void
ipi_process(struct cpu_info *ci, uint64_t ipi_mask)
{
KASSERT(cpu_intr_p());
if (ipi_mask & __BIT(IPI_NOP)) {
ci->ci_evcnt_per_ipi[IPI_NOP].ev_count++;
ipi_nop(ci);
}
if (ipi_mask & __BIT(IPI_AST)) {
ci->ci_evcnt_per_ipi[IPI_AST].ev_count++;
ipi_nop(ci);
}
if (ipi_mask & __BIT(IPI_SHOOTDOWN)) {
ci->ci_evcnt_per_ipi[IPI_SHOOTDOWN].ev_count++;
ipi_shootdown(ci);
}
if (ipi_mask & __BIT(IPI_SYNCICACHE)) {
ci->ci_evcnt_per_ipi[IPI_SYNCICACHE].ev_count++;
ipi_syncicache(ci);
}
if (ipi_mask & __BIT(IPI_SUSPEND)) {
ci->ci_evcnt_per_ipi[IPI_SUSPEND].ev_count++;
cpu_pause(NULL);
}
if (ipi_mask & __BIT(IPI_HALT)) {
ci->ci_evcnt_per_ipi[IPI_HALT].ev_count++;
ipi_halt();
}
if (ipi_mask & __BIT(IPI_XCALL)) {
ci->ci_evcnt_per_ipi[IPI_XCALL].ev_count++;
xc_ipi_handler();
}
}
int
pserialize_read_enter(void)
{
KASSERT(!cpu_intr_p());
return splsoftserial();
}
void
intr_disestablish(void *ih)
{
struct intrsource * const is = ih;
KASSERT(!cpu_intr_p());
KASSERT(!cpu_softintr_p());
pic_disestablish_source(is);
}
/*
* callout_halt:
*
* Cancel a pending callout. If in-flight, block until it completes.
* May not be called from a hard interrupt handler. If the callout
* can take locks, the caller of callout_halt() must not hold any of
* those locks, otherwise the two could deadlock. If 'interlock' is
* non-NULL and we must wait for the callout to complete, it will be
* released and re-acquired before returning.
*/
bool
callout_halt(callout_t *cs, void *interlock)
{
callout_impl_t *c = (callout_impl_t *)cs;
struct callout_cpu *cc;
struct lwp *l;
kmutex_t *lock, *relock;
bool expired;
KASSERT(c->c_magic == CALLOUT_MAGIC);
KASSERT(!cpu_intr_p());
lock = callout_lock(c);
relock = NULL;
expired = ((c->c_flags & CALLOUT_FIRED) != 0);
if ((c->c_flags & CALLOUT_PENDING) != 0)
CIRCQ_REMOVE(&c->c_list);
c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED);
l = curlwp;
for (;;) {
cc = c->c_cpu;
if (__predict_true(cc->cc_active != c || cc->cc_lwp == l))
break;
if (interlock != NULL) {
/*
* Avoid potential scheduler lock order problems by
* dropping the interlock without the callout lock
* held.
*/
mutex_spin_exit(lock);
mutex_exit(interlock);
relock = interlock;
interlock = NULL;
} else {
/* XXX Better to do priority inheritance. */
KASSERT(l->l_wchan == NULL);
cc->cc_nwait++;
cc->cc_ev_block.ev_count++;
l->l_kpriority = true;
sleepq_enter(&cc->cc_sleepq, l, cc->cc_lock);
sleepq_enqueue(&cc->cc_sleepq, cc, "callout",
&sleep_syncobj);
sleepq_block(0, false);
}
lock = callout_lock(c);
}
mutex_spin_exit(lock);
if (__predict_false(relock != NULL))
mutex_enter(relock);
return expired;
}
/*
* pcu_save_lwp: save PCU state to the given LWP.
*/
void
pcu_save(const pcu_ops_t *pcu)
{
const u_int id = pcu->pcu_id;
lwp_t * const l = curlwp;
KASSERT(!cpu_intr_p() && !cpu_softintr_p());
if (__predict_true(l->l_pcu_cpu[id] == NULL)) {
return;
}
pcu_lwp_op(pcu, l, PCU_SAVE | PCU_RELEASE);
}
/*
* pcu_load: load/initialize the PCU state of current LWP on current CPU.
