_sp(saved_state, ts32->sp);
set_saved_state_pc(saved_state, ts32->pc);
#if defined(__arm64__)
set_saved_state_cpsr(saved_state, (ts32->cpsr & ~PSR64_MODE_MASK) | PSR64_MODE_RW_32);
#elif defined(__arm__)
set_saved_state_cpsr(saved_state, (ts32->cpsr & ~PSR_USER_MASK) | (ts32->cpsr & PSR_USER_MASK));
#else
#error Unknown architecture.
#endif
for (i = 0; i < 13; i++)
set_saved_state_reg(saved_state, i, ts32->r[i]);
}
/*
* Function: unix_syscall
*
* Inputs: regs - pointer to i386 save area
*
* Outputs: none
*/
void
unix_syscall(x86_saved_state_t *state)
{
thread_t thread;
void *vt;
unsigned int code;
struct sysent *callp;
int error;
vm_offset_t params;
struct proc *p;
struct uthread *uthread;
x86_saved_state32_t *regs;
boolean_t is_vfork;
assert(is_saved_state32(state));
regs = saved_state32(state);
#if DEBUG
if (regs->eax == 0x800)
thread_exception_return();
#endif
thread = current_thread();
uthread = get_bsdthread_info(thread);
/* Get the approriate proc; may be different from task's for vfork() */
is_vfork = uthread->uu_flag & UT_VFORK;
if (__improbable(is_vfork != 0))
p = current_proc();
else
p = (struct proc *)get_bsdtask_info(current_task());
/* Verify that we are not being called from a task without a proc */
if (__improbable(p == NULL)) {
regs->eax = EPERM;
regs->efl |= EFL_CF;
task_terminate_internal(current_task());
thread_exception_return();
/* NOTREACHED */
}
code = regs->eax & I386_SYSCALL_NUMBER_MASK;
DEBUG_KPRINT_SYSCALL_UNIX("unix_syscall: code=%d(%s) eip=%u\n",
code, syscallnames[code >= NUM_SYSENT ? 63 : code], (uint32_t)regs->eip);
params = (vm_offset_t) (regs->uesp + sizeof (int));
regs->efl &= ~(EFL_CF);
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
if (__improbable(callp == sysent)) {
code = fuword(params);
params += sizeof(int);
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
}
vt = (void *)uthread->uu_arg;
if (callp->sy_arg_bytes != 0) {
#if CONFIG_REQUIRES_U32_MUNGING
sy_munge_t *mungerp;
#else
#error U32 syscalls on x86_64 kernel requires munging
#endif
uint32_t nargs;
assert((unsigned) callp->sy_arg_bytes <= sizeof (uthread->uu_arg));
nargs = callp->sy_arg_bytes;
error = copyin((user_addr_t) params, (char *) vt, nargs);
if (error) {
regs->eax = error;
regs->efl |= EFL_CF;
thread_exception_return();
/* NOTREACHED */
}
if (__probable(code != 180)) {
int *ip = (int *)vt;
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
*ip, *(ip+1), *(ip+2), *(ip+3), 0);
}
#if CONFIG_REQUIRES_U32_MUNGING
mungerp = callp->sy_arg_munge32;
if (mungerp != NULL)
(*mungerp)(vt);
#endif
} else
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
0, 0, 0, 0, 0);
/*
* Delayed binding of thread credential to process credential, if we
* are not running with an explicitly set thread credential.
*/
kauth_cred_uthread_update(uthread, p);
uthread->uu_rval[0] = 0;
uthread->uu_rval[1] = 0;
uthread->uu_flag |= UT_NOTCANCELPT;
uthread->syscall_code = code;
#ifdef JOE_DEBUG
uthread->uu_iocount = 0;
uthread->uu_vpindex = 0;
#endif
AUDIT_SYSCALL_ENTER(code, p, uthread);
error = (*(callp->sy_call))((void *) p, (void *) vt, &(uthread->uu_rval[0]));
AUDIT_SYSCALL_EXIT(code, p, uthread, error);
#ifdef JOE_DEBUG
if (uthread->uu_iocount)
printf("system call returned with uu_iocount != 0\n");
#endif
#if CONFIG_DTRACE
uthread->t_dtrace_errno = error;
#endif /* CONFIG_DTRACE */
if (__improbable(error == ERESTART)) {
/*
* Move the user's pc back to repeat the syscall:
* 5 bytes for a sysenter, or 2 for an int 8x.
* The SYSENTER_TF_CS covers single-stepping over a sysenter
* - see debug trap handler in idt.s/idt64.s
*/
pal_syscall_restart(thread, state);
}
else if (error != EJUSTRETURN) {
if (__improbable(error)) {
regs->eax = error;
regs->efl |= EFL_CF; /* carry bit */
} else { /* (not error) */
/*
* We split retval across two registers, in case the
* syscall had a 64-bit return value, in which case
* eax/edx matches the function call ABI.
