/*
* Fill in a struct kinfo_lwp.
*/
void
fill_kinfo_lwp(struct lwp *lwp, struct kinfo_lwp *kl)
{
bzero(kl, sizeof(*kl));
kl->kl_pid = lwp->lwp_proc->p_pid;
kl->kl_tid = lwp->lwp_tid;
kl->kl_flags = lwp->lwp_flags;
kl->kl_stat = lwp->lwp_stat;
kl->kl_lock = lwp->lwp_lock;
kl->kl_tdflags = lwp->lwp_thread->td_flags;
/*
* The process/lwp stat may not reflect whether the process is
* actually sleeping or not if the related thread was directly
* descheduled by LWKT. Adjust the stat if the thread is not
* runnable and not waiting to be scheduled on a cpu by the
* user process scheduler.
*/
if (kl->kl_stat == LSRUN) {
if ((kl->kl_tdflags & TDF_RUNQ) == 0 &&
(lwp->lwp_mpflags & LWP_MP_ONRUNQ) == 0) {
kl->kl_stat = LSSLEEP;
}
}
#ifdef SMP
kl->kl_mpcount = get_mplock_count(lwp->lwp_thread);
#else
kl->kl_mpcount = 0;
#endif
kl->kl_prio = lwp->lwp_usdata.bsd4.priority; /* XXX TGEN dangerous assumption */
kl->kl_tdprio = lwp->lwp_thread->td_pri;
kl->kl_rtprio = lwp->lwp_rtprio;
kl->kl_uticks = lwp->lwp_thread->td_uticks;
kl->kl_sticks = lwp->lwp_thread->td_sticks;
kl->kl_iticks = lwp->lwp_thread->td_iticks;
kl->kl_cpticks = lwp->lwp_cpticks;
kl->kl_pctcpu = lwp->lwp_pctcpu;
kl->kl_slptime = lwp->lwp_slptime;
kl->kl_origcpu = lwp->lwp_usdata.bsd4.batch;
kl->kl_estcpu = lwp->lwp_usdata.bsd4.estcpu;
kl->kl_cpuid = lwp->lwp_thread->td_gd->gd_cpuid;
kl->kl_ru = lwp->lwp_ru;
kl->kl_siglist = lwp->lwp_siglist;
kl->kl_sigmask = lwp->lwp_sigmask;
kl->kl_wchan = (uintptr_t)lwp->lwp_thread->td_wchan;
if (lwp->lwp_thread->td_wmesg) {
strncpy(kl->kl_wmesg, lwp->lwp_thread->td_wmesg, WMESGLEN);
kl->kl_wmesg[WMESGLEN] = 0;
}
}
/*
* Fill in a struct kinfo_proc for kernel threads (i.e. those without proc).
*/
void
fill_kinfo_proc_kthread(struct thread *td, struct kinfo_proc *kp)
{
bzero(kp, sizeof(*kp));
/*
* Fill in fake proc information and semi-fake lwp info.
*/
kp->kp_pid = -1;
kp->kp_tdev = NOUDEV;
strncpy(kp->kp_comm, td->td_comm, sizeof(kp->kp_comm) - 1);
kp->kp_comm[sizeof(kp->kp_comm) - 1] = 0;
kp->kp_flags = P_SYSTEM;
kp->kp_stat = SACTIVE;
kp->kp_lwp.kl_pid = -1;
kp->kp_lwp.kl_tid = -1;
kp->kp_lwp.kl_tdflags = td->td_flags;
#ifdef SMP
kp->kp_lwp.kl_mpcount = get_mplock_count(td);
#else /* !SMP */
kp->kp_lwp.kl_mpcount = 0;
#endif /* SMP */
kp->kp_lwp.kl_tdprio = td->td_pri;
kp->kp_lwp.kl_rtprio.type = RTP_PRIO_THREAD;
kp->kp_lwp.kl_rtprio.prio = td->td_pri;
kp->kp_lwp.kl_uticks = td->td_uticks;
kp->kp_lwp.kl_sticks = td->td_sticks;
kp->kp_lwp.kl_iticks = td->td_iticks;
kp->kp_lwp.kl_cpuid = td->td_gd->gd_cpuid;
kp->kp_lwp.kl_wchan = (uintptr_t)td->td_wchan;
if (td->td_flags & TDF_RUNQ)
kp->kp_lwp.kl_stat = LSRUN;
else
kp->kp_lwp.kl_stat = LSSLEEP;
if (td->td_wmesg) {
strncpy(kp->kp_lwp.kl_wmesg, td->td_wmesg, WMESGLEN);
kp->kp_lwp.kl_wmesg[WMESGLEN] = 0;
}
}
void
trap(struct trapframe *frame)
{
struct globaldata *gd = mycpu;
struct thread *td = gd->gd_curthread;
struct lwp *lp = td->td_lwp;
struct proc *p;
int sticks = 0;
int i = 0, ucode = 0, type, code;
int have_mplock = 0;
#ifdef INVARIANTS
int crit_count = td->td_critcount;
lwkt_tokref_t curstop = td->td_toks_stop;
#endif
vm_offset_t eva;
p = td->td_proc;
#ifdef DDB
/*
* We need to allow T_DNA faults when the debugger is active since
* some dumping paths do large bcopy() which use the floating
* point registers for faster copying.