*/
void
pcu_load(const pcu_ops_t *pcu)
{
const u_int id = pcu->pcu_id;
struct cpu_info *ci, *curci;
lwp_t * const l = curlwp;
uint64_t where;
int s;
KASSERT(!cpu_intr_p() && !cpu_softintr_p());
s = splsoftclock();
curci = curcpu();
ci = l->l_pcu_cpu[id];
/* Does this CPU already have our PCU state loaded? */
if (ci == curci) {
KASSERT(curci->ci_pcu_curlwp[id] == l);
pcu->pcu_state_load(l, PCU_ENABLE); /* Re-enable */
splx(s);
return;
}
/* If PCU state of this LWP is on the remote CPU - save it there. */
if (ci) {
splx(s);
/* Note: there is a race; see description in the top. */
where = xc_unicast(XC_HIGHPRI, (xcfunc_t)pcu_cpu_op,
__UNCONST(pcu), (void *)(PCU_SAVE | PCU_RELEASE), ci);
xc_wait(where);
/* Enter IPL_SOFTCLOCK and re-fetch the current CPU. */
s = splsoftclock();
curci = curcpu();
}
KASSERT(l->l_pcu_cpu[id] == NULL);
/* Save the PCU state on the current CPU, if there is any. */
pcu_cpu_op(pcu, PCU_SAVE | PCU_RELEASE);
KASSERT(curci->ci_pcu_curlwp[id] == NULL);
/*
* Finally, load the state for this LWP on this CPU. Indicate to
* load function whether PCU was used before. Note the usage.
*/
pcu_do_op(pcu, l, PCU_CLAIM | PCU_ENABLE | PCU_RELOAD);
splx(s);
}
/*
* pserialize_perform:
*
* Perform the write side of passive serialization. The calling
* thread holds an exclusive lock on the data object(s) being updated.
* We wait until every processor in the system has made at least two
* passes through cpu_swichto(). The wait is made with the caller's
* update lock held, but is short term.
*/
void
pserialize_perform(pserialize_t psz)
{
uint64_t xc;
KASSERT(!cpu_intr_p());
KASSERT(!cpu_softintr_p());
if (__predict_false(panicstr != NULL)) {
return;
}
KASSERT(psz->psz_owner == NULL);
KASSERT(ncpu > 0);
/*
* Set up the object and put it onto the queue. The lock
* activity here provides the necessary memory barrier to
* make the caller's data update completely visible to
* other processors.
*/
psz->psz_owner = curlwp;
kcpuset_copy(psz->psz_target, kcpuset_running);
kcpuset_zero(psz->psz_pass);
mutex_spin_enter(&psz_lock);
TAILQ_INSERT_TAIL(&psz_queue0, psz, psz_chain);
psz_work_todo++;
do {
mutex_spin_exit(&psz_lock);
/*
* Force some context switch activity on every CPU, as
* the system may not be busy. Pause to not flood.
*/
xc = xc_broadcast(XC_HIGHPRI, (xcfunc_t)nullop, NULL, NULL);
xc_wait(xc);
kpause("psrlz", false, 1, NULL);
mutex_spin_enter(&psz_lock);
} while (!kcpuset_iszero(psz->psz_target));
psz_ev_excl.ev_count++;
mutex_spin_exit(&psz_lock);
psz->psz_owner = NULL;
}
/*
* pcu_discard: discard the PCU state of current LWP.
* If the "usesw" flag is set, pcu_used_p() will return "true".
*/
void
pcu_discard(const pcu_ops_t *pcu, bool usesw)
{
const u_int id = pcu->pcu_id;
lwp_t * const l = curlwp;
KASSERT(!cpu_intr_p() && !cpu_softintr_p());
if (usesw)
l->l_pcu_used[PCU_USER] |= (1 << id);
else
l->l_pcu_used[PCU_USER] &= ~(1 << id);
if (__predict_true(l->l_pcu_cpu[id] == NULL)) {
return;
}
pcu_lwp_op(pcu, l, PCU_RELEASE);
}
void *
intr_establish(int irq, int ipl, int type, int (*func)(void *), void *arg)
{
KASSERT(!cpu_intr_p());
KASSERT(!cpu_softintr_p());
for (size_t slot = 0; slot < PIC_MAXPICS; slot++) {
struct pic_softc * const pic = pic_list[slot];
if (pic == NULL || pic->pic_irqbase < 0)
continue;
if (pic->pic_irqbase <= irq
&& irq < pic->pic_irqbase + pic->pic_maxsources) {
return pic_establish_intr(pic, irq - pic->pic_irqbase,
ipl, type, func, arg);
}
}
return NULL;
}
/*
* rw_vector_enter:
*
* Acquire a rwlock.
