2 * Kernel support for the ptrace() and syscall tracing interfaces.
4 * Copyright (C) 1999-2005 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
7 * Derived from the x86 and Alpha versions.
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
11 #include <linux/slab.h>
13 #include <linux/errno.h>
14 #include <linux/ptrace.h>
15 #include <linux/smp_lock.h>
16 #include <linux/user.h>
17 #include <linux/security.h>
18 #include <linux/audit.h>
19 #include <linux/signal.h>
21 #include <asm/pgtable.h>
22 #include <asm/processor.h>
23 #include <asm/ptrace_offsets.h>
25 #include <asm/system.h>
26 #include <asm/uaccess.h>
27 #include <asm/unwind.h>
29 #include <asm/perfmon.h>
35 * Bits in the PSR that we allow ptrace() to change:
36 * be, up, ac, mfl, mfh (the user mask; five bits total)
37 * db (debug breakpoint fault; one bit)
38 * id (instruction debug fault disable; one bit)
39 * dd (data debug fault disable; one bit)
40 * ri (restart instruction; two bits)
41 * is (instruction set; one bit)
43 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS \
44 | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
46 #define MASK(nbits) ((1UL << (nbits)) - 1) /* mask with NBITS bits set */
47 #define PFM_MASK MASK(38)
49 #define PTRACE_DEBUG 0
52 # define dprintk(format...) printk(format)
55 # define dprintk(format...)
58 /* Return TRUE if PT was created due to kernel-entry via a system-call. */
61 in_syscall (struct pt_regs *pt)
63 return (long) pt->cr_ifs >= 0;
67 * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
68 * bitset where bit i is set iff the NaT bit of register i is set.
71 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
73 # define GET_BITS(first, last, unat) \
75 unsigned long bit = ia64_unat_pos(&pt->r##first); \
76 unsigned long nbits = (last - first + 1); \
77 unsigned long mask = MASK(nbits) << first; \
80 dist = 64 + bit - first; \
83 ia64_rotr(unat, dist) & mask; \
88 * Registers that are stored consecutively in struct pt_regs
89 * can be handled in parallel. If the register order in
90 * struct_pt_regs changes, this code MUST be updated.
92 val = GET_BITS( 1, 1, scratch_unat);
93 val |= GET_BITS( 2, 3, scratch_unat);
94 val |= GET_BITS(12, 13, scratch_unat);
95 val |= GET_BITS(14, 14, scratch_unat);
96 val |= GET_BITS(15, 15, scratch_unat);
97 val |= GET_BITS( 8, 11, scratch_unat);
98 val |= GET_BITS(16, 31, scratch_unat);
105 * Set the NaT bits for the scratch registers according to NAT and
106 * return the resulting unat (assuming the scratch registers are
110 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
112 # define PUT_BITS(first, last, nat) \
114 unsigned long bit = ia64_unat_pos(&pt->r##first); \
115 unsigned long nbits = (last - first + 1); \
116 unsigned long mask = MASK(nbits) << first; \
119 dist = 64 + bit - first; \
121 dist = bit - first; \
122 ia64_rotl(nat & mask, dist); \
124 unsigned long scratch_unat;
127 * Registers that are stored consecutively in struct pt_regs
128 * can be handled in parallel. If the register order in
129 * struct_pt_regs changes, this code MUST be updated.
131 scratch_unat = PUT_BITS( 1, 1, nat);
132 scratch_unat |= PUT_BITS( 2, 3, nat);
133 scratch_unat |= PUT_BITS(12, 13, nat);
134 scratch_unat |= PUT_BITS(14, 14, nat);
135 scratch_unat |= PUT_BITS(15, 15, nat);
136 scratch_unat |= PUT_BITS( 8, 11, nat);
137 scratch_unat |= PUT_BITS(16, 31, nat);
144 #define IA64_MLX_TEMPLATE 0x2
145 #define IA64_MOVL_OPCODE 6
148 ia64_increment_ip (struct pt_regs *regs)
150 unsigned long w0, ri = ia64_psr(regs)->ri + 1;
155 } else if (ri == 2) {
156 get_user(w0, (char __user *) regs->cr_iip + 0);
157 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
159 * rfi'ing to slot 2 of an MLX bundle causes
160 * an illegal operation fault. We don't want
167 ia64_psr(regs)->ri = ri;
171 ia64_decrement_ip (struct pt_regs *regs)
173 unsigned long w0, ri = ia64_psr(regs)->ri - 1;
175 if (ia64_psr(regs)->ri == 0) {
178 get_user(w0, (char __user *) regs->cr_iip + 0);
179 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
181 * rfi'ing to slot 2 of an MLX bundle causes
182 * an illegal operation fault. We don't want
188 ia64_psr(regs)->ri = ri;
192 * This routine is used to read an rnat bits that are stored on the
193 * kernel backing store. Since, in general, the alignment of the user
194 * and kernel are different, this is not completely trivial. In
195 * essence, we need to construct the user RNAT based on up to two
196 * kernel RNAT values and/or the RNAT value saved in the child's
201 * +--------+ <-- lowest address
208 * | slot01 | > child_regs->ar_rnat
210 * | slot02 | / kernel rbs
211 * +--------+ +--------+
212 * <- child_regs->ar_bspstore | slot61 | <-- krbs
213 * +- - - - + +--------+
215 * +- - - - + +--------+
217 * +- - - - + +--------+
219 * +- - - - + +--------+
224 * | slot01 | > child_stack->ar_rnat
228 * <--- child_stack->ar_bspstore
230 * The way to think of this code is as follows: bit 0 in the user rnat
231 * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
232 * value. The kernel rnat value holding this bit is stored in
233 * variable rnat0. rnat1 is loaded with the kernel rnat value that
234 * form the upper bits of the user rnat value.
