static int
db_ss(struct db_variable *vp, db_expr_t *valuep, int op)
{
if (kdb_frame == NULL)
return (0);
if (op == DB_VAR_GET)
*valuep = (ISPL(kdb_frame->tf_cs)) ? kdb_frame->tf_ss : rss();
else if (ISPL(kdb_frame->tf_cs))
kdb_frame->tf_ss = *valuep;
return (1);
}
static int
db_rsp(struct db_variable *vp, db_expr_t *valuep, int op)
{
if (kdb_frame == NULL)
return (0);
if (op == DB_VAR_GET)
*valuep = get_rsp(kdb_frame);
else if (ISPL(kdb_frame->tf_cs))
kdb_frame->tf_rsp = *valuep;
return (1);
}
void
_test_frame_enter(struct trapframe *frame)
{
thread_t td = curthread;
if (ISPL(frame->tf_cs) == SEL_UPL) {
KKASSERT(td->td_lwp);
KASSERT(td->td_lwp->lwp_md.md_regs == frame,
("_test_frame_exit: Frame mismatch %p %p",
td->td_lwp->lwp_md.md_regs, frame));
td->td_lwp->lwp_saveusp = (void *)frame->tf_rsp;
td->td_lwp->lwp_saveupc = (void *)frame->tf_rip;
}
if ((char *)frame < td->td_kstack ||
(char *)frame > td->td_kstack + td->td_kstack_size) {
panic("_test_frame_exit: frame not on kstack %p kstack=%p\n",
frame, td->td_kstack);
}
}
void
_test_frame_exit(struct trapframe *frame)
{
thread_t td = curthread;
if (ISPL(frame->tf_cs) == SEL_UPL) {
KKASSERT(td->td_lwp);
KASSERT(td->td_lwp->lwp_md.md_regs == frame,
("_test_frame_exit: Frame mismatch %p %p",
td->td_lwp->lwp_md.md_regs, frame));
if (td->td_lwp->lwp_saveusp != (void *)frame->tf_rsp) {
kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n",
td->td_comm, td->td_proc->p_pid,
td->td_lwp->lwp_saveusp,
(void *)frame->tf_rsp);
}
if (td->td_lwp->lwp_saveupc != (void *)frame->tf_rip) {
kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n",
td->td_comm, td->td_proc->p_pid,
td->td_lwp->lwp_saveupc,
(void *)frame->tf_rip);
}
/*
* adulterate the fields to catch entries that
* don't run through test_frame_enter
*/
td->td_lwp->lwp_saveusp =
(void *)~(intptr_t)td->td_lwp->lwp_saveusp;
td->td_lwp->lwp_saveupc =
(void *)~(intptr_t)td->td_lwp->lwp_saveupc;
}
if ((char *)frame < td->td_kstack ||
(char *)frame > td->td_kstack + td->td_kstack_size) {
panic("_test_frame_exit: frame not on kstack %p kstack=%p\n",
frame, td->td_kstack);
}
}
static int
fasttrap_do_seg(fasttrap_tracepoint_t *tp, struct reg *rp, uintptr_t *addr)
{
proc_t *p = curproc;
#ifdef __i386__
struct segment_descriptor *desc;
#else
struct user_segment_descriptor *desc;
#endif
uint16_t sel = 0, ndx, type;
uintptr_t limit;
switch (tp->ftt_segment) {
case FASTTRAP_SEG_CS:
sel = rp->r_cs;
break;
case FASTTRAP_SEG_DS:
sel = rp->r_ds;
break;
case FASTTRAP_SEG_ES:
sel = rp->r_es;
break;
case FASTTRAP_SEG_FS:
sel = rp->r_fs;
break;
case FASTTRAP_SEG_GS:
sel = rp->r_gs;
break;
case FASTTRAP_SEG_SS:
sel = rp->r_ss;
break;
}
/*
* Make sure the given segment register specifies a user priority
* selector rather than a kernel selector.
*/
if (ISPL(sel) != SEL_UPL)
return (-1);
ndx = IDXSEL(sel);
/*
* Check the bounds and grab the descriptor out of the specified
* descriptor table.
*/
if (ISLDT(sel)) {
#ifdef __i386__
if (ndx > p->p_md.md_ldt->ldt_len)
return (-1);
desc = (struct segment_descriptor *)
p->p_md.md_ldt[ndx].ldt_base;
#else
if (ndx > max_ldt_segment)
return (-1);
desc = (struct user_segment_descriptor *)
p->p_md.md_ldt[ndx].ldt_base;
#endif
} else {
if (ndx >= NGDT)
return (-1);
#ifdef __i386__
desc = &gdt[ndx].sd;
#else
desc = &gdt[ndx];
#endif
}
/*
* The descriptor must have user privilege level and it must be
* present in memory.
*/
if (desc->sd_dpl != SEL_UPL || desc->sd_p != 1)
return (-1);
type = desc->sd_type;
/*
* If the S bit in the type field is not set, this descriptor can
* only be used in system context.
*/
if ((type & 0x10) != 0x10)
return (-1);
limit = USD_GETLIMIT(desc) * (desc->sd_gran ? PAGESIZE : 1);
if (tp->ftt_segment == FASTTRAP_SEG_CS) {
/*
* The code/data bit and readable bit must both be set.
*/
if ((type & 0xa) != 0xa)
return (-1);
if (*addr > limit)
return (-1);
} else {
/*
* The code/data bit must be clear.
*/
if ((type & 0x8) != 0)
return (-1);
/*
* If the expand-down bit is clear, we just check the limit as
* it would naturally be applied. Otherwise, we need to check
* that the address is the range [limit + 1 .. 0xffff] or
* [limit + 1 ... 0xffffffff] depending on if the default
* operand size bit is set.
*/
if ((type & 0x4) == 0) {
if (*addr > limit)
return (-1);
} else if (desc->sd_def32) {
if (*addr < limit + 1 || 0xffff < *addr)
return (-1);
} else {
if (*addr < limit + 1 || 0xffffffff < *addr)
return (-1);
}
}
*addr += USD_GETBASE(desc);
return (0);
}
static __inline long
get_rsp(struct trapframe *tf)
{
return ((ISPL(tf->tf_cs)) ? tf->tf_rsp :
(db_expr_t)tf + offsetof(struct trapframe, tf_rsp));
}
static void
trap_fatal(struct trapframe *frame, vm_offset_t eva)
{
int code, ss;
u_int type;
long rsp;
struct soft_segment_descriptor softseg;
char *msg;
code = frame->tf_err;
type = frame->tf_trapno;
sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
if (type <= MAX_TRAP_MSG)
msg = trap_msg[type];
else
msg = "UNKNOWN";
kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
/* three separate prints in case of a trap on an unmapped page */
kprintf("cpuid = %d; ", mycpu->gd_cpuid);
if (lapic_usable)
kprintf("lapic id = %u\n", LAPIC_READID);
if (type == T_PAGEFLT) {
kprintf("fault virtual address = 0x%lx\n", eva);
kprintf("fault code = %s %s %s, %s\n",
code & PGEX_U ? "user" : "supervisor",
code & PGEX_W ? "write" : "read",
code & PGEX_I ? "instruction" : "data",
code & PGEX_P ? "protection violation" : "page not present");
}
kprintf("instruction pointer = 0x%lx:0x%lx\n",
frame->tf_cs & 0xffff, frame->tf_rip);
if (ISPL(frame->tf_cs) == SEL_UPL) {
ss = frame->tf_ss & 0xffff;
rsp = frame->tf_rsp;
} else {
/*
* NOTE: in 64-bit mode traps push rsp/ss even if no ring
* change occurs.
*/
ss = GSEL(GDATA_SEL, SEL_KPL);
rsp = frame->tf_rsp;
}
kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
softseg.ssd_gran);
kprintf("processor eflags = ");
if (frame->tf_rflags & PSL_T)
kprintf("trace trap, ");
if (frame->tf_rflags & PSL_I)
kprintf("interrupt enabled, ");
if (frame->tf_rflags & PSL_NT)
kprintf("nested task, ");
if (frame->tf_rflags & PSL_RF)
kprintf("resume, ");
kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
kprintf("current process = ");
if (curproc) {
kprintf("%lu\n",
(u_long)curproc->p_pid);
} else {
kprintf("Idle\n");
}
kprintf("current thread = pri %d ", curthread->td_pri);
if (curthread->td_critcount)
kprintf("(CRIT)");
kprintf("\n");
#ifdef DDB
if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
return;
#endif
kprintf("trap number = %d\n", type);
if (type <= MAX_TRAP_MSG)
panic("%s", trap_msg[type]);
else
panic("unknown/reserved trap");
}
/*
* Exception, fault, and trap interface to the 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.