*/
void
rw_vector_enter(krwlock_t *rw, const krw_t op)
{
uintptr_t owner, incr, need_wait, set_wait, curthread, next;
turnstile_t *ts;
int queue;
lwp_t *l;
LOCKSTAT_TIMER(slptime);
LOCKSTAT_TIMER(slpcnt);
LOCKSTAT_TIMER(spintime);
LOCKSTAT_COUNTER(spincnt);
LOCKSTAT_FLAG(lsflag);
l = curlwp;
curthread = (uintptr_t)l;
RW_ASSERT(rw, !cpu_intr_p());
RW_ASSERT(rw, curthread != 0);
RW_WANTLOCK(rw, op);
if (panicstr == NULL) {
LOCKDEBUG_BARRIER(&kernel_lock, 1);
}
/*
* We play a slight trick here. If we're a reader, we want
* increment the read count. If we're a writer, we want to
* set the owner field and whe WRITE_LOCKED bit.
*
* In the latter case, we expect those bits to be zero,
* therefore we can use an add operation to set them, which
* means an add operation for both cases.
*/
if (__predict_true(op == RW_READER)) {
incr = RW_READ_INCR;
set_wait = RW_HAS_WAITERS;
need_wait = RW_WRITE_LOCKED | RW_WRITE_WANTED;
queue = TS_READER_Q;
} else {
RW_DASSERT(rw, op == RW_WRITER);
incr = curthread | RW_WRITE_LOCKED;
set_wait = RW_HAS_WAITERS | RW_WRITE_WANTED;
need_wait = RW_WRITE_LOCKED | RW_THREAD;
queue = TS_WRITER_Q;
}
LOCKSTAT_ENTER(lsflag);
KPREEMPT_DISABLE(curlwp);
for (owner = rw->rw_owner; ;) {
/*
* Read the lock owner field. If the need-to-wait
* indicator is clear, then try to acquire the lock.
*/
if ((owner & need_wait) == 0) {
next = rw_cas(rw, owner, (owner + incr) &
~RW_WRITE_WANTED);
if (__predict_true(next == owner)) {
/* Got it! */
membar_enter();
break;
}
/*
* Didn't get it -- spin around again (we'll
* probably sleep on the next iteration).
*/
owner = next;
continue;
}
if (__predict_false(panicstr != NULL)) {
kpreempt_enable();
return;
}
if (__predict_false(RW_OWNER(rw) == curthread)) {
rw_abort(rw, __func__, "locking against myself");
}
/*
* If the lock owner is running on another CPU, and
* there are no existing waiters, then spin.
*/
if (rw_oncpu(owner)) {
LOCKSTAT_START_TIMER(lsflag, spintime);
u_int count = SPINLOCK_BACKOFF_MIN;
do {
KPREEMPT_ENABLE(curlwp);
SPINLOCK_BACKOFF(count);
KPREEMPT_DISABLE(curlwp);
owner = rw->rw_owner;
} while (rw_oncpu(owner));
LOCKSTAT_STOP_TIMER(lsflag, spintime);
LOCKSTAT_COUNT(spincnt, 1);
if ((owner & need_wait) == 0)
continue;
}
/*
* Grab the turnstile chain lock. Once we have that, we
* can adjust the waiter bits and sleep queue.
*/
ts = turnstile_lookup(rw);
/*
* Mark the rwlock as having waiters. If the set fails,
* then we may not need to sleep and should spin again.
* Reload rw_owner because turnstile_lookup() may have
* spun on the turnstile chain lock.
*/
owner = rw->rw_owner;
if ((owner & need_wait) == 0 || rw_oncpu(owner)) {
turnstile_exit(rw);
continue;
}
next = rw_cas(rw, owner, owner | set_wait);
if (__predict_false(next != owner)) {
turnstile_exit(rw);
owner = next;
continue;
}
LOCKSTAT_START_TIMER(lsflag, slptime);
turnstile_block(ts, queue, rw, &rw_syncobj);
LOCKSTAT_STOP_TIMER(lsflag, slptime);
LOCKSTAT_COUNT(slpcnt, 1);
/*
* No need for a memory barrier because of context switch.