238 * o when reading the rnat "below" the first rnat slot on the kernel
239 * backing store, rnat0/rnat1 are set to 0 and the low order bits are
240 * merged in from pt->ar_rnat.
242 * o when reading the rnat "above" the last rnat slot on the kernel
243 * backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
246 get_rnat (struct task_struct *task, struct switch_stack *sw,
247 unsigned long *krbs, unsigned long *urnat_addr,
248 unsigned long *urbs_end)
250 unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
251 unsigned long umask = 0, mask, m;
252 unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
253 long num_regs, nbits;
256 pt = task_pt_regs(task);
257 kbsp = (unsigned long *) sw->ar_bspstore;
258 ubspstore = (unsigned long *) pt->ar_bspstore;
260 if (urbs_end < urnat_addr)
261 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
266 * First, figure out which bit number slot 0 in user-land maps
267 * to in the kernel rnat. Do this by figuring out how many
268 * register slots we're beyond the user's backingstore and
269 * then computing the equivalent address in kernel space.
271 num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
272 slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
273 shift = ia64_rse_slot_num(slot0_kaddr);
274 rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
275 rnat0_kaddr = rnat1_kaddr - 64;
277 if (ubspstore + 63 > urnat_addr) {
278 /* some bits need to be merged in from pt->ar_rnat */
279 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
280 urnat = (pt->ar_rnat & umask);
287 if (rnat0_kaddr >= kbsp)
289 else if (rnat0_kaddr > krbs)
290 rnat0 = *rnat0_kaddr;
291 urnat |= (rnat0 & m) >> shift;
293 m = mask >> (63 - shift);
294 if (rnat1_kaddr >= kbsp)
296 else if (rnat1_kaddr > krbs)
297 rnat1 = *rnat1_kaddr;
298 urnat |= (rnat1 & m) << (63 - shift);
303 * The reverse of get_rnat.
306 put_rnat (struct task_struct *task, struct switch_stack *sw,
307 unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
308 unsigned long *urbs_end)
310 unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
311 unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
312 long num_regs, nbits;
314 unsigned long cfm, *urbs_kargs;
316 pt = task_pt_regs(task);
317 kbsp = (unsigned long *) sw->ar_bspstore;
318 ubspstore = (unsigned long *) pt->ar_bspstore;
320 urbs_kargs = urbs_end;
321 if (in_syscall(pt)) {
323 * If entered via syscall, don't allow user to set rnat bits
327 urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
330 if (urbs_kargs >= urnat_addr)
333 if ((urnat_addr - 63) >= urbs_kargs)
335 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
340 * First, figure out which bit number slot 0 in user-land maps
341 * to in the kernel rnat. Do this by figuring out how many
342 * register slots we're beyond the user's backingstore and
343 * then computing the equivalent address in kernel space.
345 num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
346 slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
347 shift = ia64_rse_slot_num(slot0_kaddr);
348 rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
349 rnat0_kaddr = rnat1_kaddr - 64;
351 if (ubspstore + 63 > urnat_addr) {
352 /* some bits need to be place in pt->ar_rnat: */
353 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
354 pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
360 * Note: Section 11.1 of the EAS guarantees that bit 63 of an
361 * rnat slot is ignored. so we don't have to clear it here.
363 rnat0 = (urnat << shift);
365 if (rnat0_kaddr >= kbsp)
366 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
367 else if (rnat0_kaddr > krbs)
368 *rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
370 rnat1 = (urnat >> (63 - shift));
371 m = mask >> (63 - shift);
372 if (rnat1_kaddr >= kbsp)
373 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
374 else if (rnat1_kaddr > krbs)
375 *rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
379 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
380 unsigned long urbs_end)
382 unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
384 return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
388 * Read a word from the user-level backing store of task CHILD. ADDR
389 * is the user-level address to read the word from, VAL a pointer to
390 * the return value, and USER_BSP gives the end of the user-level
391 * backing store (i.e., it's the address that would be in ar.bsp after
392 * the user executed a "cover" instruction).
394 * This routine takes care of accessing the kernel register backing
395 * store for those registers that got spilled there. It also takes
396 * care of calculating the appropriate RNaT collection words.