*
* This function is also called from doreti in an interlock to handle ASTs.
* For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
*
* NOTE! We have to retrieve the fault address prior to potentially
* blocking, including blocking on any token.
*
* NOTE! NMI and kernel DBG traps remain on their respective pcpu IST
* stacks if taken from a kernel RPL. trap() cannot block in this
* situation. DDB entry or a direct report-and-return is ok.
*
* XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
* if an attempt is made to switch from a fast interrupt or IPI.
*/
void
trap(struct trapframe *frame)
{
static struct krate sscpubugrate = { 1 };
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;
#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;
clear_quickret();
#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 ? frame->tf_addr : 0);
++gd->gd_trap_nesting_level;
trap_fatal(frame, eva);
--gd->gd_trap_nesting_level;
goto out2;
}
#endif
eva = 0;
if ((frame->tf_rflags & PSL_I) == 0) {
/*
* 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) {
/* JG curproc can be NULL */
kprintf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curproc->p_comm, type);
} else if ((type == T_STKFLT || type == T_PROTFLT ||
type == T_SEGNPFLT) &&
frame->tf_rip == (long)doreti_iret) {
/*
* iretq fault from kernel mode during return to
* userland.
*
* This situation is expected, don't complain.
*/
} else if (type != T_NMI && type != T_BPTFLT &&
type != T_TRCTRAP) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
kprintf("kernel trap %d (%s @ 0x%016jx) with "
"interrupts disabled\n",
type,
td->td_comm,
frame->tf_rip);
}
cpu_enable_intr();
}
type = frame->tf_trapno;
code = frame->tf_err;
if (ISPL(frame->tf_cs) == SEL_UPL) {
/* user trap */
KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
frame->tf_trapno, eva);
userenter(td, p);
sticks = (int)td->td_sticks;
KASSERT(lp->lwp_md.md_regs == frame,
("Frame mismatch %p %p", 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_rflags &= ~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;
case T_PROTFLT: /* general protection fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_STKFLT: /* stack fault */
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);
#ifdef DDB
if (frame->tf_rip == 0) {
/* used for kernel debugging only */
while (freeze_on_seg_fault)
tsleep(p, 0, "freeze", hz * 20);
}
#endif
if (i == -1 || 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:
/* 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 /* 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;
}
/*
* 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()) {
gd->gd_cnt.v_trap++;
goto out;
}
i = SIGFPE;
ucode = FPE_FPU_NP_TRAP;
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 */
switch (type) {
case T_PAGEFLT: /* page fault */
trap_pfault(frame, FALSE);
goto out2;
case T_DNA:
/*
* The kernel is apparently using fpu for copying.
* XXX this should be fatal unless the kernel has
* registered such use.
*/
if (npxdna()) {
gd->gd_cnt.v_trap++;
goto out2;
}
break;
case T_STKFLT: /* stack fault */
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
/*
* Invalid segment selectors and out of bounds
* %rip's and %rsp'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.
*/
if (mycpu->gd_intr_nesting_level == 0) {
/*
* NOTE: in 64-bit mode traps push rsp/ss
* even if no ring change occurs.
*/
if (td->td_pcb->pcb_onfault &&
td->td_pcb->pcb_onfault_sp ==
frame->tf_rsp) {
frame->tf_rip = (register_t)
td->td_pcb->pcb_onfault;
goto out2;
}
/*
* If the iretq in doreti faults during
* return to user, it will be special-cased
* in IDTVEC(prot) to get here. We want
* to 'return' to doreti_iret_fault in
* ipl.s in approximately the same state we
* were in at the iretq.
*/
if (frame->tf_rip == (long)doreti_iret) {
frame->tf_rip = (long)doreti_iret_fault;
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_rflags & PSL_NT) {
frame->tf_rflags &= ~PSL_NT;
#if 0
/* do we need this? */
if (frame->tf_rip == (long)doreti_iret)
frame->tf_rip = (long)doreti_iret_fault;
#endif
goto out2;
}
break;
case T_TRCTRAP: /* trace trap */
/*
* Detect historical CPU artifact on syscall or int $3
* entry (if not shortcutted in exception.s via
* DIRECT_DISALLOW_SS_CPUBUG).
*/
gd->gd_cnt.v_trap++;
if (frame->tf_rip == (register_t)IDTVEC(fast_syscall)) {
krateprintf(&sscpubugrate,
"Caught #DB at syscall cpu artifact\n");
goto out2;
}
if (frame->tf_rip == (register_t)IDTVEC(bpt)) {
krateprintf(&sscpubugrate,
"Caught #DB at int $N cpu artifact\n");
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() & ~0xf);
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
if (kdb_trap(type, 0, frame))
goto out2;
#endif
break;
#if NISA > 0
case T_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 /* NISA > 0 */
}
trap_fatal(frame, 0);
goto out2;
}
/*
* Fault from user mode, virtual kernel interecept.
*
* 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);
gd->gd_cnt.v_trap++;
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", frame->tf_addr);
uprintf("\n");
}
#endif
out:
userret(lp, frame, sticks);
userexit(lp);
out2: ;
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! %ld/%ld",
curstop - &td->td_toks_base,
td->td_toks_stop - &td->td_toks_base));
#endif
}
void
trap(struct trapframe *frame)
{
struct thread *td = curthread;
struct proc *p = td->td_proc;
int i = 0, ucode = 0, code;
u_int type;
register_t addr = 0;
ksiginfo_t ksi;
PCPU_INC(cnt.v_trap);
type = frame->tf_trapno;
#ifdef SMP
#ifdef STOP_NMI
/* Handler for NMI IPIs used for stopping CPUs. */
if (type == T_NMI) {
if (ipi_nmi_handler() == 0)
goto out;
}
#endif /* STOP_NMI */
#endif /* SMP */
#ifdef KDB
if (kdb_active) {
kdb_reenter();
goto out;
}
#endif
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI. If the PMC module is
* active, pass the 'rip' value to the PMC module's interrupt
* handler. A return value of '1' from the handler means that
* the NMI was handled by it and we can return immediately.
*/
if (type == T_NMI && pmc_intr &&
(*pmc_intr)(PCPU_GET(cpuid), frame))
goto out;
#endif
if (type == T_MCHK) {
if (!mca_intr())
trap_fatal(frame, 0);
goto out;
}
#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 it's 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 (dtrace_trap_func != NULL)
if ((*dtrace_trap_func)(frame, type))
goto out;
#endif
if ((frame->tf_rflags & PSL_I) == 0) {
/*
* 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.
*/
if (ISPL(frame->tf_cs) == SEL_UPL)
printf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curproc->p_comm, type);
else if (type != T_NMI && type != T_BPTFLT &&
type != T_TRCTRAP) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
printf("kernel trap %d with interrupts disabled\n",
type);
/*
* We shouldn't enable interrupts while holding a
* spin lock or servicing an NMI.
*/
if (type != T_NMI && td->td_md.md_spinlock_count == 0)
enable_intr();
}
}
code = frame->tf_err;
if (type == T_PAGEFLT) {
/*
* If we get a page fault while in a critical section, then
* it is most likely a fatal kernel page fault. The kernel
* is already going to panic trying to get a sleep lock to
* do the VM lookup, so just consider it a fatal trap so the
* kernel can print out a useful trap message and even get
* to the debugger.
*
* If we get a page fault while holding a non-sleepable
* lock, then it is most likely a fatal kernel page fault.
* If WITNESS is enabled, then it's going to whine about
* bogus LORs with various VM locks, so just skip to the
* fatal trap handling directly.