* If not handed the lock, then spin again.
*/
if (op == RW_READER || (rw->rw_owner & RW_THREAD) == curthread)
break;
owner = rw->rw_owner;
}
KPREEMPT_ENABLE(curlwp);
LOCKSTAT_EVENT(lsflag, rw, LB_RWLOCK |
(op == RW_WRITER ? LB_SLEEP1 : LB_SLEEP2), slpcnt, slptime);
LOCKSTAT_EVENT(lsflag, rw, LB_RWLOCK | LB_SPIN, spincnt, spintime);
LOCKSTAT_EXIT(lsflag);
RW_DASSERT(rw, (op != RW_READER && RW_OWNER(rw) == curthread) ||
(op == RW_READER && RW_COUNT(rw) != 0));
RW_LOCKED(rw, op);
}
/*
* trap(frame): exception, fault, and trap interface to BSD kernel.
*
* This common code is called from assembly language IDT gate entry routines
* that prepare a suitable stack frame, and restore this frame after the
* exception has been processed. Note that the effect is as if the arguments
* were passed call by reference.
*/
void
trap(struct trapframe *frame)
{
struct lwp *l = curlwp;
struct proc *p;
struct pcb *pcb;
extern char fusubail[], kcopy_fault[], return_address_fault[],
IDTVEC(osyscall)[];
struct trapframe *vframe;
ksiginfo_t ksi;
void *onfault;
int type, error;
uint32_t cr2;
bool pfail;
if (__predict_true(l != NULL)) {
pcb = lwp_getpcb(l);
p = l->l_proc;
} else {
/*
* this can happen eg. on break points in early on boot.
*/
pcb = NULL;
p = NULL;
}
type = frame->tf_trapno;
#ifdef DEBUG
if (trapdebug) {
trap_print(frame, l);
}
#endif
if (type != T_NMI &&
!KERNELMODE(frame->tf_cs, frame->tf_eflags)) {
type |= T_USER;
l->l_md.md_regs = frame;
pcb->pcb_cr2 = 0;
LWP_CACHE_CREDS(l, p);
}
#ifdef KDTRACE_HOOKS
/*
* A trap can occur while DTrace executes a probe. Before
* executing the probe, DTrace blocks re-scheduling and sets
* a flag in its per-cpu flags to indicate that it doesn't
* want to fault. On returning from the the probe, the no-fault
* flag is cleared and finally re-scheduling is enabled.
*
* If the DTrace kernel module has registered a trap handler,
* call it and if it returns non-zero, assume that it has
* handled the trap and modified the trap frame so that this
* function can return normally.
*/
if ((type == T_PROTFLT || type == T_PAGEFLT) &&
dtrace_trap_func != NULL) {
if ((*dtrace_trap_func)(frame, type)) {
return;
}
}
#endif
switch (type) {
case T_ASTFLT:
/*FALLTHROUGH*/
default:
we_re_toast:
if (type == T_TRCTRAP)
check_dr0();
else
trap_print(frame, l);
if (kdb_trap(type, 0, frame))
return;
if (kgdb_trap(type, frame))
return;
/*
* If this is a breakpoint, don't panic if we're not connected.
*/
if (type == T_BPTFLT && kgdb_disconnected()) {
printf("kgdb: ignored %s\n", trap_type[type]);
return;
}
panic("trap");
/*NOTREACHED*/
case T_PROTFLT:
case T_SEGNPFLT:
case T_ALIGNFLT:
case T_TSSFLT:
if (p == NULL)
goto we_re_toast;
/* Check for copyin/copyout fault. */
onfault = onfault_handler(pcb, frame);
if (onfault != NULL) {
copyefault:
error = EFAULT;
copyfault:
frame->tf_eip = (uintptr_t)onfault;
frame->tf_eax = error;
return;
}
/*
* Check for failure during return to user mode.
* This can happen loading invalid values into the segment
* registers, or during the 'iret' itself.
*
* We do this by looking at the instruction we faulted on.
* The specific instructions we recognize only happen when
* returning from a trap, syscall, or interrupt.
*/
kernelfault:
KSI_INIT_TRAP(&ksi);
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_ACCERR;
ksi.ksi_trap = type;
switch (*(u_char *)frame->tf_eip) {
case 0xcf: /* iret */
/*
* The 'iret' instruction faulted, so we have the
* 'user' registers saved after the kernel %eip:%cs:%fl
* of the 'iret' and below that the user %eip:%cs:%fl
* the 'iret' was processing.
