2 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
3 * Copyright (C) 1995, 1996 TooLs GmbH.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by TooLs GmbH.
17 * 4. The name of TooLs GmbH may not be used to endorse or promote products
18 * derived from this software without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
29 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 * $NetBSD: trap.c,v 1.58 2002/03/04 04:07:35 dbj Exp $
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_kdtrace.h"
39 #include <sys/param.h>
44 #include <sys/mutex.h>
45 #include <sys/pioctl.h>
46 #include <sys/ptrace.h>
47 #include <sys/reboot.h>
48 #include <sys/syscall.h>
49 #include <sys/sysent.h>
50 #include <sys/systm.h>
52 #include <sys/signalvar.h>
53 #include <sys/vmmeter.h>
55 #include <security/audit/audit.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_param.h>
61 #include <vm/vm_kern.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_page.h>
65 #include <machine/_inttypes.h>
66 #include <machine/altivec.h>
67 #include <machine/cpu.h>
68 #include <machine/db_machdep.h>
69 #include <machine/fpu.h>
70 #include <machine/frame.h>
71 #include <machine/pcb.h>
72 #include <machine/pmap.h>
73 #include <machine/psl.h>
74 #include <machine/trap.h>
75 #include <machine/spr.h>
76 #include <machine/sr.h>
78 static void trap_fatal(struct trapframe *frame);
79 static void printtrap(u_int vector, struct trapframe *frame, int isfatal,
81 static int trap_pfault(struct trapframe *frame, int user);
82 static int fix_unaligned(struct thread *td, struct trapframe *frame);
83 static int handle_onfault(struct trapframe *frame);
84 static void syscall(struct trapframe *frame);
87 void handle_kernel_slb_spill(int, register_t, register_t);
88 static int handle_user_slb_spill(pmap_t pm, vm_offset_t addr);
92 struct powerpc_exception {
98 #include <sys/dtrace_bsd.h>
100 int (*dtrace_invop_jump_addr)(struct trapframe *);
103 static struct powerpc_exception powerpc_exceptions[] = {
104 { 0x0100, "system reset" },
105 { 0x0200, "machine check" },
106 { 0x0300, "data storage interrupt" },
107 { 0x0380, "data segment exception" },
108 { 0x0400, "instruction storage interrupt" },
109 { 0x0480, "instruction segment exception" },
110 { 0x0500, "external interrupt" },
111 { 0x0600, "alignment" },
112 { 0x0700, "program" },
113 { 0x0800, "floating-point unavailable" },
114 { 0x0900, "decrementer" },
115 { 0x0c00, "system call" },
117 { 0x0e00, "floating-point assist" },
118 { 0x0f00, "performance monitoring" },
119 { 0x0f20, "altivec unavailable" },
120 { 0x1000, "instruction tlb miss" },
121 { 0x1100, "data load tlb miss" },
122 { 0x1200, "data store tlb miss" },
123 { 0x1300, "instruction breakpoint" },
124 { 0x1400, "system management" },
125 { 0x1600, "altivec assist" },
126 { 0x1700, "thermal management" },
127 { 0x2000, "run mode/trace" },
132 trapname(u_int vector)
134 struct powerpc_exception *pe;
136 for (pe = powerpc_exceptions; pe->vector != 0x3000; pe++) {
137 if (pe->vector == vector)
145 trap(struct trapframe *frame)
156 PCPU_INC(cnt.v_trap);
161 type = ucode = frame->exc;
163 user = frame->srr1 & PSL_PR;
165 CTR3(KTR_TRAP, "trap: %s type=%s (%s)", td->td_name,
166 trapname(type), user ? "user" : "kernel");
170 * A trap can occur while DTrace executes a probe. Before
171 * executing the probe, DTrace blocks re-scheduling and sets
172 * a flag in it's per-cpu flags to indicate that it doesn't
173 * want to fault. On returning from the probe, the no-fault
174 * flag is cleared and finally re-scheduling is enabled.
176 * If the DTrace kernel module has registered a trap handler,
177 * call it and if it returns non-zero, assume that it has
178 * handled the trap and modified the trap frame so that this
179 * function can return normally.