*/
if (td->td_critnest != 0 ||
WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
"Kernel page fault") != 0)
trap_fatal(frame, frame->tf_addr);
}
if (ISPL(frame->tf_cs) == SEL_UPL) {
/* user trap */
td->td_pticks = 0;
td->td_frame = frame;
addr = frame->tf_rip;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
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 */
enable_intr();
frame->tf_rflags &= ~PSL_T;
i = SIGTRAP;
ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
break;
case T_ARITHTRAP: /* arithmetic trap */
ucode = fputrap();
if (ucode == -1)
goto userout;
i = SIGFPE;
break;
case T_PROTFLT: /* general protection fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_STKFLT: /* stack fault */
case T_SEGNPFLT: /* segment not present fault */
i = SIGBUS;
ucode = BUS_ADRERR;
break;
case T_TSSFLT: /* invalid TSS fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_DOUBLEFLT: /* double fault */
default:
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_PAGEFLT: /* page fault */
addr = frame->tf_addr;
#ifdef KSE
if (td->td_pflags & TDP_SA)
thread_user_enter(td);
#endif
i = trap_pfault(frame, TRUE);
if (i == -1)
goto userout;
if (i == 0)
goto user;
if (i == SIGSEGV)
ucode = SEGV_MAPERR;
else {
if (prot_fault_translation == 0) {
/*
* Autodetect.
* This check also covers the images
* without the ABI-tag ELF note.
*/
if ((curproc->p_sysent ==
&elf64_freebsd_sysvec
#ifdef COMPAT_IA32
|| curproc->p_sysent ==
&ia32_freebsd_sysvec
#endif
) && p->p_osrel >= 700004) {
i = SIGSEGV;
ucode = SEGV_ACCERR;
} else {
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
}
} else if (prot_fault_translation == 1) {
/*
* Always compat mode.
*/
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
} else {
/*
* Always SIGSEGV mode.
*/
i = SIGSEGV;
ucode = SEGV_ACCERR;
}
}
break;
case T_DIVIDE: /* integer divide fault */
ucode = FPE_INTDIV;
i = SIGFPE;
break;
#ifdef DEV_ISA
case T_NMI:
/* machine/parity/power fail/"kitchen sink" faults */
/* XXX Giant */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto userout;
} else if (panic_on_nmi)
panic("NMI indicates hardware failure");
break;
#endif /* DEV_ISA */
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:
/* transparent fault (due to context switch "late") */
fpudna();
goto userout;
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 */
KASSERT(cold || td->td_ucred != NULL,
("kernel trap doesn't have ucred"));
switch (type) {
case T_PAGEFLT: /* page fault */
(void) trap_pfault(frame, FALSE);
goto out;
case T_DNA:
/*
* The kernel is apparently using fpu for copying.
* XXX this should be fatal unless the kernel has
* registered such use.
*/
fpudna();
printf("fpudna in kernel mode!\n");
goto out;
case T_STKFLT: /* stack fault */
break;
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
if (td->td_intr_nesting_level != 0)
break;
/*
* Invalid segment selectors and out of bounds
* %rip's and %rsp'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.
*/
if (frame->tf_rip == (long)doreti_iret) {
frame->tf_rip = (long)doreti_iret_fault;
goto out;
}
if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
frame->tf_rip =
(long)PCPU_GET(curpcb)->pcb_onfault;
goto out;
}
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_rflags & PSL_NT) {
frame->tf_rflags &= ~PSL_NT;
goto out;
}
break;
case T_TRCTRAP: /* trace trap */
/*
* 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
*/
/* XXX check upper bits here */
load_dr6(rdr6() & 0xfffffff0);
goto out;
}
/*
* FALLTHROUGH (TRCTRAP kernel mode, kernel address)
*/
case T_BPTFLT:
/*
* If KDB is enabled, let it handle the debugger trap.
* Otherwise, debugger traps "can't happen".
*/
#ifdef KDB
if (kdb_trap(type, 0, frame))
goto out;
#endif
break;
#ifdef DEV_ISA
case T_NMI:
/* XXX Giant */
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto out;
} else if (panic_on_nmi == 0)
goto out;
/* FALLTHROUGH */
#endif /* DEV_ISA */
}
trap_fatal(frame, 0);
goto out;
}
/* Translate fault for emulators (e.g. Linux) */
if (*p->p_sysent->sv_transtrap)
i = (*p->p_sysent->sv_transtrap)(i, type);
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = i;
ksi.ksi_code = ucode;
ksi.ksi_trapno = type;
ksi.ksi_addr = (void *)addr;
trapsignal(td, &ksi);
#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", frame->tf_addr);
uprintf("\n");
}
#endif
user:
userret(td, frame);
mtx_assert(&Giant, MA_NOTOWNED);
userout:
out:
return;
}
static int
trap_pfault(struct trapframe *frame, int usermode)
{
vm_offset_t va;
struct vmspace *vm = NULL;
vm_map_t map;
int rv = 0;
int fault_flags;
vm_prot_t ftype;
thread_t td = curthread;
struct lwp *lp = td->td_lwp;
struct proc *p;
va = trunc_page(frame->tf_addr);
if (va >= VM_MIN_KERNEL_ADDRESS) {
/*
* Don't allow user-mode faults in kernel address space.
*/
if (usermode) {
fault_flags = -1;
ftype = -1;
goto nogo;
}
map = &kernel_map;
} else {
/*
* This is a fault on non-kernel virtual memory.
* vm is initialized above to NULL. If curproc is NULL
* or curproc->p_vmspace is NULL the fault is fatal.
*/
if (lp != NULL)
vm = lp->lwp_vmspace;
if (vm == NULL) {
fault_flags = -1;
ftype = -1;
goto nogo;
}
/*
* Debugging, try to catch kernel faults on the user address space when not inside
* on onfault (e.g. copyin/copyout) routine.
*/
if (usermode == 0 && (td->td_pcb == NULL || td->td_pcb->pcb_onfault == NULL)) {
#ifdef DDB
if (freeze_on_seg_fault) {
kprintf("trap_pfault: user address fault from kernel mode "
"%016lx\n", (long)frame->tf_addr);
while (freeze_on_seg_fault)
tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20);
}
#endif
}
map = &vm->vm_map;
}
/*
* PGEX_I is defined only if the execute disable bit capability is
* supported and enabled.
*/
if (frame->tf_err & PGEX_W)
ftype = VM_PROT_WRITE;
#if JG
else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
ftype = VM_PROT_EXECUTE;
#endif
else
ftype = VM_PROT_READ;
if (map != &kernel_map) {
/*
* Keep swapout from messing with us during this
* critical time.
*/
PHOLD(lp->lwp_proc);
/*
* Issue fault
*/
fault_flags = 0;
if (usermode)
fault_flags |= VM_FAULT_BURST;
if (ftype & VM_PROT_WRITE)
fault_flags |= VM_FAULT_DIRTY;
else
fault_flags |= VM_FAULT_NORMAL;
rv = vm_fault(map, va, ftype, fault_flags);
PRELE(lp->lwp_proc);
} else {
/*
* Don't have to worry about process locking or stacks in the
* kernel.
*/
fault_flags = VM_FAULT_NORMAL;
rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
}
if (rv == KERN_SUCCESS)
return (0);
nogo:
if (!usermode) {
if (td->td_gd->gd_intr_nesting_level == 0 &&
td->td_pcb->pcb_onfault) {
frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
return (0);
}
trap_fatal(frame, frame->tf_addr);
return (-1);
}
/*
* NOTE: on x86_64 we have a tf_addr field in the trapframe, no
* kludge is needed to pass the fault address to signal handlers.