* We must delete the 3 words of kernel return address
* from the stack to generate a normal stack frame
* (eg for sending a SIGSEGV).
*/
vframe = (void *)((int *)frame + 3);
if (KERNELMODE(vframe->tf_cs, vframe->tf_eflags))
goto we_re_toast;
memmove(vframe, frame,
offsetof(struct trapframe, tf_eip));
/* Set the faulting address to the user %eip */
ksi.ksi_addr = (void *)vframe->tf_eip;
break;
case 0x8e:
switch (*(uint32_t *)frame->tf_eip) {
case 0x8e242c8e: /* mov (%esp,%gs), then */
case 0x0424648e: /* mov 0x4(%esp),%fs */
case 0x0824448e: /* mov 0x8(%esp),%es */
case 0x0c245c8e: /* mov 0xc(%esp),%ds */
break;
default:
goto we_re_toast;
}
/*
* We faulted loading one if the user segment registers.
* The stack frame containing the user registers is
* still valid and is just below the %eip:%cs:%fl of
* the kernel fault frame.
*/
vframe = (void *)(&frame->tf_eflags + 1);
if (KERNELMODE(vframe->tf_cs, vframe->tf_eflags))
goto we_re_toast;
/* There is no valid address for the fault */
break;
default:
goto we_re_toast;
}
/*
* We might have faulted trying to execute the
* trampoline for a local (nested) signal handler.
* Only generate SIGSEGV if the user %cs isn't changed.
* (This is only strictly necessary in the 'iret' case.)
*/
if (!pmap_exec_fixup(&p->p_vmspace->vm_map, vframe, pcb)) {
/* Save outer frame for any signal return */
l->l_md.md_regs = vframe;
(*p->p_emul->e_trapsignal)(l, &ksi);
}
/* Return to user by reloading the user frame */
trap_return_fault_return(vframe);
/* NOTREACHED */
case T_PROTFLT|T_USER: /* protection fault */
case T_TSSFLT|T_USER:
case T_SEGNPFLT|T_USER:
case T_STKFLT|T_USER:
case T_ALIGNFLT|T_USER:
KSI_INIT_TRAP(&ksi);
ksi.ksi_addr = (void *)rcr2();
switch (type) {
case T_SEGNPFLT|T_USER:
case T_STKFLT|T_USER:
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_ADRERR;
break;
case T_TSSFLT|T_USER:
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_OBJERR;
break;
case T_ALIGNFLT|T_USER:
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_ADRALN;
break;
case T_PROTFLT|T_USER:
#ifdef VM86
if (frame->tf_eflags & PSL_VM) {
vm86_gpfault(l, type & ~T_USER);
goto out;
}
#endif
/*
* If pmap_exec_fixup does something,
* let's retry the trap.
*/
if (pmap_exec_fixup(&p->p_vmspace->vm_map, frame, pcb)){
goto out;
}
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_ACCERR;
break;
default:
KASSERT(0);
break;
}
goto trapsignal;
case T_PRIVINFLT|T_USER: /* privileged instruction fault */
case T_FPOPFLT|T_USER: /* coprocessor operand fault */
KSI_INIT_TRAP(&ksi);
ksi.ksi_signo = SIGILL;
ksi.ksi_addr = (void *) frame->tf_eip;
switch (type) {
case T_PRIVINFLT|T_USER:
ksi.ksi_code = ILL_PRVOPC;
break;
case T_FPOPFLT|T_USER:
ksi.ksi_code = ILL_COPROC;
break;
default:
ksi.ksi_code = 0;
break;
}
goto trapsignal;
case T_ASTFLT|T_USER:
/* Allow process switch. */
//curcpu()->ci_data.cpu_nast++;
if (l->l_pflag & LP_OWEUPC) {
l->l_pflag &= ~LP_OWEUPC;
ADDUPROF(l);
}
/* Allow a forced task switch. */
if (curcpu()->ci_want_resched) {
preempt();
}
goto out;
case T_BOUND|T_USER:
case T_OFLOW|T_USER:
case T_DIVIDE|T_USER:
KSI_INIT_TRAP(&ksi);
ksi.ksi_signo = SIGFPE;
ksi.ksi_addr = (void *)frame->tf_eip;
switch (type) {
case T_BOUND|T_USER:
ksi.ksi_code = FPE_FLTSUB;
break;
case T_OFLOW|T_USER:
ksi.ksi_code = FPE_INTOVF;
break;
case T_DIVIDE|T_USER:
ksi.ksi_code = FPE_INTDIV;
break;
default:
ksi.ksi_code = 0;
break;
}
goto trapsignal;
case T_PAGEFLT:
/* Allow page faults in kernel mode. */
if (__predict_false(l == NULL))
goto we_re_toast;
/*
* fusubail is used by [fs]uswintr() to prevent page faulting
* from inside the profiling interrupt.