399 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
400 unsigned long user_rbs_end, unsigned long addr, long *val)
402 unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
403 struct pt_regs *child_regs;
407 urbs_end = (long *) user_rbs_end;
408 laddr = (unsigned long *) addr;
409 child_regs = task_pt_regs(child);
410 bspstore = (unsigned long *) child_regs->ar_bspstore;
411 krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
412 if (on_kernel_rbs(addr, (unsigned long) bspstore,
413 (unsigned long) urbs_end))
416 * Attempt to read the RBS in an area that's actually
417 * on the kernel RBS => read the corresponding bits in
420 rnat_addr = ia64_rse_rnat_addr(laddr);
421 ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
423 if (laddr == rnat_addr) {
424 /* return NaT collection word itself */
429 if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
431 * It is implementation dependent whether the
432 * data portion of a NaT value gets saved on a
433 * st8.spill or RSE spill (e.g., see EAS 2.6,
434 * 4.4.4.6 Register Spill and Fill). To get
435 * consistent behavior across all possible
436 * IA-64 implementations, we return zero in
443 if (laddr < urbs_end) {
445 * The desired word is on the kernel RBS and
448 regnum = ia64_rse_num_regs(bspstore, laddr);
449 *val = *ia64_rse_skip_regs(krbs, regnum);
453 copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
454 if (copied != sizeof(ret))
461 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
462 unsigned long user_rbs_end, unsigned long addr, long val)
464 unsigned long *bspstore, *krbs, regnum, *laddr;
465 unsigned long *urbs_end = (long *) user_rbs_end;
466 struct pt_regs *child_regs;
468 laddr = (unsigned long *) addr;
469 child_regs = task_pt_regs(child);
470 bspstore = (unsigned long *) child_regs->ar_bspstore;
471 krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
472 if (on_kernel_rbs(addr, (unsigned long) bspstore,
473 (unsigned long) urbs_end))
476 * Attempt to write the RBS in an area that's actually
477 * on the kernel RBS => write the corresponding bits
480 if (ia64_rse_is_rnat_slot(laddr))
481 put_rnat(child, child_stack, krbs, laddr, val,
484 if (laddr < urbs_end) {
485 regnum = ia64_rse_num_regs(bspstore, laddr);
486 *ia64_rse_skip_regs(krbs, regnum) = val;
489 } else if (access_process_vm(child, addr, &val, sizeof(val), 1)
496 * Calculate the address of the end of the user-level register backing
497 * store. This is the address that would have been stored in ar.bsp
498 * if the user had executed a "cover" instruction right before
499 * entering the kernel. If CFMP is not NULL, it is used to return the
500 * "current frame mask" that was active at the time the kernel was
504 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
507 unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
510 krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
511 bspstore = (unsigned long *) pt->ar_bspstore;
512 ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
515 ndirty += (cfm & 0x7f);
517 cfm &= ~(1UL << 63); /* clear valid bit */
521 return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
525 * Synchronize (i.e, write) the RSE backing store living in kernel
526 * space to the VM of the CHILD task. SW and PT are the pointers to
527 * the switch_stack and pt_regs structures, respectively.
528 * USER_RBS_END is the user-level address at which the backing store
532 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
533 unsigned long user_rbs_start, unsigned long user_rbs_end)
535 unsigned long addr, val;
538 /* now copy word for word from kernel rbs to user rbs: */
539 for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
540 ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
543 if (access_process_vm(child, addr, &val, sizeof(val), 1)
551 ia64_sync_kernel_rbs (struct task_struct *child, struct switch_stack *sw,
552 unsigned long user_rbs_start, unsigned long user_rbs_end)
554 unsigned long addr, val;
557 /* now copy word for word from user rbs to kernel rbs: */
558 for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
559 if (access_process_vm(child, addr, &val, sizeof(val), 0)
563 ret = ia64_poke(child, sw, user_rbs_end, addr, val);
570 typedef long (*syncfunc_t)(struct task_struct *, struct switch_stack *,
571 unsigned long, unsigned long);
573 static void do_sync_rbs(struct unw_frame_info *info, void *arg)
576 unsigned long urbs_end;
579 if (unw_unwind_to_user(info) < 0)
581 pt = task_pt_regs(info->task);
582 urbs_end = ia64_get_user_rbs_end(info->task, pt, NULL);
584 fn(info->task, info->sw, pt->ar_bspstore, urbs_end);
588 * when a thread is stopped (ptraced), debugger might change thread's user
589 * stack (change memory directly), and we must avoid the RSE stored in kernel
590 * to override user stack (user space's RSE is newer than kernel's in the
591 * case). To workaround the issue, we copy kernel RSE to user RSE before the
592 * task is stopped, so user RSE has updated data. we then copy user RSE to
593 * kernel after the task is resummed from traced stop and kernel will use the
594 * newer RSE to return to user. TIF_RESTORE_RSE is the flag to indicate we need
595 * synchronize user RSE to kernel.