182 * XXXDTRACE: add pid probe handler here (if ever)
184 if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
190 td->td_frame = frame;
191 if (td->td_ucred != p->p_ucred)
192 cred_update_thread(td);
194 /* User Mode Traps */
198 frame->srr1 &= ~PSL_SE;
206 if (handle_user_slb_spill(&p->p_vmspace->vm_pmap,
207 (type == EXC_ISE) ? frame->srr0 :
208 frame->cpu.aim.dar) != 0) {
216 sig = trap_pfault(frame, 1);
226 KASSERT((td->td_pcb->pcb_flags & PCB_FPU) != PCB_FPU,
227 ("FPU already enabled for thread"));
232 KASSERT((td->td_pcb->pcb_flags & PCB_VEC) != PCB_VEC,
233 ("Altivec already enabled for thread"));
240 * We get a VPU assist exception for IEEE mode
241 * vector operations on denormalized floats.
242 * Emulating this is a giant pain, so for now,
243 * just switch off IEEE mode and treat them as
248 td->td_pcb->pcb_vec.vscr |= ALTIVEC_VSCR_NJ;
253 if (fix_unaligned(td, frame) != 0) {
262 /* Identify the trap reason */
263 if (frame->srr1 & EXC_PGM_TRAP) {
265 inst = fuword32((const void *)frame->srr0);
266 if (inst == 0x0FFFDDDD && dtrace_pid_probe_ptr != NULL) {
268 fill_regs(td, ®s);
269 (*dtrace_pid_probe_ptr)(®s);
276 sig = ppc_instr_emulate(frame, td->td_pcb);
278 if (frame->srr1 & EXC_PGM_PRIV)
280 else if (frame->srr1 & EXC_PGM_ILLEGAL)
282 } else if (sig == SIGFPE)
283 ucode = FPE_FLTINV; /* Punt for now, invalid operation. */
289 * Note that this may not be recoverable for the user
290 * process, depending on the type of machine check,
291 * but it at least prevents the kernel from dying.
301 /* Kernel Mode Traps */
303 KASSERT(cold || td->td_ucred != NULL,
304 ("kernel trap doesn't have ucred"));
308 if (frame->srr1 & EXC_PGM_TRAP) {
309 if (*(uint32_t *)frame->srr0 == 0x7c810808) {
310 if (dtrace_invop_jump_addr != NULL) {
311 dtrace_invop_jump_addr(frame);
320 if ((frame->cpu.aim.dar & SEGMENT_MASK) == USER_ADDR) {
321 __asm __volatile ("slbmte %0, %1" ::
322 "r"(td->td_pcb->pcb_cpu.aim.usr_vsid),
329 if (trap_pfault(frame, 0) == 0)
333 if (handle_onfault(frame))
343 if (p->p_sysent->sv_transtrap != NULL)
344 sig = (p->p_sysent->sv_transtrap)(sig, type);
345 ksiginfo_init_trap(&ksi);
347 ksi.ksi_code = (int) ucode; /* XXX, not POSIX */
348 /* ksi.ksi_addr = ? */
349 ksi.ksi_trapno = type;
350 trapsignal(td, &ksi);
357 trap_fatal(struct trapframe *frame)
360 printtrap(frame->exc, frame, 1, (frame->srr1 & PSL_PR));
362 if ((debugger_on_panic || kdb_active) &&
363 kdb_trap(frame->exc, 0, frame))
366 panic("%s trap", trapname(frame->exc));
370 printtrap(u_int vector, struct trapframe *frame, int isfatal, int user)
374 printf("%s %s trap:\n", isfatal ? "fatal" : "handled",
375 user ? "user" : "kernel");
377 printf(" exception = 0x%x (%s)\n", vector, trapname(vector));
381 printf(" virtual address = 0x%" PRIxPTR "\n",
383 printf(" dsisr = 0x%" PRIxPTR "\n",
384 frame->cpu.aim.dsisr);
388 printf(" virtual address = 0x%" PRIxPTR "\n", frame->srr0);
391 printf(" srr0 = 0x%" PRIxPTR "\n", frame->srr0);
392 printf(" srr1 = 0x%" PRIxPTR "\n", frame->srr1);
393 printf(" lr = 0x%" PRIxPTR "\n", frame->lr);
394 printf(" curthread = %p\n", curthread);
395 if (curthread != NULL)
396 printf(" pid = %d, comm = %s\n",
397 curthread->td_proc->p_pid, curthread->td_name);
402 * Handles a fatal fault when we have onfault state to recover. Returns
403 * non-zero if there was onfault recovery state available.