*/
p = td->td_proc;
if (td->td_lwp->lwp_vkernel == NULL) {
#ifdef DDB
if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) {
#else
if (bootverbose) {
#endif
kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
"pid=%d cpu=%d p_comm=%s\n",
ftype, fault_flags,
(void *)frame->tf_addr,
(void *)frame->tf_rip,
p->p_pid, mycpu->gd_cpuid, p->p_comm);
}
#ifdef DDB
while (freeze_on_seg_fault) {
tsleep(p, 0, "freeze", hz * 20);
}
if (ddb_on_seg_fault)
Debugger("ddb_on_seg_fault");
#endif
}
return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
}
static void
trap_fatal(struct trapframe *frame, vm_offset_t eva)
{
int code, ss;
u_int type;
long rsp;
struct soft_segment_descriptor softseg;
char *msg;
code = frame->tf_err;
type = frame->tf_trapno;
sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
if (type <= MAX_TRAP_MSG)
msg = trap_msg[type];
else
msg = "UNKNOWN";
kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
#ifdef SMP
/* three separate prints in case of a trap on an unmapped page */
kprintf("cpuid = %d; ", mycpu->gd_cpuid);
kprintf("lapic->id = %08x\n", lapic->id);
#endif
if (type == T_PAGEFLT) {
kprintf("fault virtual address = 0x%lx\n", eva);
kprintf("fault code = %s %s %s, %s\n",
code & PGEX_U ? "user" : "supervisor",
code & PGEX_W ? "write" : "read",
code & PGEX_I ? "instruction" : "data",
code & PGEX_P ? "protection violation" : "page not present");
}
kprintf("instruction pointer = 0x%lx:0x%lx\n",
frame->tf_cs & 0xffff, frame->tf_rip);
if (ISPL(frame->tf_cs) == SEL_UPL) {
ss = frame->tf_ss & 0xffff;
rsp = frame->tf_rsp;
} else {
ss = GSEL(GDATA_SEL, SEL_KPL);
rsp = (long)&frame->tf_rsp;
}
kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
softseg.ssd_gran);
kprintf("processor eflags = ");
if (frame->tf_rflags & PSL_T)
kprintf("trace trap, ");
if (frame->tf_rflags & PSL_I)
kprintf("interrupt enabled, ");
if (frame->tf_rflags & PSL_NT)
kprintf("nested task, ");
if (frame->tf_rflags & PSL_RF)
kprintf("resume, ");
kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
kprintf("current process = ");
if (curproc) {
kprintf("%lu\n",
(u_long)curproc->p_pid);
} else {
kprintf("Idle\n");
}
kprintf("current thread = pri %d ", curthread->td_pri);
if (curthread->td_critcount)
kprintf("(CRIT)");
kprintf("\n");
#ifdef DDB
if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
return;
#endif
kprintf("trap number = %d\n", type);
if (type <= MAX_TRAP_MSG)
panic("%s", trap_msg[type]);
else
panic("unknown/reserved trap");
}
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
}
/*
* syscall2 - MP aware system call request C handler
*
* A system call is essentially treated as a trap. The MP lock is not
* held on entry or return. We are responsible for handling ASTs
* (e.g. a task switch) prior to return.
*
* MPSAFE
*/
void
syscall2(struct trapframe *frame)
{
struct thread *td = curthread;
struct proc *p = td->td_proc;
struct lwp *lp = td->td_lwp;
caddr_t params;
struct sysent *callp;
register_t orig_tf_eflags;
int sticks;
int error;
int narg;
#ifdef INVARIANTS
int crit_count = td->td_critcount;
#endif
int have_mplock = 0;
u_int code;
union sysunion args;
#ifdef DIAGNOSTIC
if (ISPL(frame->tf_cs) != SEL_UPL) {
get_mplock();
panic("syscall");
/* NOT REACHED */
}
#endif
KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
frame->tf_eax);
userenter(td, p); /* lazy raise our priority */
/*
* Misc
*/
sticks = (int)td->td_sticks;
orig_tf_eflags = frame->tf_eflags;
/*
* Virtual kernel intercept - if a VM context managed by a virtual
* kernel issues a system call the virtual kernel handles it, not us.
* Restore the virtual kernel context and return from its system
* call. The current frame is copied out to the virtual kernel.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
vkernel_trap(lp, frame);
error = EJUSTRETURN;
callp = NULL;
goto out;
}
/*
* Get the system call parameters and account for time
*/
lp->lwp_md.md_regs = frame;
params = (caddr_t)frame->tf_esp + sizeof(int);
code = frame->tf_eax;
if (p->p_sysent->sv_prepsyscall) {
(*p->p_sysent->sv_prepsyscall)(
frame, (int *)(&args.nosys.sysmsg + 1),
&code, ¶ms);
} else {
/*
* Need to check if this is a 32 bit or 64 bit syscall.
* fuword is MP aware.
*/
if (code == SYS_syscall) {
/*
* Code is first argument, followed by actual args.
*/
code = fuword(params);
params += sizeof(int);
} else if (code == SYS___syscall) {
/*
* Like syscall, but code is a quad, so as to maintain
* quad alignment for the rest of the arguments.
*/
code = fuword(params);
params += sizeof(quad_t);
}
}
code &= p->p_sysent->sv_mask;
if (code >= p->p_sysent->sv_size)
callp = &p->p_sysent->sv_table[0];
else
callp = &p->p_sysent->sv_table[code];
narg = callp->sy_narg & SYF_ARGMASK;
#if 0
if (p->p_sysent->sv_name[0] == 'L')
kprintf("Linux syscall, code = %d\n", code);
#endif
/*
* copyin is MP aware, but the tracing code is not
*/
if (narg && params) {
error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
narg * sizeof(register_t));
if (error) {
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSCALL)) {
MAKEMPSAFE(have_mplock);
ktrsyscall(lp, code, narg,
(void *)(&args.nosys.sysmsg + 1));
}
#endif
goto bad;
}
}
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSCALL)) {
MAKEMPSAFE(have_mplock);
ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
}
#endif
/*
* For traditional syscall code edx is left untouched when 32 bit
* results are returned. Since edx is loaded from fds[1] when the
* system call returns we pre-set it here.
*/
args.sysmsg_fds[0] = 0;
args.sysmsg_fds[1] = frame->tf_edx;
/*
* The syscall might manipulate the trap frame. If it does it
* will probably return EJUSTRETURN.
*/
args.sysmsg_frame = frame;
STOPEVENT(p, S_SCE, narg); /* MP aware */
/*
* NOTE: All system calls run MPSAFE now. The system call itself
* is responsible for getting the MP lock.
*/
error = (*callp->sy_call)(&args);
out:
/*
* MP SAFE (we may or may not have the MP lock at this point)
*/
switch (error) {
case 0:
/*
* Reinitialize proc pointer `p' as it may be different
* if this is a child returning from fork syscall.
*/
p = curproc;
lp = curthread->td_lwp;
frame->tf_eax = args.sysmsg_fds[0];
frame->tf_edx = args.sysmsg_fds[1];
frame->tf_eflags &= ~PSL_C;
break;
case ERESTART:
/*
* Reconstruct pc, assuming lcall $X,y is 7 bytes,
* int 0x80 is 2 bytes. We saved this in tf_err.
*/
frame->tf_eip -= frame->tf_err;
break;
case EJUSTRETURN:
break;
case EASYNC:
panic("Unexpected EASYNC return value (for now)");
default:
bad:
if (p->p_sysent->sv_errsize) {
if (error >= p->p_sysent->sv_errsize)
error = -1; /* XXX */
else
error = p->p_sysent->sv_errtbl[error];
}
frame->tf_eax = error;
frame->tf_eflags |= PSL_C;
break;
}
/*
* Traced syscall. trapsignal() is not MP aware.
*/
if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
MAKEMPSAFE(have_mplock);
frame->tf_eflags &= ~PSL_T;
trapsignal(lp, SIGTRAP, TRAP_TRACE);
}
/*
* Handle reschedule and other end-of-syscall issues
*/
userret(lp, frame, sticks);
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSRET)) {
MAKEMPSAFE(have_mplock);
ktrsysret(lp, code, error, args.sysmsg_result);
}
#endif
/*
* This works because errno is findable through the
* register set. If we ever support an emulation where this
* is not the case, this code will need to be revisited.