597 void ia64_ptrace_stop(void)
599 if (test_and_set_tsk_thread_flag(current, TIF_RESTORE_RSE))
601 tsk_set_notify_resume(current);
602 unw_init_running(do_sync_rbs, ia64_sync_user_rbs);
606 * This is called to read back the register backing store.
608 void ia64_sync_krbs(void)
610 clear_tsk_thread_flag(current, TIF_RESTORE_RSE);
611 tsk_clear_notify_resume(current);
613 unw_init_running(do_sync_rbs, ia64_sync_kernel_rbs);
617 * After PTRACE_ATTACH, a thread's register backing store area in user
618 * space is assumed to contain correct data whenever the thread is
619 * stopped. arch_ptrace_stop takes care of this on tracing stops.
620 * But if the child was already stopped for job control when we attach
621 * to it, then it might not ever get into ptrace_stop by the time we
622 * want to examine the user memory containing the RBS.
625 ptrace_attach_sync_user_rbs (struct task_struct *child)
628 struct unw_frame_info info;
631 * If the child is in TASK_STOPPED, we need to change that to
632 * TASK_TRACED momentarily while we operate on it. This ensures
633 * that the child won't be woken up and return to user mode while
634 * we are doing the sync. (It can only be woken up for SIGKILL.)
637 read_lock(&tasklist_lock);
639 spin_lock_irq(&child->sighand->siglock);
640 if (child->state == TASK_STOPPED &&
641 !test_and_set_tsk_thread_flag(child, TIF_RESTORE_RSE)) {
642 tsk_set_notify_resume(child);
644 child->state = TASK_TRACED;
647 spin_unlock_irq(&child->sighand->siglock);
649 read_unlock(&tasklist_lock);
654 unw_init_from_blocked_task(&info, child);
655 do_sync_rbs(&info, ia64_sync_user_rbs);
658 * Now move the child back into TASK_STOPPED if it should be in a
659 * job control stop, so that SIGCONT can be used to wake it up.
661 read_lock(&tasklist_lock);
663 spin_lock_irq(&child->sighand->siglock);
664 if (child->state == TASK_TRACED &&
665 (child->signal->flags & SIGNAL_STOP_STOPPED)) {
666 child->state = TASK_STOPPED;
668 spin_unlock_irq(&child->sighand->siglock);
670 read_unlock(&tasklist_lock);
674 thread_matches (struct task_struct *thread, unsigned long addr)
676 unsigned long thread_rbs_end;
677 struct pt_regs *thread_regs;
679 if (ptrace_check_attach(thread, 0) < 0)
681 * If the thread is not in an attachable state, we'll
682 * ignore it. The net effect is that if ADDR happens
683 * to overlap with the portion of the thread's
684 * register backing store that is currently residing
685 * on the thread's kernel stack, then ptrace() may end
686 * up accessing a stale value. But if the thread
687 * isn't stopped, that's a problem anyhow, so we're
688 * doing as well as we can...
692 thread_regs = task_pt_regs(thread);
693 thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
694 if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
697 return 1; /* looks like we've got a winner */
701 * Write f32-f127 back to task->thread.fph if it has been modified.
704 ia64_flush_fph (struct task_struct *task)
706 struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
709 * Prevent migrating this task while
710 * we're fiddling with the FPU state
713 if (ia64_is_local_fpu_owner(task) && psr->mfh) {
715 task->thread.flags |= IA64_THREAD_FPH_VALID;
716 ia64_save_fpu(&task->thread.fph[0]);
722 * Sync the fph state of the task so that it can be manipulated
723 * through thread.fph. If necessary, f32-f127 are written back to
724 * thread.fph or, if the fph state hasn't been used before, thread.fph
725 * is cleared to zeroes. Also, access to f32-f127 is disabled to
726 * ensure that the task picks up the state from thread.fph when it
730 ia64_sync_fph (struct task_struct *task)
732 struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
734 ia64_flush_fph(task);
735 if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
736 task->thread.flags |= IA64_THREAD_FPH_VALID;
737 memset(&task->thread.fph, 0, sizeof(task->thread.fph));
744 access_fr (struct unw_frame_info *info, int regnum, int hi,
745 unsigned long *data, int write_access)
747 struct ia64_fpreg fpval;
750 ret = unw_get_fr(info, regnum, &fpval);
755 fpval.u.bits[hi] = *data;
756 ret = unw_set_fr(info, regnum, fpval);
758 *data = fpval.u.bits[hi];
763 * Change the machine-state of CHILD such that it will return via the normal
764 * kernel exit-path, rather than the syscall-exit path.