406 handle_onfault(struct trapframe *frame)
412 fb = td->td_pcb->pcb_onfault;
414 frame->srr0 = (*fb)[0];
415 frame->fixreg[1] = (*fb)[1];
416 frame->fixreg[2] = (*fb)[2];
417 frame->fixreg[3] = 1;
418 frame->cr = (*fb)[3];
419 bcopy(&(*fb)[4], &frame->fixreg[13],
420 19 * sizeof(register_t));
427 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
430 struct trapframe *frame;
436 frame = td->td_frame;
438 sa->code = frame->fixreg[0];
439 params = (caddr_t)(frame->fixreg + FIRSTARG);
442 if (sa->code == SYS_syscall) {
444 * code is first argument,
445 * followed by actual args.
447 sa->code = *(register_t *) params;
448 params += sizeof(register_t);
450 } else if (sa->code == SYS___syscall) {
452 * Like syscall, but code is a quad,
453 * so as to maintain quad alignment
454 * for the rest of the args.
456 if (SV_PROC_FLAG(p, SV_ILP32)) {
457 params += sizeof(register_t);
458 sa->code = *(register_t *) params;
459 params += sizeof(register_t);
462 sa->code = *(register_t *) params;
463 params += sizeof(register_t);
468 if (p->p_sysent->sv_mask)
469 sa->code &= p->p_sysent->sv_mask;
470 if (sa->code >= p->p_sysent->sv_size)
471 sa->callp = &p->p_sysent->sv_table[0];
473 sa->callp = &p->p_sysent->sv_table[sa->code];
475 sa->narg = sa->callp->sy_narg;
477 if (SV_PROC_FLAG(p, SV_ILP32)) {
478 argsz = sizeof(uint32_t);
480 for (i = 0; i < n; i++)
481 sa->args[i] = ((u_register_t *)(params))[i] &
484 argsz = sizeof(uint64_t);
486 for (i = 0; i < n; i++)
487 sa->args[i] = ((u_register_t *)(params))[i];
491 error = copyin(MOREARGS(frame->fixreg[1]), sa->args + n,
492 (sa->narg - n) * argsz);
497 if (SV_PROC_FLAG(p, SV_ILP32) && sa->narg > n) {
498 /* Expand the size of arguments copied from the stack */
500 for (i = sa->narg; i >= n; i--)
501 sa->args[i] = ((uint32_t *)(&sa->args[n]))[i-n];
506 td->td_retval[0] = 0;
507 td->td_retval[1] = frame->fixreg[FIRSTARG + 1];
512 #include "../../kern/subr_syscall.c"
515 syscall(struct trapframe *frame)
518 struct syscall_args sa;
522 td->td_frame = frame;
526 * Speculatively restore last user SLB segment, which we know is
527 * invalid already, since we are likely to do copyin()/copyout().