*/
STOPEVENT(p, S_SCX, code);
userexit(lp);
/*
* Release the MP lock if we had to get it
*/
if (have_mplock)
rel_mplock();
KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
#ifdef INVARIANTS
KASSERT(crit_count == td->td_critcount,
("syscall: critical section count mismatch! %d/%d",
crit_count, td->td_pri));
KASSERT(&td->td_toks_base == td->td_toks_stop,
("syscall: extra tokens held after trap! %zd",
td->td_toks_stop - &td->td_toks_base));
#endif
}
static void
trap_fatal(struct trapframe *frame, vm_offset_t eva)
{
int code, type, ss, esp;
struct soft_segment_descriptor softseg;
code = frame->tf_err;
type = frame->tf_trapno;
sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
if (type <= MAX_TRAP_MSG)
kprintf("\n\nFatal trap %d: %s while in %s mode\n",
type, trap_msg[type],
frame->tf_eflags & PSL_VM ? "vm86" :
ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
/* three separate prints in case of a trap on an unmapped page */
kprintf("cpuid = %d; ", mycpu->gd_cpuid);
kprintf("lapic.id = %08x\n", lapic->id);
if (type == T_PAGEFLT) {
kprintf("fault virtual address = %p\n", (void *)eva);
kprintf("fault code = %s %s, %s\n",
code & PGEX_U ? "user" : "supervisor",
code & PGEX_W ? "write" : "read",
code & PGEX_P ? "protection violation" : "page not present");
}
kprintf("instruction pointer = 0x%x:0x%x\n",
frame->tf_cs & 0xffff, frame->tf_eip);
if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
ss = frame->tf_ss & 0xffff;
esp = frame->tf_esp;
} else {
ss = GSEL(GDATA_SEL, SEL_KPL);
esp = (int)&frame->tf_esp;
}
kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
softseg.ssd_gran);
kprintf("processor eflags = ");
if (frame->tf_eflags & PSL_T)
kprintf("trace trap, ");
if (frame->tf_eflags & PSL_I)
kprintf("interrupt enabled, ");
if (frame->tf_eflags & PSL_NT)
kprintf("nested task, ");
if (frame->tf_eflags & PSL_RF)
kprintf("resume, ");
if (frame->tf_eflags & PSL_VM)
kprintf("vm86, ");
kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
kprintf("current process = ");
if (curproc) {
kprintf("%lu (%s)\n",
(u_long)curproc->p_pid, curproc->p_comm ?
curproc->p_comm : "");
} else {
kprintf("Idle\n");
}
kprintf("current thread = pri %d ", curthread->td_pri);
if (curthread->td_critcount)
kprintf("(CRIT)");
kprintf("\n");
/**
* XXX FIXME:
* we probably SHOULD have stopped the other CPUs before now!
* another CPU COULD have been touching cpl at this moment...
*/
kprintf(" <- SMP: XXX");
kprintf("\n");
#ifdef KDB
if (kdb_trap(&psl))
return;
#endif
#ifdef DDB
if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
return;
#endif
kprintf("trap number = %d\n", type);
if (type <= MAX_TRAP_MSG)
panic("%s", trap_msg[type]);
else
panic("unknown/reserved trap");
}
/*
* kdb_trap - field a TRACE or BPT trap
*/
int
kdb_trap(int type, int code, struct i386_saved_state *regs)
{
volatile int ddb_mode = !(boothowto & RB_GDB);
/*
* XXX try to do nothing if the console is in graphics mode.
* Handle trace traps (and hardware breakpoints...) by ignoring
* them except for forgetting about them. Return 0 for other
* traps to say that we haven't done anything. The trap handler
* will usually panic. We should handle breakpoint traps for
* our breakpoints by disarming our breakpoints and fixing up
* %eip.
*/
if (cons_unavail && ddb_mode) {
if (type == T_TRCTRAP) {
regs->tf_eflags &= ~PSL_T;
return (1);
}
return (0);
}
switch (type) {
case T_BPTFLT: /* breakpoint */
case T_TRCTRAP: /* debug exception */
break;
default:
/*
* XXX this is almost useless now. In most cases,
* trap_fatal() has already printed a much more verbose
* message. However, it is dangerous to print things in
* trap_fatal() - kprintf() might be reentered and trap.
* The debugger should be given control first.
*/
if (ddb_mode)
db_printf("kernel: type %d trap, code=%x\n", type, code);
if (db_nofault) {
jmp_buf *no_fault = db_nofault;
db_nofault = NULL;
longjmp(*no_fault, 1);
}
}
/*
* This handles unexpected traps in ddb commands, including calls to
* non-ddb functions. db_nofault only applies to memory accesses by
* internal ddb commands.
*/
if (db_global_jmpbuf_valid)
longjmp(db_global_jmpbuf, 1);
/*
* XXX We really should switch to a local stack here.
*/
ddb_regs = *regs;
/*
* If in kernel mode, esp and ss are not saved, so dummy them up.
*/
if (ISPL(regs->tf_cs) == 0) {
ddb_regs.tf_esp = (int)®s->tf_esp;
ddb_regs.tf_ss = rss();
}
crit_enter();
#ifdef SMP
db_printf("\nCPU%d stopping CPUs: 0x%08x\n",
mycpu->gd_cpuid, mycpu->gd_other_cpus);
/* We stop all CPUs except ourselves (obviously) */
stop_cpus(mycpu->gd_other_cpus);
db_printf(" stopped\n");
#endif /* SMP */
setjmp(db_global_jmpbuf);
db_global_jmpbuf_valid = TRUE;
db_active++;
if (ddb_mode) {
cndbctl(TRUE);
db_trap(type, code);
cndbctl(FALSE);
} else
gdb_handle_exception(&ddb_regs, type, code);
db_active--;
db_global_jmpbuf_valid = FALSE;
#ifdef SMP
db_printf("\nCPU%d restarting CPUs: 0x%08x\n",
mycpu->gd_cpuid, stopped_cpus);
/* Restart all the CPUs we previously stopped */
if (stopped_cpus != mycpu->gd_other_cpus) {
db_printf("whoa, other_cpus: 0x%08x, stopped_cpus: 0x%08x\n",
mycpu->gd_other_cpus, stopped_cpus);
panic("stop_cpus() failed");
}
restart_cpus(stopped_cpus);
db_printf(" restarted\n");
#endif /* SMP */
crit_exit();
regs->tf_eip = ddb_regs.tf_eip;
regs->tf_eflags = ddb_regs.tf_eflags;
regs->tf_eax = ddb_regs.tf_eax;
regs->tf_ecx = ddb_regs.tf_ecx;
regs->tf_edx = ddb_regs.tf_edx;
regs->tf_ebx = ddb_regs.tf_ebx;
/*
* If in user mode, the saved ESP and SS were valid, restore them.
*/
if (ISPL(regs->tf_cs)) {
regs->tf_esp = ddb_regs.tf_esp;
regs->tf_ss = ddb_regs.tf_ss & 0xffff;
}
regs->tf_ebp = ddb_regs.tf_ebp;
regs->tf_esi = ddb_regs.tf_esi;
regs->tf_edi = ddb_regs.tf_edi;
regs->tf_es = ddb_regs.tf_es & 0xffff;
regs->tf_fs = ddb_regs.tf_fs & 0xffff;
regs->tf_gs = ddb_regs.tf_gs & 0xffff;
regs->tf_cs = ddb_regs.tf_cs & 0xffff;
regs->tf_ds = ddb_regs.tf_ds & 0xffff;
return (1);
}
void
trap(struct trapframe *frame)
{
struct thread *td = curthread;
struct proc *p = td->td_proc;
int i = 0, ucode = 0, code;
u_int type;
register_t addr = 0;
vm_offset_t eva;
ksiginfo_t ksi;
#ifdef POWERFAIL_NMI
static int lastalert = 0;
#endif
PCPU_INC(cnt.v_trap);
type = frame->tf_trapno;
#ifdef SMP
/* Handler for NMI IPIs used for stopping CPUs. */
if (type == T_NMI) {
if (ipi_nmi_handler() == 0)
goto out;
}
#endif /* SMP */
#ifdef KDB
if (kdb_active) {
kdb_reenter();
goto out;
}
#endif
if (type == T_RESERVED) {
trap_fatal(frame, 0);
goto out;
}
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI so we check for that first.
* If the HWPMC module is active, 'pmc_hook' will point to
* the function to be called. A return value of '1' from the
* hook means that the NMI was handled by it and that we can
* return immediately.