767 convert_to_non_syscall (struct task_struct *child, struct pt_regs *pt,
770 struct unw_frame_info info, prev_info;
771 unsigned long ip, sp, pr;
773 unw_init_from_blocked_task(&info, child);
776 if (unw_unwind(&info) < 0)
779 unw_get_sp(&info, &sp);
780 if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
781 < IA64_PT_REGS_SIZE) {
782 dprintk("ptrace.%s: ran off the top of the kernel "
783 "stack\n", __func__);
786 if (unw_get_pr (&prev_info, &pr) < 0) {
787 unw_get_rp(&prev_info, &ip);
788 dprintk("ptrace.%s: failed to read "
789 "predicate register (ip=0x%lx)\n",
793 if (unw_is_intr_frame(&info)
794 && (pr & (1UL << PRED_USER_STACK)))
799 * Note: at the time of this call, the target task is blocked
800 * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
801 * (aka, "pLvSys") we redirect execution from
802 * .work_pending_syscall_end to .work_processed_kernel.
804 unw_get_pr(&prev_info, &pr);
805 pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
806 pr |= (1UL << PRED_NON_SYSCALL);
807 unw_set_pr(&prev_info, pr);
809 pt->cr_ifs = (1UL << 63) | cfm;
811 * Clear the memory that is NOT written on syscall-entry to
812 * ensure we do not leak kernel-state to user when execution
818 memset(&pt->r16, 0, 16*8); /* clear r16-r31 */
819 memset(&pt->f6, 0, 6*16); /* clear f6-f11 */
827 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
828 struct unw_frame_info *info,
829 unsigned long *data, int write_access)
831 unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
836 scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
837 if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
838 dprintk("ptrace: failed to set ar.unat\n");
841 for (regnum = 4; regnum <= 7; ++regnum) {
842 unw_get_gr(info, regnum, &dummy, &nat);
843 unw_set_gr(info, regnum, dummy,
844 (nat_bits >> regnum) & 1);
847 if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
848 dprintk("ptrace: failed to read ar.unat\n");
851 nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
852 for (regnum = 4; regnum <= 7; ++regnum) {
853 unw_get_gr(info, regnum, &dummy, &nat);
854 nat_bits |= (nat != 0) << regnum;
862 access_uarea (struct task_struct *child, unsigned long addr,
863 unsigned long *data, int write_access)
865 unsigned long *ptr, regnum, urbs_end, cfm;
866 struct switch_stack *sw;
868 # define pt_reg_addr(pt, reg) ((void *) \
869 ((unsigned long) (pt) \
870 + offsetof(struct pt_regs, reg)))
873 pt = task_pt_regs(child);
874 sw = (struct switch_stack *) (child->thread.ksp + 16);
876 if ((addr & 0x7) != 0) {
877 dprintk("ptrace: unaligned register address 0x%lx\n", addr);
881 if (addr < PT_F127 + 16) {
884 ia64_sync_fph(child);
886 ia64_flush_fph(child);
887 ptr = (unsigned long *)
888 ((unsigned long) &child->thread.fph + addr);
889 } else if ((addr >= PT_F10) && (addr < PT_F11 + 16)) {
890 /* scratch registers untouched by kernel (saved in pt_regs) */
891 ptr = pt_reg_addr(pt, f10) + (addr - PT_F10);
892 } else if (addr >= PT_F12 && addr < PT_F15 + 16) {
894 * Scratch registers untouched by kernel (saved in
897 ptr = (unsigned long *) ((long) sw
898 + (addr - PT_NAT_BITS - 32));
899 } else if (addr < PT_AR_LC + 8) {
900 /* preserved state: */
901 struct unw_frame_info info;
905 unw_init_from_blocked_task(&info, child);
906 if (unw_unwind_to_user(&info) < 0)
911 return access_nat_bits(child, pt, &info,
914 case PT_R4: case PT_R5: case PT_R6: case PT_R7:
916 /* read NaT bit first: */
919 ret = unw_get_gr(&info, (addr - PT_R4)/8 + 4,
924 return unw_access_gr(&info, (addr - PT_R4)/8 + 4, data,
927 case PT_B1: case PT_B2: case PT_B3:
928 case PT_B4: case PT_B5:
929 return unw_access_br(&info, (addr - PT_B1)/8 + 1, data,
933 return unw_access_ar(&info, UNW_AR_EC, data,
937 return unw_access_ar(&info, UNW_AR_LC, data,
941 if (addr >= PT_F2 && addr < PT_F5 + 16)
942 return access_fr(&info, (addr - PT_F2)/16 + 2,
943 (addr & 8) != 0, data,
945 else if (addr >= PT_F16 && addr < PT_F31 + 16)
946 return access_fr(&info,
947 (addr - PT_F16)/16 + 16,
951 dprintk("ptrace: rejecting access to register "
952 "address 0x%lx\n", addr);
956 } else if (addr < PT_F9+16) {
961 * By convention, we use PT_AR_BSP to refer to
962 * the end of the user-level backing store.
963 * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
964 * to get the real value of ar.bsp at the time
965 * the kernel was entered.