*/
onfault = pcb->pcb_onfault;
if (onfault == fusubail || onfault == return_address_fault) {
goto copyefault;
}
if (cpu_intr_p() || (l->l_pflag & LP_INTR) != 0) {
goto we_re_toast;
}
cr2 = rcr2();
goto faultcommon;
case T_PAGEFLT|T_USER: { /* page fault */
register vaddr_t va;
register struct vmspace *vm;
register struct vm_map *map;
vm_prot_t ftype;
extern struct vm_map *kernel_map;
cr2 = rcr2();
faultcommon:
vm = p->p_vmspace;
if (__predict_false(vm == NULL)) {
goto we_re_toast;
}
pcb->pcb_cr2 = cr2;
va = trunc_page((vaddr_t)cr2);
/*
* It is only a kernel address space fault iff:
* 1. (type & T_USER) == 0 and
* 2. pcb_onfault not set or
* 3. pcb_onfault set but supervisor space fault
* The last can occur during an exec() copyin where the
* argument space is lazy-allocated.
*/
if (type == T_PAGEFLT && va >= KERNBASE)
map = kernel_map;
else
map = &vm->vm_map;
if (frame->tf_err & PGEX_W)
ftype = VM_PROT_WRITE;
else if (frame->tf_err & PGEX_X)
ftype = VM_PROT_EXECUTE;
else
ftype = VM_PROT_READ;
#ifdef DIAGNOSTIC
if (map == kernel_map && va == 0) {
printf("trap: bad kernel access at %lx\n", va);
goto we_re_toast;
}
#endif
/* Fault the original page in. */
onfault = pcb->pcb_onfault;
pcb->pcb_onfault = NULL;
error = uvm_fault(map, va, ftype);
pcb->pcb_onfault = onfault;
if (error == 0) {
if (map != kernel_map && (void *)va >= vm->vm_maxsaddr)
uvm_grow(p, va);
pfail = false;
while (type == T_PAGEFLT) {
/*
* we need to switch pmap now if we're in
* the middle of copyin/out.
*
* but we don't need to do so for kcopy as
* it never touch userspace.
*/
kpreempt_disable();
if (curcpu()->ci_want_pmapload) {
onfault = onfault_handler(pcb, frame);
if (onfault != kcopy_fault) {
pmap_load();
}
}
/*
* We need to keep the pmap loaded and
* so avoid being preempted until back
* into the copy functions. Disable
* interrupts at the hardware level before
* re-enabling preemption. Interrupts
* will be re-enabled by 'iret' when
* returning back out of the trap stub.
* They'll only be re-enabled when the
* program counter is once again in
* the copy functions, and so visible
* to cpu_kpreempt_exit().
*/
#ifndef XEN
x86_disable_intr();
#endif
l->l_nopreempt--;
if (l->l_nopreempt > 0 || !l->l_dopreempt ||
pfail) {
return;
}
#ifndef XEN
x86_enable_intr();
#endif
/*
* If preemption fails for some reason,
* don't retry it. The conditions won't
* change under our nose.