*/
regs->eax = uthread->uu_rval[0];
regs->edx = uthread->uu_rval[1];
}
}
DEBUG_KPRINT_SYSCALL_UNIX(
"unix_syscall: error=%d retval=(%u,%u)\n",
error, regs->eax, regs->edx);
uthread->uu_flag &= ~UT_NOTCANCELPT;
if (__improbable(uthread->uu_lowpri_window)) {
/*
* task is marked as a low priority I/O type
* and the I/O we issued while in this system call
* collided with normal I/O operations... we'll
* delay in order to mitigate the impact of this
* task on the normal operation of the system
*/
throttle_lowpri_io(1);
}
if (__probable(code != 180))
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
if (__improbable(!is_vfork && callp->sy_call == (sy_call_t *)execve && !error)) {
pal_execve_return(thread);
}
thread_exception_return();
/* NOTREACHED */
}
void
mach_call_munger(x86_saved_state_t *state)
{
int argc;
int call_number;
mach_call_t mach_call;
kern_return_t retval;
struct mach_call_args args = { 0, 0, 0, 0, 0, 0, 0, 0, 0 };
x86_saved_state32_t *regs;
#if PROC_REF_DEBUG
struct uthread *ut = get_bsdthread_info(current_thread());
uthread_reset_proc_refcount(ut);
#endif
assert(is_saved_state32(state));
regs = saved_state32(state);
call_number = -(regs->eax);
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: code=%d(%s)\n",
call_number, mach_syscall_name_table[call_number]);
#if DEBUG_TRACE
kprintf("mach_call_munger(0x%08x) code=%d\n", regs, call_number);
#endif
if (call_number < 0 || call_number >= mach_trap_count) {
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
mach_call = (mach_call_t)mach_trap_table[call_number].mach_trap_function;
if (mach_call == (mach_call_t)kern_invalid) {
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: kern_invalid 0x%x\n", regs->eax);
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
argc = mach_trap_table[call_number].mach_trap_arg_count;
if (argc) {
retval = mach_call_arg_munger32(regs->uesp, &args, &mach_trap_table[call_number]);
if (retval != KERN_SUCCESS) {
regs->eax = retval;
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: retval=0x%x\n", retval);
thread_exception_return();
/* NOTREACHED */
}
}
#ifdef MACH_BSD
mach_kauth_cred_uthread_update();
#endif
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC, (call_number)) | DBG_FUNC_START,
args.arg1, args.arg2, args.arg3, args.arg4, 0);
retval = mach_call(&args);
DEBUG_KPRINT_SYSCALL_MACH("mach_call_munger: retval=0x%x\n", retval);
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC,(call_number)) | DBG_FUNC_END,
retval, 0, 0, 0, 0);
regs->eax = retval;
throttle_lowpri_io(1);
#if PROC_REF_DEBUG
if (__improbable(uthread_get_proc_refcount(ut) != 0)) {
panic("system call returned with uu_proc_refcount != 0");
}
#endif
thread_exception_return();
/* NOTREACHED */
}
void
machdep_syscall(x86_saved_state_t *state)
{
int args[machdep_call_count];
int trapno;
int nargs;
const machdep_call_t *entry;
x86_saved_state32_t *regs;
assert(is_saved_state32(state));
regs = saved_state32(state);
trapno = regs->eax;
#if DEBUG_TRACE
kprintf("machdep_syscall(0x%08x) code=%d\n", regs, trapno);
#endif
DEBUG_KPRINT_SYSCALL_MDEP(
"machdep_syscall: trapno=%d\n", trapno);
if (trapno < 0 || trapno >= machdep_call_count) {
regs->eax = (unsigned int)kern_invalid(NULL);
thread_exception_return();
/* NOTREACHED */
}
entry = &machdep_call_table[trapno];
nargs = entry->nargs;
if (nargs != 0) {
if (copyin((user_addr_t) regs->uesp + sizeof (int),
(char *) args, (nargs * sizeof (int)))) {
regs->eax = KERN_INVALID_ADDRESS;
thread_exception_return();
/* NOTREACHED */
}
}
switch (nargs) {
case 0:
regs->eax = (*entry->routine.args_0)();
break;
case 1:
regs->eax = (*entry->routine.args_1)(args[0]);
break;
case 2:
regs->eax = (*entry->routine.args_2)(args[0],args[1]);
break;
case 3:
if (!entry->bsd_style)
regs->eax = (*entry->routine.args_3)(args[0],args[1],args[2]);
else {
int error;
uint32_t rval;
error = (*entry->routine.args_bsd_3)(&rval, args[0], args[1], args[2]);
if (error) {
regs->eax = error;