*/
if (type == T_NMI && pmc_intr &&
(*pmc_intr)(PCPU_GET(cpuid), frame))
goto out;
#endif
if (type == T_MCHK) {
if (!mca_intr())
trap_fatal(frame, 0);
goto out;
}
#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 it's per-cpu flags to indicate that it doesn't
* want to fault. On returning from 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))
goto out;
if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
type == T_BPTFLT) {
struct reg regs;
fill_frame_regs(frame, ®s);
if (type == T_DTRACE_PROBE &&
dtrace_fasttrap_probe_ptr != NULL &&
dtrace_fasttrap_probe_ptr(®s) == 0)
goto out;
if (type == T_BPTFLT &&
dtrace_pid_probe_ptr != NULL &&
dtrace_pid_probe_ptr(®s) == 0)
goto out;
if (type == T_DTRACE_RET &&
dtrace_return_probe_ptr != NULL &&
dtrace_return_probe_ptr(®s) == 0)
goto out;
}
#endif
if ((frame->tf_eflags & PSL_I) == 0) {
/*
* 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.
*/
if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
uprintf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curthread->td_name, type);
else if (type != T_BPTFLT && type != T_TRCTRAP &&
frame->tf_eip != (int)cpu_switch_load_gs) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
printf("kernel trap %d with interrupts disabled\n",
type);
/*
* Page faults need interrupts disabled until later,
* and we shouldn't enable interrupts while holding
* a spin lock or if servicing an NMI.
*/
if (type != T_NMI && type != T_PAGEFLT &&
td->td_md.md_spinlock_count == 0)
enable_intr();
}
}
eva = 0;
code = frame->tf_err;
if (type == 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.
*
* If we get a page fault while in a critical section, then
* it is most likely a fatal kernel page fault. The kernel
* is already going to panic trying to get a sleep lock to
* do the VM lookup, so just consider it a fatal trap so the
* kernel can print out a useful trap message and even get
* to the debugger.
*
* If we get a page fault while holding a non-sleepable
* lock, then it is most likely a fatal kernel page fault.
* If WITNESS is enabled, then it's going to whine about
* bogus LORs with various VM locks, so just skip to the
* fatal trap handling directly.
*/
eva = rcr2();
if (td->td_critnest != 0 ||
WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
"Kernel page fault") != 0)
trap_fatal(frame, eva);
else
enable_intr();
}
if ((ISPL(frame->tf_cs) == SEL_UPL) ||
((frame->tf_eflags & PSL_VM) &&
!(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL))) {
/* user trap */
td->td_pticks = 0;
td->td_frame = frame;
addr = frame->tf_eip;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
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 */
enable_intr();
frame->tf_eflags &= ~PSL_T;
i = SIGTRAP;
ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
break;
case T_ARITHTRAP: /* arithmetic trap */
#ifdef DEV_NPX
ucode = npxtrap();
if (ucode == -1)
goto userout;
#else
ucode = 0;
#endif
i = SIGFPE;
break;
/*
* 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 user;
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 */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_DOUBLEFLT: /* double fault */
default:
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_PAGEFLT: /* page fault */
i = trap_pfault(frame, TRUE, eva);
#if defined(I586_CPU) && !defined(NO_F00F_HACK)
if (i == -2) {
/*
* The f00f hack workaround has triggered, so
* treat the fault as an illegal instruction
* (T_PRIVINFLT) instead of a page fault.
*/
type = frame->tf_trapno = T_PRIVINFLT;
/* Proceed as in that case. */
ucode = ILL_PRVOPC;
i = SIGILL;
break;
}
#endif
if (i == -1)
goto userout;
if (i == 0)
goto user;
if (i == SIGSEGV)
ucode = SEGV_MAPERR;
else {
if (prot_fault_translation == 0) {
/*
* Autodetect.
* This check also covers the images
* without the ABI-tag ELF note.
*/
if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
&& p->p_osrel >= P_OSREL_SIGSEGV) {
i = SIGSEGV;
ucode = SEGV_ACCERR;
} else {
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
}
} else if (prot_fault_translation == 1) {
/*
* Always compat mode.
*/
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
} else {
/*
* Always SIGSEGV mode.
*/
i = SIGSEGV;
ucode = SEGV_ACCERR;
}
}
addr = eva;
break;
case T_DIVIDE: /* integer divide fault */
ucode = FPE_INTDIV;
i = SIGFPE;
break;
#ifdef DEV_ISA
case T_NMI:
#ifdef POWERFAIL_NMI
#ifndef TIMER_FREQ
# define TIMER_FREQ 1193182
#endif
if (time_second - lastalert > 10) {
log(LOG_WARNING, "NMI: power fail\n");
sysbeep(880, hz);
lastalert = time_second;
}
goto userout;
#else /* !POWERFAIL_NMI */
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto userout;
} else if (panic_on_nmi)
panic("NMI indicates hardware failure");
break;
#endif /* POWERFAIL_NMI */
#endif /* DEV_ISA */
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:
#ifdef DEV_NPX
KASSERT(PCB_USER_FPU(td->td_pcb),
("kernel FPU ctx has leaked"));
/* transparent fault (due to context switch "late") */
if (npxdna())
goto userout;
#endif
uprintf("pid %d killed due to lack of floating point\n",
p->p_pid);
i = SIGKILL;
ucode = 0;
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 */
KASSERT(cold || td->td_ucred != NULL,
("kernel trap doesn't have ucred"));
switch (type) {
case T_PAGEFLT: /* page fault */
(void) trap_pfault(frame, FALSE, eva);
goto out;
case T_DNA:
#ifdef DEV_NPX
KASSERT(!PCB_USER_FPU(td->td_pcb),
("Unregistered use of FPU in kernel"));
if (npxdna())
goto out;
#endif
break;
case T_ARITHTRAP: /* arithmetic trap */
case T_XMMFLT: /* SIMD floating-point exception */
case T_FPOPFLT: /* FPU operand fetch fault */
/*
* XXXKIB for now disable any FPU traps in kernel
* handler registration seems to be overkill
*/
trap_fatal(frame, 0);
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)
/*
* returns to original process
*/
vm86_trap((struct vm86frame *)frame);
goto out;
}
if (type == T_STKFLT)
break;
/* FALL THROUGH */
case T_SEGNPFLT: /* segment not present fault */
if (PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)
break;
/*
* 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.
*/
if (frame->tf_eip == (int)cpu_switch_load_gs) {
PCPU_GET(curpcb)->pcb_gs = 0;
#if 0
PROC_LOCK(p);
kern_psignal(p, SIGBUS);
PROC_UNLOCK(p);
#endif
goto out;
}
if (td->td_intr_nesting_level != 0)
break;
/*
* 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.
*/
if (frame->tf_eip == (int)doreti_iret) {
frame->tf_eip = (int)doreti_iret_fault;
goto out;
}
if (frame->tf_eip == (int)doreti_popl_ds) {
frame->tf_eip = (int)doreti_popl_ds_fault;
goto out;
}
if (frame->tf_eip == (int)doreti_popl_es) {
frame->tf_eip = (int)doreti_popl_es_fault;
goto out;
}
if (frame->tf_eip == (int)doreti_popl_fs) {
frame->tf_eip = (int)doreti_popl_fs_fault;
goto out;
}
if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
frame->tf_eip =
(int)PCPU_GET(curpcb)->pcb_onfault;
goto out;
}
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 out;
}
break;
case T_TRCTRAP: /* trace trap */
if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
/*
* We've just entered system mode via the
* syscall lcall. Continue single stepping
* silently until the syscall handler has
* saved the flags.
*/
goto out;
}
if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
/*
* The syscall handler has now saved the
* flags. Stop single stepping it.
*/
frame->tf_eflags &= ~PSL_T;
goto out;
}
/*
* 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() &&
!(PCPU_GET(curpcb)->pcb_flags & PCB_VM86CALL)) {
/*
* Reset breakpoint bits because the
* processor doesn't
*/
load_dr6(rdr6() & 0xfffffff0);
goto out;
}
/*
* FALLTHROUGH (TRCTRAP kernel mode, kernel address)
*/
case T_BPTFLT:
/*
* If KDB is enabled, let it handle the debugger trap.
* Otherwise, debugger traps "can't happen".