*/
pfail = kpreempt(0);
}
goto out;
}
if (type == T_PAGEFLT) {
onfault = onfault_handler(pcb, frame);
if (onfault != NULL)
goto copyfault;
printf("uvm_fault(%p, %#lx, %d) -> %#x\n",
map, va, ftype, error);
goto kernelfault;
}
KSI_INIT_TRAP(&ksi);
ksi.ksi_trap = type & ~T_USER;
ksi.ksi_addr = (void *)cr2;
switch (error) {
case EINVAL:
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_ADRERR;
break;
case EACCES:
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_ACCERR;
error = EFAULT;
break;
case ENOMEM:
ksi.ksi_signo = SIGKILL;
printf("UVM: pid %d.%d (%s), uid %d killed: "
"out of swap\n", p->p_pid, l->l_lid, p->p_comm,
l->l_cred ? kauth_cred_geteuid(l->l_cred) : -1);
break;
default:
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = SEGV_MAPERR;
break;
}
#ifdef TRAP_SIGDEBUG
printf("pid %d.%d (%s): signal %d at eip %x addr %lx "
"error %d\n", p->p_pid, l->l_lid, p->p_comm, ksi.ksi_signo,
frame->tf_eip, va, error);
#endif
(*p->p_emul->e_trapsignal)(l, &ksi);
break;
}
case T_TRCTRAP:
/* Check whether they single-stepped into a lcall. */
if (frame->tf_eip == (int)IDTVEC(osyscall))
return;
if (frame->tf_eip == (int)IDTVEC(osyscall) + 1) {
frame->tf_eflags &= ~PSL_T;
return;
}
goto we_re_toast;
case T_BPTFLT|T_USER: /* bpt instruction fault */
case T_TRCTRAP|T_USER: /* trace trap */
/*
* Don't go single-stepping into a RAS.
*/
if (p->p_raslist == NULL ||
(ras_lookup(p, (void *)frame->tf_eip) == (void *)-1)) {
KSI_INIT_TRAP(&ksi);
ksi.ksi_signo = SIGTRAP;
ksi.ksi_trap = type & ~T_USER;
if (type == (T_BPTFLT|T_USER))
ksi.ksi_code = TRAP_BRKPT;
else
ksi.ksi_code = TRAP_TRACE;
ksi.ksi_addr = (void *)frame->tf_eip;
(*p->p_emul->e_trapsignal)(l, &ksi);
}
break;
case T_NMI:
if (nmi_dispatch(frame))
return;
/* NMI can be hooked up to a pushbutton for debugging */
if (kgdb_trap(type, frame))
return;
if (kdb_trap(type, 0, frame))
return;
/* machine/parity/power fail/"kitchen sink" faults */
#if NMCA > 0
mca_nmi();
#endif
x86_nmi();
}
if ((type & T_USER) == 0)
return;
out:
userret(l);
return;
trapsignal:
ksi.ksi_trap = type & ~T_USER;
(*p->p_emul->e_trapsignal)(l, &ksi);
userret(l);
}
/*
* General page fault handler.
*/
void
do_fault(struct trapframe *tf, struct lwp *l,
struct vm_map *map, vaddr_t va, vm_prot_t atype)
{
int error;
if (pmap_fault(map->pmap, va, atype))
return;
struct pcb * const pcb = lwp_getpcb(l);
void * const onfault = pcb->pcb_onfault;
const bool user = TRAP_USERMODE(tf);
if (cpu_intr_p()) {
KASSERT(!user);
error = EFAULT;
} else {
pcb->pcb_onfault = NULL;
error = uvm_fault(map, va, atype);
pcb->pcb_onfault = onfault;
}
if (error != 0) {
ksiginfo_t ksi;
if (onfault != NULL) {
tf->tf_r0 = error;
tf->tf_r15 = (tf->tf_r15 & ~R15_PC) |
(register_t)onfault;
return;
}
#ifdef DDB
if (db_validating) {
db_faulted = true;
tf->tf_r15 += INSN_SIZE;
return;
}
#endif
if (!user) {
#ifdef DDB
db_printf("Unhandled data abort in kernel mode\n");
kdb_trap(T_FAULT, tf);
#else
#ifdef DEBUG
printf("Unhandled data abort:\n");
printregs(tf);
#endif
panic("unhandled data abort in kernel mode");
#endif
}
KSI_INIT_TRAP(&ksi);
if (error == ENOMEM) {
printf("UVM: pid %d (%s), uid %d killed: "
"out of swap\n",
l->l_proc->p_pid, l->l_proc->p_comm,
l->l_cred ? kauth_cred_geteuid(l->l_cred) : -1);
ksi.ksi_signo = SIGKILL;
} else
ksi.ksi_signo = SIGSEGV;
ksi.ksi_code = (error == EPERM) ? SEGV_ACCERR : SEGV_MAPERR;
ksi.ksi_addr = (void *) va;
trapsignal(l, &ksi);
} else if (!user) {
ucas_ras_check(tf);
}
}