967 * Furthermore, when changing the contents of
968 * PT_AR_BSP (or PT_CFM) while the task is
969 * blocked in a system call, convert the state
970 * so that the non-system-call exit
971 * path is used. This ensures that the proper
972 * state will be picked up when resuming
973 * execution. However, it *also* means that
974 * once we write PT_AR_BSP/PT_CFM, it won't be
975 * possible to modify the syscall arguments of
976 * the pending system call any longer. This
977 * shouldn't be an issue because modifying
978 * PT_AR_BSP/PT_CFM generally implies that
979 * we're either abandoning the pending system
980 * call or that we defer it's re-execution
981 * (e.g., due to GDB doing an inferior
984 urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
986 if (*data != urbs_end) {
988 convert_to_non_syscall(child,
992 * Simulate user-level write
996 pt->ar_bspstore = *data;
1003 urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
1005 if (((cfm ^ *data) & PFM_MASK) != 0) {
1007 convert_to_non_syscall(child,
1010 pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
1011 | (*data & PFM_MASK));
1019 unsigned long tmp = *data;
1020 /* psr.ri==3 is a reserved value: SDM 2:25 */
1021 if ((tmp & IA64_PSR_RI) == IA64_PSR_RI)
1022 tmp &= ~IA64_PSR_RI;
1023 pt->cr_ipsr = ((tmp & IPSR_MASK)
1024 | (pt->cr_ipsr & ~IPSR_MASK));
1026 *data = (pt->cr_ipsr & IPSR_MASK);
1031 pt->ar_rsc = *data | (3 << 2); /* force PL3 */
1037 ptr = pt_reg_addr(pt, ar_rnat);
1040 ptr = pt_reg_addr(pt, r1);
1042 case PT_R2: case PT_R3:
1043 ptr = pt_reg_addr(pt, r2) + (addr - PT_R2);
1045 case PT_R8: case PT_R9: case PT_R10: case PT_R11:
1046 ptr = pt_reg_addr(pt, r8) + (addr - PT_R8);
1048 case PT_R12: case PT_R13:
1049 ptr = pt_reg_addr(pt, r12) + (addr - PT_R12);
1052 ptr = pt_reg_addr(pt, r14);
1055 ptr = pt_reg_addr(pt, r15);
1057 case PT_R16: case PT_R17: case PT_R18: case PT_R19:
1058 case PT_R20: case PT_R21: case PT_R22: case PT_R23:
1059 case PT_R24: case PT_R25: case PT_R26: case PT_R27:
1060 case PT_R28: case PT_R29: case PT_R30: case PT_R31:
1061 ptr = pt_reg_addr(pt, r16) + (addr - PT_R16);
1064 ptr = pt_reg_addr(pt, b0);
1067 ptr = pt_reg_addr(pt, b6);
1070 ptr = pt_reg_addr(pt, b7);
1072 case PT_F6: case PT_F6+8: case PT_F7: case PT_F7+8:
1073 case PT_F8: case PT_F8+8: case PT_F9: case PT_F9+8:
1074 ptr = pt_reg_addr(pt, f6) + (addr - PT_F6);
1076 case PT_AR_BSPSTORE:
1077 ptr = pt_reg_addr(pt, ar_bspstore);
1080 ptr = pt_reg_addr(pt, ar_unat);
1083 ptr = pt_reg_addr(pt, ar_pfs);
1086 ptr = pt_reg_addr(pt, ar_ccv);
1089 ptr = pt_reg_addr(pt, ar_fpsr);
1092 ptr = pt_reg_addr(pt, cr_iip);
1095 ptr = pt_reg_addr(pt, pr);
1097 /* scratch register */
1100 /* disallow accessing anything else... */
1101 dprintk("ptrace: rejecting access to register "
1102 "address 0x%lx\n", addr);
1105 } else if (addr <= PT_AR_SSD) {
1106 ptr = pt_reg_addr(pt, ar_csd) + (addr - PT_AR_CSD);
1108 /* access debug registers */
1110 if (addr >= PT_IBR) {
1111 regnum = (addr - PT_IBR) >> 3;
1112 ptr = &child->thread.ibr[0];
1114 regnum = (addr - PT_DBR) >> 3;
1115 ptr = &child->thread.dbr[0];
1119 dprintk("ptrace: rejecting access to register "
1120 "address 0x%lx\n", addr);
1123 #ifdef CONFIG_PERFMON
1125 * Check if debug registers are used by perfmon. This
1126 * test must be done once we know that we can do the
1127 * operation, i.e. the arguments are all valid, but
1128 * before we start modifying the state.
1130 * Perfmon needs to keep a count of how many processes
1131 * are trying to modify the debug registers for system
1132 * wide monitoring sessions.
1134 * We also include read access here, because they may
1135 * cause the PMU-installed debug register state
1136 * (dbr[], ibr[]) to be reset. The two arrays are also
1137 * used by perfmon, but we do not use
1138 * IA64_THREAD_DBG_VALID. The registers are restored
1139 * by the PMU context switch code.