*/
#ifdef KDB
if (kdb_trap(type, 0, frame))
goto out;
#endif
break;
#ifdef DEV_ISA
case T_NMI:
#ifdef POWERFAIL_NMI
if (time_second - lastalert > 10) {
log(LOG_WARNING, "NMI: power fail\n");
sysbeep(880, hz);
lastalert = time_second;
}
goto out;
#else /* !POWERFAIL_NMI */
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto out;
} else if (panic_on_nmi == 0)
goto out;
/* FALLTHROUGH */
#endif /* POWERFAIL_NMI */
#endif /* DEV_ISA */
}
trap_fatal(frame, eva);
goto out;
}
/* Translate fault for emulators (e.g. Linux) */
if (*p->p_sysent->sv_transtrap)
i = (*p->p_sysent->sv_transtrap)(i, type);
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = i;
ksi.ksi_code = ucode;
ksi.ksi_addr = (void *)addr;
ksi.ksi_trapno = type;
trapsignal(td, &ksi);
#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
user:
userret(td, frame);
mtx_assert(&Giant, MA_NOTOWNED);
KASSERT(PCB_USER_FPU(td->td_pcb),
("Return from trap with kernel FPU ctx leaked"));
userout:
out:
return;
}
/*
* syscall2 - MP aware system call request C handler
*
* A system call is essentially treated as a trap except that the
* MP lock is not held on entry or return. We are responsible for
* obtaining the MP lock if necessary and for handling ASTs
* (e.g. a task switch) prior to return.
*
* MPSAFE
*/
void
syscall2(struct trapframe *frame)
{
struct thread *td = curthread;
struct proc *p = td->td_proc;
struct lwp *lp = td->td_lwp;
struct sysent *callp;
register_t orig_tf_rflags;
int sticks;
int error;
int narg;
#ifdef INVARIANTS
int crit_count = td->td_critcount;
#endif
register_t *argp;
u_int code;
int regcnt, optimized_regcnt;
union sysunion args;
register_t *argsdst;
mycpu->gd_cnt.v_syscall++;
#ifdef DIAGNOSTIC
if (ISPL(frame->tf_cs) != SEL_UPL) {
panic("syscall");
/* NOT REACHED */
}
#endif
KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
frame->tf_rax);
userenter(td, p); /* lazy raise our priority */
regcnt = 6;
optimized_regcnt = 6;
/*
* Misc
*/
sticks = (int)td->td_sticks;
orig_tf_rflags = frame->tf_rflags;
/*
* Virtual kernel intercept - if a VM context managed by a virtual
* kernel issues a system call the virtual kernel handles it, not us.
* Restore the virtual kernel context and return from its system
* call. The current frame is copied out to the virtual kernel.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
vkernel_trap(lp, frame);
error = EJUSTRETURN;
callp = NULL;
code = 0;
goto out;
}
/*
* Get the system call parameters and account for time
*/
KASSERT(lp->lwp_md.md_regs == frame,
("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
code = (u_int)frame->tf_rax;
if (code == SYS_syscall || code == SYS___syscall) {
code = frame->tf_rdi;
regcnt--;
argp = &frame->tf_rdi + 1;
} else {
argp = &frame->tf_rdi;
}
if (code >= p->p_sysent->sv_size)
callp = &p->p_sysent->sv_table[0];
else
callp = &p->p_sysent->sv_table[code];
narg = callp->sy_narg & SYF_ARGMASK;
/*
* On x86_64 we get up to six arguments in registers. The rest are
* on the stack. The first six members of 'struct trapframe' happen
* to be the registers used to pass arguments, in exactly the right
* order.
*/
argsdst = (register_t *)(&args.nosys.sysmsg + 1);
/*
* Its easier to copy up to the highest number of syscall arguments
* passed in registers, which is 6, than to conditionalize it.
*/
bcopy(argp, argsdst, sizeof(register_t) * optimized_regcnt);
/*
* Any arguments beyond available argument-passing registers must
* be copyin()'d from the user stack.
*/
if (narg > regcnt) {
caddr_t params;
params = (caddr_t)frame->tf_rsp + sizeof(register_t);
error = copyin(params, &argsdst[regcnt],
(narg - regcnt) * sizeof(register_t));
if (error) {
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSCALL)) {
ktrsyscall(lp, code, narg,
(void *)(&args.nosys.sysmsg + 1));
}
#endif
goto bad;
}
}
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSCALL)) {
ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
}
#endif
/*
* Default return value is 0 (will be copied to %rax). Double-value
* returns use %rax and %rdx. %rdx is left unchanged for system
* calls which return only one result.
*/
args.sysmsg_fds[0] = 0;
args.sysmsg_fds[1] = frame->tf_rdx;
/*
* The syscall might manipulate the trap frame. If it does it
* will probably return EJUSTRETURN.
*/
args.sysmsg_frame = frame;
STOPEVENT(p, S_SCE, narg); /* MP aware */
/*
* NOTE: All system calls run MPSAFE now. The system call itself
* is responsible for getting the MP lock.
*/
#ifdef SYSCALL_DEBUG
tsc_uclock_t tscval = rdtsc();
#endif
error = (*callp->sy_call)(&args);
#ifdef SYSCALL_DEBUG
tscval = rdtsc() - tscval;
tscval = tscval * 1000000 / tsc_frequency;
if (SysCallsWorstCase[code] < tscval)
SysCallsWorstCase[code] = tscval;
#endif
out:
/*
* MP SAFE (we may or may not have the MP lock at this point)
*/
//kprintf("SYSMSG %d ", error);
switch (error) {
case 0:
/*
* Reinitialize proc pointer `p' as it may be different
* if this is a child returning from fork syscall.
*/
p = curproc;
lp = curthread->td_lwp;
frame->tf_rax = args.sysmsg_fds[0];
frame->tf_rdx = args.sysmsg_fds[1];
frame->tf_rflags &= ~PSL_C;
break;
case ERESTART:
/*
* Reconstruct pc, we know that 'syscall' is 2 bytes.
* We have to do a full context restore so that %r10
* (which was holding the value of %rcx) is restored for
* the next iteration.
*/
if (frame->tf_err != 0 && frame->tf_err != 2)
kprintf("lp %s:%d frame->tf_err is weird %ld\n",
td->td_comm, lp->lwp_proc->p_pid, frame->tf_err);
frame->tf_rip -= frame->tf_err;
frame->tf_r10 = frame->tf_rcx;
break;
case EJUSTRETURN:
break;
case EASYNC:
panic("Unexpected EASYNC return value (for now)");
default:
bad:
if (p->p_sysent->sv_errsize) {
if (error >= p->p_sysent->sv_errsize)
error = -1; /* XXX */
else
error = p->p_sysent->sv_errtbl[error];
}
frame->tf_rax = error;
frame->tf_rflags |= PSL_C;
break;
}
/*
* Traced syscall. trapsignal() should now be MP aware
*/
if (orig_tf_rflags & PSL_T) {
frame->tf_rflags &= ~PSL_T;
trapsignal(lp, SIGTRAP, TRAP_TRACE);
}
/*
* Handle reschedule and other end-of-syscall issues
*/
userret(lp, frame, sticks);
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSRET)) {
ktrsysret(lp, code, error, args.sysmsg_result);
}
#endif
/*
* This works because errno is findable through the
* register set. If we ever support an emulation where this
* is not the case, this code will need to be revisited.
*/
STOPEVENT(p, S_SCX, code);
userexit(lp);
KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
#ifdef INVARIANTS
KASSERT(crit_count == td->td_critcount,
("syscall: critical section count mismatch! %d/%d",
crit_count, td->td_pri));
KASSERT(&td->td_toks_base == td->td_toks_stop,
("syscall: %ld extra tokens held after trap! syscall %p",
td->td_toks_stop - &td->td_toks_base,
callp->sy_call));
#endif
}
void
trap(struct trapframe *frame)
{
#ifdef KDTRACE_HOOKS
struct reg regs;
#endif
struct thread *td = curthread;
struct proc *p = td->td_proc;
int i = 0, ucode = 0, code;
u_int type;
register_t addr = 0;
ksiginfo_t ksi;
PCPU_INC(cnt.v_trap);
type = frame->tf_trapno;
#ifdef SMP
/* Handler for NMI IPIs used for stopping CPUs. */
if (type == T_NMI) {
if (ipi_nmi_handler() == 0)
goto out;
}
#endif /* SMP */
#ifdef KDB
if (kdb_active) {
kdb_reenter();
goto out;
}
#endif
if (type == T_RESERVED) {
trap_fatal(frame, 0);
goto out;
}
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI. If the PMC module is
* active, pass the 'rip' value to the PMC module's interrupt
* handler. A return value of '1' from the handler means that
* the NMI was handled by it and we can return immediately.