529 __asm __volatile ("slbmte %0, %1; isync" ::
530 "r"(td->td_pcb->pcb_cpu.aim.usr_vsid), "r"(USER_SLB_SLBE));
533 error = syscallenter(td, &sa);
534 syscallret(td, error, &sa);
538 /* Handle kernel SLB faults -- runs in real mode, all seat belts off */
540 handle_kernel_slb_spill(int type, register_t dar, register_t srr0)
542 struct slb *slbcache;
547 addr = (type == EXC_ISE) ? srr0 : dar;
548 slbcache = PCPU_GET(slb);
549 esid = (uintptr_t)addr >> ADDR_SR_SHFT;
550 slbe = (esid << SLBE_ESID_SHIFT) | SLBE_VALID;
552 /* See if the hardware flushed this somehow (can happen in LPARs) */
553 for (i = 0; i < n_slbs; i++)
554 if (slbcache[i].slbe == (slbe | (uint64_t)i))
557 /* Not in the map, needs to actually be added */
558 slbv = kernel_va_to_slbv(addr);
559 if (slbcache[USER_SLB_SLOT].slbe == 0) {
560 for (i = 0; i < n_slbs; i++) {
561 if (i == USER_SLB_SLOT)
563 if (!(slbcache[i].slbe & SLBE_VALID))
568 slbcache[USER_SLB_SLOT].slbe = 1;
571 /* Sacrifice a random SLB entry that is not the user entry */
573 if (i == USER_SLB_SLOT)
577 /* Write new entry */
578 slbcache[i].slbv = slbv;
579 slbcache[i].slbe = slbe | (uint64_t)i;
581 /* Trap handler will restore from cache on exit */
585 handle_user_slb_spill(pmap_t pm, vm_offset_t addr)
587 struct slb *user_entry;
591 esid = (uintptr_t)addr >> ADDR_SR_SHFT;
594 user_entry = user_va_to_slb_entry(pm, addr);
596 if (user_entry == NULL) {
597 /* allocate_vsid auto-spills it */
598 (void)allocate_user_vsid(pm, esid, 0);
601 * Check that another CPU has not already mapped this.
602 * XXX: Per-thread SLB caches would be better.
604 for (i = 0; i < pm->pm_slb_len; i++)
605 if (pm->pm_slb[i] == user_entry)
608 if (i == pm->pm_slb_len)
609 slb_insert_user(pm, user_entry);
618 trap_pfault(struct trapframe *frame, int user)
630 if (frame->exc == EXC_ISI) {
632 ftype = VM_PROT_EXECUTE;
633 if (frame->srr1 & SRR1_ISI_PFAULT)
634 ftype |= VM_PROT_READ;
636 eva = frame->cpu.aim.dar;
637 if (frame->cpu.aim.dsisr & DSISR_STORE)
638 ftype = VM_PROT_WRITE;
640 ftype = VM_PROT_READ;
644 map = &p->p_vmspace->vm_map;
646 if ((eva >> ADDR_SR_SHFT) == (USER_ADDR >> ADDR_SR_SHFT)) {
647 if (p->p_vmspace == NULL)
650 map = &p->p_vmspace->vm_map;
652 user_sr = td->td_pcb->pcb_cpu.aim.usr_segm;
653 eva &= ADDR_PIDX | ADDR_POFF;
654 eva |= user_sr << ADDR_SR_SHFT;
659 va = trunc_page(eva);
661 if (map != kernel_map) {
663 * Keep swapout from messing with us during this
670 /* Fault in the user page: */
671 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
677 * XXXDTRACE: add dtrace_doubletrap_func here?
681 * Don't have to worry about process locking or stacks in the
684 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
687 if (rv == KERN_SUCCESS)
690 if (!user && handle_onfault(frame))
697 * For now, this only deals with the particular unaligned access case
698 * that gcc tends to generate. Eventually it should handle all of the
699 * possibilities that can happen on a 32-bit PowerPC in big-endian mode.
703 fix_unaligned(struct thread *td, struct trapframe *frame)
705 struct thread *fputhread;
709 indicator = EXC_ALI_OPCODE_INDICATOR(frame->cpu.aim.dsisr);
714 reg = EXC_ALI_RST(frame->cpu.aim.dsisr);
715 fpr = &td->td_pcb->pcb_fpu.fpr[reg];
716 fputhread = PCPU_GET(fputhread);
718 /* Juggle the FPU to ensure that we've initialized
719 * the FPRs, and that their current state is in
722 if (fputhread != td) {
729 if (indicator == EXC_ALI_LFD) {
730 if (copyin((void *)frame->cpu.aim.dar, fpr,
731 sizeof(double)) != 0)
735 if (copyout(fpr, (void *)frame->cpu.aim.dar,
736 sizeof(double)) != 0)