*/
if (db_active && frame->tf_trapno != T_DNA) {
eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
++gd->gd_trap_nesting_level;
MAKEMPSAFE(have_mplock);
trap_fatal(frame, eva);
--gd->gd_trap_nesting_level;
goto out2;
}
#endif
eva = 0;
++gd->gd_trap_nesting_level;
if (frame->tf_trapno == T_PAGEFLT) {
/*
* For some Cyrix CPUs, %cr2 is clobbered by interrupts.
* This problem is worked around by using an interrupt
* gate for the pagefault handler. We are finally ready
* to read %cr2 and then must reenable interrupts.
*
* XXX this should be in the switch statement, but the
* NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
* flow of control too much for this to be obviously
* correct.
*/
eva = rcr2();
cpu_enable_intr();
}
--gd->gd_trap_nesting_level;
if (!(frame->tf_eflags & PSL_I)) {
/*
* Buggy application or kernel code has disabled interrupts
* and then trapped. Enabling interrupts now is wrong, but
* it is better than running with interrupts disabled until
* they are accidentally enabled later.
*/
type = frame->tf_trapno;
if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
MAKEMPSAFE(have_mplock);
kprintf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curproc->p_comm, type);
} else if (type != T_BPTFLT && type != T_TRCTRAP) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
MAKEMPSAFE(have_mplock);
kprintf("kernel trap %d with interrupts disabled\n",
type);
}
cpu_enable_intr();
}
#if defined(I586_CPU) && !defined(NO_F00F_HACK)
restart:
#endif
type = frame->tf_trapno;
code = frame->tf_err;
if (in_vm86call) {
if (frame->tf_eflags & PSL_VM &&
(type == T_PROTFLT || type == T_STKFLT)) {
KKASSERT(get_mplock_count(curthread) > 0);
i = vm86_emulate((struct vm86frame *)frame);
KKASSERT(get_mplock_count(curthread) > 0);
if (i != 0) {
/*
* returns to original process
*/
vm86_trap((struct vm86frame *)frame,
have_mplock);
KKASSERT(0); /* NOT REACHED */
}
goto out2;
}
switch (type) {
/*
* these traps want either a process context, or
* assume a normal userspace trap.
*/
case T_PROTFLT:
case T_SEGNPFLT:
trap_fatal(frame, eva);
goto out2;
case T_TRCTRAP:
type = T_BPTFLT; /* kernel breakpoint */
/* FALL THROUGH */
}
goto kernel_trap; /* normal kernel trap handling */
}
if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
/* user trap */
KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
frame->tf_trapno, eva);
userenter(td, p);
sticks = (int)td->td_sticks;
lp->lwp_md.md_regs = frame;
switch (type) {
case T_PRIVINFLT: /* privileged instruction fault */
i = SIGILL;
ucode = ILL_PRVOPC;
break;
case T_BPTFLT: /* bpt instruction fault */
case T_TRCTRAP: /* trace trap */
frame->tf_eflags &= ~PSL_T;
i = SIGTRAP;
ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
break;
case T_ARITHTRAP: /* arithmetic trap */
ucode = code;
i = SIGFPE;
break;
case T_ASTFLT: /* Allow process switch */
mycpu->gd_cnt.v_soft++;
if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
atomic_clear_int(&mycpu->gd_reqflags,
RQF_AST_OWEUPC);
addupc_task(p, p->p_prof.pr_addr,
p->p_prof.pr_ticks);
}
goto out;
/*
* The following two traps can happen in
* vm86 mode, and, if so, we want to handle
* them specially.