1141 if (pfm_use_debug_registers(child)) return -1;
1144 if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
1145 child->thread.flags |= IA64_THREAD_DBG_VALID;
1146 memset(child->thread.dbr, 0,
1147 sizeof(child->thread.dbr));
1148 memset(child->thread.ibr, 0,
1149 sizeof(child->thread.ibr));
1154 if ((regnum & 1) && write_access) {
1155 /* don't let the user set kernel-level breakpoints: */
1156 *ptr = *data & ~(7UL << 56);
1168 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1170 unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
1171 struct unw_frame_info info;
1172 struct ia64_fpreg fpval;
1173 struct switch_stack *sw;
1175 long ret, retval = 0;
1179 if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
1182 pt = task_pt_regs(child);
1183 sw = (struct switch_stack *) (child->thread.ksp + 16);
1184 unw_init_from_blocked_task(&info, child);
1185 if (unw_unwind_to_user(&info) < 0) {
1189 if (((unsigned long) ppr & 0x7) != 0) {
1190 dprintk("ptrace:unaligned register address %p\n", ppr);
1194 if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
1195 || access_uarea(child, PT_AR_EC, &ec, 0) < 0
1196 || access_uarea(child, PT_AR_LC, &lc, 0) < 0
1197 || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
1198 || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
1199 || access_uarea(child, PT_CFM, &cfm, 0)
1200 || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
1205 retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
1206 retval |= __put_user(psr, &ppr->cr_ipsr);
1210 retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1211 retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
1212 retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1213 retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1214 retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1215 retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1217 retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
1218 retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
1219 retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1220 retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
1221 retval |= __put_user(cfm, &ppr->cfm);
1225 retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
1226 retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
1230 for (i = 4; i < 8; i++) {
1231 if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
1233 retval |= __put_user(val, &ppr->gr[i]);
1238 retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
1242 retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
1243 retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
1244 retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
1248 retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
1252 retval |= __put_user(pt->b0, &ppr->br[0]);
1256 for (i = 1; i < 6; i++) {
1257 if (unw_access_br(&info, i, &val, 0) < 0)
1259 __put_user(val, &ppr->br[i]);
1264 retval |= __put_user(pt->b6, &ppr->br[6]);
1265 retval |= __put_user(pt->b7, &ppr->br[7]);
1269 for (i = 2; i < 6; i++) {
1270 if (unw_get_fr(&info, i, &fpval) < 0)
1272 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1277 retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
1278 sizeof(struct ia64_fpreg) * 6);
1280 /* fp scratch regs(12-15) */
1282 retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
1283 sizeof(struct ia64_fpreg) * 4);
1287 for (i = 16; i < 32; i++) {
1288 if (unw_get_fr(&info, i, &fpval) < 0)
1290 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1295 ia64_flush_fph(child);
1296 retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
1297 sizeof(ppr->fr[32]) * 96);
1301 retval |= __put_user(pt->pr, &ppr->pr);
1305 retval |= __put_user(nat_bits, &ppr->nat);
1307 ret = retval ? -EIO : 0;
1312 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1314 unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
1315 struct unw_frame_info info;
1316 struct switch_stack *sw;
1317 struct ia64_fpreg fpval;
1319 long ret, retval = 0;
1322 memset(&fpval, 0, sizeof(fpval));
1324 if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
1327 pt = task_pt_regs(child);
1328 sw = (struct switch_stack *) (child->thread.ksp + 16);
1329 unw_init_from_blocked_task(&info, child);
1330 if (unw_unwind_to_user(&info) < 0) {
1334 if (((unsigned long) ppr & 0x7) != 0) {
1335 dprintk("ptrace:unaligned register address %p\n", ppr);
1341 retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
1342 retval |= __get_user(psr, &ppr->cr_ipsr);
1346 retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1347 retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
1348 retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1349 retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1350 retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1351 retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1353 retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1354 retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1355 retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1356 retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1357 retval |= __get_user(cfm, &ppr->cfm);
1361 retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1362 retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1366 for (i = 4; i < 8; i++) {
1367 retval |= __get_user(val, &ppr->gr[i]);
1368 /* NaT bit will be set via PT_NAT_BITS: */
1369 if (unw_set_gr(&info, i, val, 0) < 0)
1375 retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1379 retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1380 retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1381 retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1385 retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1389 retval |= __get_user(pt->b0, &ppr->br[0]);
1393 for (i = 1; i < 6; i++) {
1394 retval |= __get_user(val, &ppr->br[i]);
1395 unw_set_br(&info, i, val);
1400 retval |= __get_user(pt->b6, &ppr->br[6]);
1401 retval |= __get_user(pt->b7, &ppr->br[7]);
1405 for (i = 2; i < 6; i++) {
1406 retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1407 if (unw_set_fr(&info, i, fpval) < 0)
1413 retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1414 sizeof(ppr->fr[6]) * 6);
1416 /* fp scratch regs(12-15) */
1418 retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1419 sizeof(ppr->fr[12]) * 4);
1423 for (i = 16; i < 32; i++) {
1424 retval |= __copy_from_user(&fpval, &ppr->fr[i],
1426 if (unw_set_fr(&info, i, fpval) < 0)
1432 ia64_sync_fph(child);
1433 retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1434 sizeof(ppr->fr[32]) * 96);
1438 retval |= __get_user(pt->pr, &ppr->pr);
1442 retval |= __get_user(nat_bits, &ppr->nat);
1444 retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1445 retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
1446 retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1447 retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1448 retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1449 retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1450 retval |= access_uarea(child, PT_CFM, &cfm, 1);
1451 retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1453 ret = retval ? -EIO : 0;
1458 user_enable_single_step (struct task_struct *child)
1460 struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1462 set_tsk_thread_flag(child, TIF_SINGLESTEP);
1467 user_enable_block_step (struct task_struct *child)
1469 struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1471 set_tsk_thread_flag(child, TIF_SINGLESTEP);
1476 user_disable_single_step (struct task_struct *child)
1478 struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1480 /* make sure the single step/taken-branch trap bits are not set: */
1481 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
1487 * Called by kernel/ptrace.c when detaching..