*/
if (type == T_NMI && pmc_intr &&
(*pmc_intr)(PCPU_GET(cpuid), frame))
goto out;
#endif
if (type == T_MCHK) {
mca_intr();
goto out;
}
#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 probe, the no-fault
* flag is cleared and finally re-scheduling is enabled.
*/
if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
goto out;
#endif
if ((frame->tf_rflags & PSL_I) == 0) {
/*
* 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.
*/
if (ISPL(frame->tf_cs) == SEL_UPL)
uprintf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curthread->td_name, type);
else if (type != T_NMI && type != T_BPTFLT &&
type != T_TRCTRAP) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
printf("kernel trap %d with interrupts disabled\n",
type);
/*
* We shouldn't enable interrupts while holding a
* spin lock.
*/
if (td->td_md.md_spinlock_count == 0)
enable_intr();
}
}
code = frame->tf_err;
if (ISPL(frame->tf_cs) == SEL_UPL) {
/* user trap */
td->td_pticks = 0;
td->td_frame = frame;
addr = frame->tf_rip;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
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 */
enable_intr();
#ifdef KDTRACE_HOOKS
if (type == T_BPTFLT) {
fill_frame_regs(frame, ®s);
if (dtrace_pid_probe_ptr != NULL &&
dtrace_pid_probe_ptr(®s) == 0)
goto out;
}
#endif
frame->tf_rflags &= ~PSL_T;
i = SIGTRAP;
ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
break;
case T_ARITHTRAP: /* arithmetic trap */
ucode = fputrap_x87();
if (ucode == -1)
goto userout;
i = SIGFPE;
break;
case T_PROTFLT: /* general protection fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_STKFLT: /* stack fault */
case T_SEGNPFLT: /* segment not present fault */
i = SIGBUS;
ucode = BUS_ADRERR;
break;
case T_TSSFLT: /* invalid TSS fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_DOUBLEFLT: /* double fault */
default:
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_PAGEFLT: /* page fault */
addr = frame->tf_addr;
i = trap_pfault(frame, TRUE);
if (i == -1)
goto userout;
if (i == 0)
goto user;
if (i == SIGSEGV)
ucode = SEGV_MAPERR;
else {
if (prot_fault_translation == 0) {
/*
* Autodetect.
* This check also covers the images
* without the ABI-tag ELF note.
*/
if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
&& p->p_osrel >= P_OSREL_SIGSEGV) {
i = SIGSEGV;
ucode = SEGV_ACCERR;
} else {
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
}
} else if (prot_fault_translation == 1) {
/*
* Always compat mode.
*/
i = SIGBUS;
ucode = BUS_PAGE_FAULT;
} else {
/*
* Always SIGSEGV mode.
*/
i = SIGSEGV;
ucode = SEGV_ACCERR;
}
}
break;
case T_DIVIDE: /* integer divide fault */
ucode = FPE_INTDIV;
i = SIGFPE;
break;
#ifdef DEV_ISA
case T_NMI:
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto userout;
} else if (panic_on_nmi)
panic("NMI indicates hardware failure");
break;
#endif /* DEV_ISA */
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:
/* transparent fault (due to context switch "late") */
KASSERT(PCB_USER_FPU(td->td_pcb),
("kernel FPU ctx has leaked"));
fpudna();
goto userout;
case T_FPOPFLT: /* FPU operand fetch fault */
ucode = ILL_COPROC;
i = SIGILL;
break;
case T_XMMFLT: /* SIMD floating-point exception */
ucode = fputrap_sse();
if (ucode == -1)
goto userout;
i = SIGFPE;
break;
#ifdef KDTRACE_HOOKS
case T_DTRACE_RET:
enable_intr();
fill_frame_regs(frame, ®s);
if (dtrace_return_probe_ptr != NULL &&
dtrace_return_probe_ptr(®s) == 0)
goto out;
break;
#endif
}
} else {
/* kernel trap */
KASSERT(cold || td->td_ucred != NULL,
("kernel trap doesn't have ucred"));
switch (type) {
case T_PAGEFLT: /* page fault */
(void) trap_pfault(frame, FALSE);
goto out;
case T_DNA:
KASSERT(!PCB_USER_FPU(td->td_pcb),
("Unregistered use of FPU in kernel"));
fpudna();
goto out;
case T_ARITHTRAP: /* arithmetic trap */
case T_XMMFLT: /* SIMD floating-point exception */
case T_FPOPFLT: /* FPU operand fetch fault */
/*
* XXXKIB for now disable any FPU traps in kernel
* handler registration seems to be overkill
*/
trap_fatal(frame, 0);
goto out;
case T_STKFLT: /* stack fault */
break;
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
if (td->td_intr_nesting_level != 0)
break;
/*
* Invalid segment selectors and out of bounds
* %rip's and %rsp'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.
*/
if (frame->tf_rip == (long)doreti_iret) {
frame->tf_rip = (long)doreti_iret_fault;
goto out;
}
if (frame->tf_rip == (long)ld_ds) {
frame->tf_rip = (long)ds_load_fault;
goto out;
}
if (frame->tf_rip == (long)ld_es) {
frame->tf_rip = (long)es_load_fault;
goto out;
}
if (frame->tf_rip == (long)ld_fs) {
frame->tf_rip = (long)fs_load_fault;
goto out;
}
if (frame->tf_rip == (long)ld_gs) {
frame->tf_rip = (long)gs_load_fault;
goto out;
}
if (frame->tf_rip == (long)ld_gsbase) {
frame->tf_rip = (long)gsbase_load_fault;
goto out;
}
if (frame->tf_rip == (long)ld_fsbase) {
frame->tf_rip = (long)fsbase_load_fault;
goto out;
}
if (curpcb->pcb_onfault != NULL) {
frame->tf_rip = (long)curpcb->pcb_onfault;
goto out;
}
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_rflags & PSL_NT) {
frame->tf_rflags &= ~PSL_NT;
goto out;
}
break;
case T_TRCTRAP: /* trace trap */
/*
* 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
*/
/* XXX check upper bits here */
load_dr6(rdr6() & 0xfffffff0);
goto out;
}
/*
* FALLTHROUGH (TRCTRAP kernel mode, kernel address)
*/
case T_BPTFLT:
/*
* If KDB is enabled, let it handle the debugger trap.
* Otherwise, debugger traps "can't happen".
*/
#ifdef KDB
if (kdb_trap(type, 0, frame))
goto out;
#endif
break;
#ifdef DEV_ISA
case T_NMI:
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef KDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (kdb_on_nmi) {
printf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* KDB */
goto out;
} else if (panic_on_nmi == 0)
goto out;
/* FALLTHROUGH */
#endif /* DEV_ISA */
}
trap_fatal(frame, 0);
goto out;
}
/* Translate fault for emulators (e.g. Linux) */
if (*p->p_sysent->sv_transtrap)
i = (*p->p_sysent->sv_transtrap)(i, type);
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = i;
ksi.ksi_code = ucode;
ksi.ksi_trapno = type;
ksi.ksi_addr = (void *)addr;
if (uprintf_signal) {
uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
"addr 0x%lx rsp 0x%lx rip 0x%lx "
"<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
frame->tf_rsp, frame->tf_rip,
fubyte((void *)(frame->tf_rip + 0)),
fubyte((void *)(frame->tf_rip + 1)),
fubyte((void *)(frame->tf_rip + 2)),
fubyte((void *)(frame->tf_rip + 3)),
fubyte((void *)(frame->tf_rip + 4)),
fubyte((void *)(frame->tf_rip + 5)),
fubyte((void *)(frame->tf_rip + 6)),
fubyte((void *)(frame->tf_rip + 7)));
}
KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
trapsignal(td, &ksi);
user:
userret(td, frame);
KASSERT(PCB_USER_FPU(td->td_pcb),
("Return from trap with kernel FPU ctx leaked"));
userout:
out:
return;
}
static __inline int
get_esp(struct trapframe *tf)
{
return ((ISPL(tf->tf_cs)) ? tf->tf_esp :
(db_expr_t)tf + (uintptr_t)DB_OFFSET(tf_esp));
}