*/
case T_PROTFLT: /* general protection fault */
case T_STKFLT: /* stack fault */
if (frame->tf_eflags & PSL_VM) {
i = vm86_emulate((struct vm86frame *)frame);
if (i == 0)
goto out;
break;
}
i = SIGBUS;
ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
break;
case T_SEGNPFLT: /* segment not present fault */
i = SIGBUS;
ucode = BUS_ADRERR;
break;
case T_TSSFLT: /* invalid TSS fault */
case T_DOUBLEFLT: /* double fault */
default:
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_PAGEFLT: /* page fault */
i = trap_pfault(frame, TRUE, eva);
if (i == -1)
goto out;
#if defined(I586_CPU) && !defined(NO_F00F_HACK)
if (i == -2)
goto restart;
#endif
if (i == 0)
goto out;
if (i == SIGSEGV)
ucode = SEGV_MAPERR;
else {
i = SIGSEGV;
ucode = SEGV_ACCERR;
}
break;
case T_DIVIDE: /* integer divide fault */
ucode = FPE_INTDIV;
i = SIGFPE;
break;
#if NISA > 0
case T_NMI:
MAKEMPSAFE(have_mplock);
#ifdef POWERFAIL_NMI
goto handle_powerfail;
#else /* !POWERFAIL_NMI */
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef DDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (ddb_on_nmi) {
kprintf ("NMI ... going to debugger\n");
kdb_trap (type, 0, frame);
}
#endif /* DDB */
goto out2;
} else if (panic_on_nmi)
panic("NMI indicates hardware failure");
break;
#endif /* POWERFAIL_NMI */
#endif /* NISA > 0 */
case T_OFLOW: /* integer overflow fault */
ucode = FPE_INTOVF;
i = SIGFPE;
break;
case T_BOUND: /* bounds check fault */
ucode = FPE_FLTSUB;
i = SIGFPE;
break;
case T_DNA:
/*
* Virtual kernel intercept - pass the DNA exception
* to the virtual kernel if it asked to handle it.
* This occurs when the virtual kernel is holding
* onto the FP context for a different emulated
* process then the one currently running.
*
* We must still call npxdna() since we may have
* saved FP state that the virtual kernel needs
* to hand over to a different emulated process.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
(td->td_pcb->pcb_flags & FP_VIRTFP)
) {
npxdna();
break;
}
#if NNPX > 0
/*
* The kernel may have switched out the FP unit's
* state, causing the user process to take a fault
* when it tries to use the FP unit. Restore the
* state here
*/
if (npxdna())
goto out;
#endif
if (!pmath_emulate) {
i = SIGFPE;
ucode = FPE_FPU_NP_TRAP;
break;
}
i = (*pmath_emulate)(frame);
if (i == 0) {
if (!(frame->tf_eflags & PSL_T))
goto out2;
frame->tf_eflags &= ~PSL_T;
i = SIGTRAP;
}
/* else ucode = emulator_only_knows() XXX */
break;
case T_FPOPFLT: /* FPU operand fetch fault */
ucode = ILL_COPROC;
i = SIGILL;
break;
case T_XMMFLT: /* SIMD floating-point exception */
ucode = 0; /* XXX */
i = SIGFPE;
break;
}
} else {
kernel_trap:
/* kernel trap */
switch (type) {
case T_PAGEFLT: /* page fault */
trap_pfault(frame, FALSE, eva);
goto out2;
case T_DNA:
#if NNPX > 0
/*
* The kernel may be using npx for copying or other
* purposes.
*/
if (npxdna())
goto out2;
#endif
break;
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
/*
* Invalid segment selectors and out of bounds
* %eip's and %esp's can be set up in user mode.
* This causes a fault in kernel mode when the
* kernel tries to return to user mode. We want
* to get this fault so that we can fix the
* problem here and not have to check all the
* selectors and pointers when the user changes
* them.
*/
#define MAYBE_DORETI_FAULT(where, whereto) \
do { \
if (frame->tf_eip == (int)where) { \
frame->tf_eip = (int)whereto; \
goto out2; \
} \
} while (0)
if (mycpu->gd_intr_nesting_level == 0) {
/*
* Invalid %fs's and %gs's can be created using
* procfs or PT_SETREGS or by invalidating the
* underlying LDT entry. This causes a fault
* in kernel mode when the kernel attempts to
* switch contexts. Lose the bad context
* (XXX) so that we can continue, and generate
* a signal.