1489 * Make sure the single step bit is not set.
1492 ptrace_disable (struct task_struct *child)
1494 user_disable_single_step(child);
1498 arch_ptrace (struct task_struct *child, long request, long addr, long data)
1501 case PTRACE_PEEKTEXT:
1502 case PTRACE_PEEKDATA:
1503 /* read word at location addr */
1504 if (access_process_vm(child, addr, &data, sizeof(data), 0)
1507 /* ensure return value is not mistaken for error code */
1508 force_successful_syscall_return();
1511 /* PTRACE_POKETEXT and PTRACE_POKEDATA is handled
1512 * by the generic ptrace_request().
1515 case PTRACE_PEEKUSR:
1516 /* read the word at addr in the USER area */
1517 if (access_uarea(child, addr, &data, 0) < 0)
1519 /* ensure return value is not mistaken for error code */
1520 force_successful_syscall_return();
1523 case PTRACE_POKEUSR:
1524 /* write the word at addr in the USER area */
1525 if (access_uarea(child, addr, &data, 1) < 0)
1529 case PTRACE_OLD_GETSIGINFO:
1530 /* for backwards-compatibility */
1531 return ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1533 case PTRACE_OLD_SETSIGINFO:
1534 /* for backwards-compatibility */
1535 return ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1537 case PTRACE_GETREGS:
1538 return ptrace_getregs(child,
1539 (struct pt_all_user_regs __user *) data);
1541 case PTRACE_SETREGS:
1542 return ptrace_setregs(child,
1543 (struct pt_all_user_regs __user *) data);
1546 return ptrace_request(child, request, addr, data);
1552 syscall_trace (void)
1555 * The 0x80 provides a way for the tracing parent to
1556 * distinguish between a syscall stop and SIGTRAP delivery.
1558 ptrace_notify(SIGTRAP
1559 | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
1562 * This isn't the same as continuing with a signal, but it
1563 * will do for normal use. strace only continues with a
1564 * signal if the stopping signal is not SIGTRAP. -brl
1566 if (current->exit_code) {
1567 send_sig(current->exit_code, current, 1);
1568 current->exit_code = 0;
1572 /* "asmlinkage" so the input arguments are preserved... */
1575 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1576 long arg4, long arg5, long arg6, long arg7,
1577 struct pt_regs regs)
1579 if (test_thread_flag(TIF_SYSCALL_TRACE)
1580 && (current->ptrace & PT_PTRACED))
1583 /* copy user rbs to kernel rbs */
1584 if (test_thread_flag(TIF_RESTORE_RSE))
1587 if (unlikely(current->audit_context)) {
1591 if (IS_IA32_PROCESS(®s)) {
1593 arch = AUDIT_ARCH_I386;
1596 arch = AUDIT_ARCH_IA64;
1599 audit_syscall_entry(arch, syscall, arg0, arg1, arg2, arg3);
1604 /* "asmlinkage" so the input arguments are preserved... */
1607 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1608 long arg4, long arg5, long arg6, long arg7,
1609 struct pt_regs regs)
1611 if (unlikely(current->audit_context)) {
1612 int success = AUDITSC_RESULT(regs.r10);
1613 long result = regs.r8;
1615 if (success != AUDITSC_SUCCESS)
1617 audit_syscall_exit(success, result);
1620 if ((test_thread_flag(TIF_SYSCALL_TRACE)
1621 || test_thread_flag(TIF_SINGLESTEP))
1622 && (current->ptrace & PT_PTRACED))
1625 /* copy user rbs to kernel rbs */
1626 if (test_thread_flag(TIF_RESTORE_RSE))