regs->efl |= EFL_CF; /* carry bit */
} else {
regs->eax = rval;
regs->efl &= ~EFL_CF;
}
}
break;
case 4:
regs->eax = (*entry->routine.args_4)(args[0], args[1], args[2], args[3]);
break;
default:
panic("machdep_syscall: too many args");
}
DEBUG_KPRINT_SYSCALL_MDEP("machdep_syscall: retval=%u\n", regs->eax);
throttle_lowpri_io(1);
thread_exception_return();
/* NOTREACHED */
}
void
mach_call_munger(x86_saved_state_t *state)
{
int argc;
int call_number;
mach_call_t mach_call;
kern_return_t retval;
struct mach_call_args args = { 0, 0, 0, 0, 0, 0, 0, 0, 0 };
x86_saved_state32_t *regs;
assert(is_saved_state32(state));
regs = saved_state32(state);
call_number = -(regs->eax);
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: code=%d(%s)\n",
call_number, mach_syscall_name_table[call_number]);
#if DEBUG_TRACE
kprintf("mach_call_munger(0x%08x) code=%d\n", regs, call_number);
#endif
if (call_number < 0 || call_number >= mach_trap_count) {
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
mach_call = (mach_call_t)mach_trap_table[call_number].mach_trap_function;
if (mach_call == (mach_call_t)kern_invalid) {
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: kern_invalid 0x%x\n", regs->eax);
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
argc = mach_trap_table[call_number].mach_trap_arg_count;
if (argc) {
retval = mach_call_arg_munger32(regs->uesp, argc, call_number, &args);
if (retval != KERN_SUCCESS) {
regs->eax = retval;
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: retval=0x%x\n", retval);
thread_exception_return();
/* NOTREACHED */
}
}
#ifdef MACH_BSD
mach_kauth_cred_uthread_update();
#endif
KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_EXCP_SC, (call_number)) | DBG_FUNC_START,
args.arg1, args.arg2, args.arg3, args.arg4, 0);
retval = mach_call(&args);
DEBUG_KPRINT_SYSCALL_MACH("mach_call_munger: retval=0x%x\n", retval);
KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_EXCP_SC,(call_number)) | DBG_FUNC_END,
retval, 0, 0, 0, 0);
regs->eax = retval;
throttle_lowpri_io(TRUE);
thread_exception_return();
/* NOTREACHED */
}
int
diagCall(x86_saved_state_t * state)
{
uint32_t stk, curpos, i, j;
uint32_t selector, data;
int err;
uint64_t currNap, durNap;
x86_saved_state32_t *regs;
assert(is_saved_state32(state));
regs = saved_state32(state);
if (!(dgWork.dgFlags & enaDiagSCs))
return 0; /* If not enabled, cause an exception */
stk = regs->uesp; /* Point to the stack */
err = copyin((user_addr_t) (stk + 4), (char *) &selector, sizeof(uint32_t)); /* Get the selector */
if (err) {
return 0; /* Failed to fetch stack */
}
switch (selector) { /* Select the routine */
case dgRuptStat: /* Suck Interruption statistics */
err = copyin((user_addr_t) (stk + 8), (char *) &data, sizeof(uint32_t)); /* Get the selector */
if (data == 0) {/* If number of processors is 0, clear all
* counts */
for (i = 0; i < real_ncpus; i++) { /* Cycle through
* processors */
for (j = 0; j < 256; j++)
cpu_data_ptr[i]->cpu_hwIntCnt[j] = 0;
}
lastRuptClear = mach_absolute_time(); /* Get the time of clear */
return 1; /* Normal return */
}
(void) copyout((char *) &real_ncpus, data, sizeof(real_ncpus)); /* Copy out number of
* processors */
currNap = mach_absolute_time(); /* Get the time now */
durNap = currNap - lastRuptClear; /* Get the last interval
* duration */
if (durNap == 0)
durNap = 1; /* This is a very short time, make it
* bigger */
curpos = data + sizeof(real_ncpus); /* Point to the next
* available spot */
for (i = 0; i < real_ncpus; i++) { /* Move 'em all out */
(void) copyout((char *) &durNap, curpos, 8); /* Copy out the time
* since last clear */
(void) copyout((char *) &cpu_data_ptr[i]->cpu_hwIntCnt, curpos + 8, 256 * sizeof(uint32_t)); /* Copy out interrupt
* data for this
* processor */
curpos = curpos + (256 * sizeof(uint32_t) + 8); /* Point to next out put
* slot */
}
break;
default: /* Handle invalid ones */
return 0; /* Return an exception */
}
return 1; /* Normal non-ast check return */
}