*/
MAYBE_DORETI_FAULT(doreti_iret,
doreti_iret_fault);
MAYBE_DORETI_FAULT(doreti_popl_ds,
doreti_popl_ds_fault);
MAYBE_DORETI_FAULT(doreti_popl_es,
doreti_popl_es_fault);
MAYBE_DORETI_FAULT(doreti_popl_fs,
doreti_popl_fs_fault);
MAYBE_DORETI_FAULT(doreti_popl_gs,
doreti_popl_gs_fault);
/*
* NOTE: cpu doesn't push esp on kernel trap
*/
if (td->td_pcb->pcb_onfault &&
td->td_pcb->pcb_onfault_sp ==
(int)&frame->tf_esp) {
frame->tf_eip =
(register_t)td->td_pcb->pcb_onfault;
goto out2;
}
}
break;
case T_TSSFLT:
/*
* PSL_NT can be set in user mode and isn't cleared
* automatically when the kernel is entered. This
* causes a TSS fault when the kernel attempts to
* `iret' because the TSS link is uninitialized. We
* want to get this fault so that we can fix the
* problem here and not every time the kernel is
* entered.
*/
if (frame->tf_eflags & PSL_NT) {
frame->tf_eflags &= ~PSL_NT;
goto out2;
}
break;
case T_TRCTRAP: /* trace trap */
if (frame->tf_eip == (int)IDTVEC(syscall)) {
/*
* We've just entered system mode via the
* syscall lcall. Continue single stepping
* silently until the syscall handler has
* saved the flags.
*/
goto out2;
}
if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
/*
* The syscall handler has now saved the
* flags. Stop single stepping it.
*/
frame->tf_eflags &= ~PSL_T;
goto out2;
}
/*
* Ignore debug register trace traps due to
* accesses in the user's address space, which
* can happen under several conditions such as
* if a user sets a watchpoint on a buffer and
* then passes that buffer to a system call.
* We still want to get TRCTRAPS for addresses
* in kernel space because that is useful when
* debugging the kernel.
*/
if (user_dbreg_trap()) {
/*
* Reset breakpoint bits because the
* processor doesn't
*/
load_dr6(rdr6() & 0xfffffff0);
goto out2;
}
/*
* FALLTHROUGH (TRCTRAP kernel mode, kernel address)
*/
case T_BPTFLT:
/*
* If DDB is enabled, let it handle the debugger trap.
* Otherwise, debugger traps "can't happen".
*/
ucode = TRAP_BRKPT;
#ifdef DDB
MAKEMPSAFE(have_mplock);
if (kdb_trap (type, 0, frame))
goto out2;
#endif
break;
#if NISA > 0
case T_NMI:
MAKEMPSAFE(have_mplock);
#ifdef POWERFAIL_NMI
#ifndef TIMER_FREQ
# define TIMER_FREQ 1193182
#endif
handle_powerfail:
{
static unsigned lastalert = 0;
if (time_uptime - lastalert > 10) {
log(LOG_WARNING, "NMI: power fail\n");
sysbeep(TIMER_FREQ/880, hz);
lastalert = time_uptime;
}
/* YYY mp count */
goto out2;
}
#else /* !POWERFAIL_NMI */
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef DDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (ddb_on_nmi) {
kprintf ("NMI ... going to debugger\n");
kdb_trap (type, 0, frame);
}
#endif /* DDB */
goto out2;
} else if (panic_on_nmi == 0)
goto out2;
/* FALL THROUGH */
#endif /* POWERFAIL_NMI */
#endif /* NISA > 0 */
}
MAKEMPSAFE(have_mplock);
trap_fatal(frame, eva);
goto out2;
}
/*
* Virtual kernel intercept - if the fault is directly related to a
* VM context managed by a virtual kernel then let the virtual kernel
* handle it.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
vkernel_trap(lp, frame);
goto out;
}
/* Translate fault for emulators (e.g. Linux) */
if (*p->p_sysent->sv_transtrap)
i = (*p->p_sysent->sv_transtrap)(i, type);
MAKEMPSAFE(have_mplock);
trapsignal(lp, i, ucode);
#ifdef DEBUG
if (type <= MAX_TRAP_MSG) {
uprintf("fatal process exception: %s",
trap_msg[type]);
if ((type == T_PAGEFLT) || (type == T_PROTFLT))
uprintf(", fault VA = 0x%lx", (u_long)eva);
uprintf("\n");
}
#endif
out:
userret(lp, frame, sticks);
userexit(lp);
out2: ;
if (have_mplock)
rel_mplock();
if (p != NULL && lp != NULL)
KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
#ifdef INVARIANTS
KASSERT(crit_count == td->td_critcount,
("trap: critical section count mismatch! %d/%d",
crit_count, td->td_pri));
KASSERT(curstop == td->td_toks_stop,
("trap: extra tokens held after trap! %zd/%zd",
curstop - &td->td_toks_base,
td->td_toks_stop - &td->td_toks_base));
#endif
}
