NFS: Update xdr_encode_foo() functions that we're keeping
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / hw_breakpoint.c
blobe5325825aeb6e1e4ea0514ee37cfa53412ec4e3c
1 /*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License as published by
4 * the Free Software Foundation; either version 2 of the License, or
5 * (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
16 * Copyright (C) 2007 Alan Stern
17 * Copyright (C) IBM Corporation, 2009
18 * Copyright (C) 2009, Frederic Weisbecker <fweisbec@gmail.com>
20 * Thanks to Ingo Molnar for his many suggestions.
22 * Authors: Alan Stern <stern@rowland.harvard.edu>
23 * K.Prasad <prasad@linux.vnet.ibm.com>
24 * Frederic Weisbecker <fweisbec@gmail.com>
28 * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility,
29 * using the CPU's debug registers.
30 * This file contains the arch-independent routines.
33 #include <linux/irqflags.h>
34 #include <linux/kallsyms.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/kernel.h>
39 #include <linux/module.h>
40 #include <linux/percpu.h>
41 #include <linux/sched.h>
42 #include <linux/init.h>
43 #include <linux/slab.h>
44 #include <linux/list.h>
45 #include <linux/cpu.h>
46 #include <linux/smp.h>
48 #include <linux/hw_breakpoint.h>
52 * Constraints data
55 /* Number of pinned cpu breakpoints in a cpu */
56 static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]);
58 /* Number of pinned task breakpoints in a cpu */
59 static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]);
61 /* Number of non-pinned cpu/task breakpoints in a cpu */
62 static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]);
64 static int nr_slots[TYPE_MAX];
66 /* Keep track of the breakpoints attached to tasks */
67 static LIST_HEAD(bp_task_head);
69 static int constraints_initialized;
71 /* Gather the number of total pinned and un-pinned bp in a cpuset */
72 struct bp_busy_slots {
73 unsigned int pinned;
74 unsigned int flexible;
77 /* Serialize accesses to the above constraints */
78 static DEFINE_MUTEX(nr_bp_mutex);
80 __weak int hw_breakpoint_weight(struct perf_event *bp)
82 return 1;
85 static inline enum bp_type_idx find_slot_idx(struct perf_event *bp)
87 if (bp->attr.bp_type & HW_BREAKPOINT_RW)
88 return TYPE_DATA;
90 return TYPE_INST;
94 * Report the maximum number of pinned breakpoints a task
95 * have in this cpu
97 static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type)
99 int i;
100 unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
102 for (i = nr_slots[type] - 1; i >= 0; i--) {
103 if (tsk_pinned[i] > 0)
104 return i + 1;
107 return 0;
111 * Count the number of breakpoints of the same type and same task.
112 * The given event must be not on the list.
114 static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type)
116 struct task_struct *tsk = bp->hw.bp_target;
117 struct perf_event *iter;
118 int count = 0;
120 list_for_each_entry(iter, &bp_task_head, hw.bp_list) {
121 if (iter->hw.bp_target == tsk && find_slot_idx(iter) == type)
122 count += hw_breakpoint_weight(iter);
125 return count;
129 * Report the number of pinned/un-pinned breakpoints we have in
130 * a given cpu (cpu > -1) or in all of them (cpu = -1).
132 static void
133 fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp,
134 enum bp_type_idx type)
136 int cpu = bp->cpu;
137 struct task_struct *tsk = bp->hw.bp_target;
139 if (cpu >= 0) {
140 slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu);
141 if (!tsk)
142 slots->pinned += max_task_bp_pinned(cpu, type);
143 else
144 slots->pinned += task_bp_pinned(bp, type);
145 slots->flexible = per_cpu(nr_bp_flexible[type], cpu);
147 return;
150 for_each_online_cpu(cpu) {
151 unsigned int nr;
153 nr = per_cpu(nr_cpu_bp_pinned[type], cpu);
154 if (!tsk)
155 nr += max_task_bp_pinned(cpu, type);
156 else
157 nr += task_bp_pinned(bp, type);
159 if (nr > slots->pinned)
160 slots->pinned = nr;
162 nr = per_cpu(nr_bp_flexible[type], cpu);
164 if (nr > slots->flexible)
165 slots->flexible = nr;
170 * For now, continue to consider flexible as pinned, until we can
171 * ensure no flexible event can ever be scheduled before a pinned event
172 * in a same cpu.
174 static void
175 fetch_this_slot(struct bp_busy_slots *slots, int weight)
177 slots->pinned += weight;
181 * Add a pinned breakpoint for the given task in our constraint table
183 static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable,
184 enum bp_type_idx type, int weight)
186 unsigned int *tsk_pinned;
187 int old_count = 0;
188 int old_idx = 0;
189 int idx = 0;
191 old_count = task_bp_pinned(bp, type);
192 old_idx = old_count - 1;
193 idx = old_idx + weight;
195 /* tsk_pinned[n] is the number of tasks having n breakpoints */
196 tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
197 if (enable) {
198 tsk_pinned[idx]++;
199 if (old_count > 0)
200 tsk_pinned[old_idx]--;
201 } else {
202 tsk_pinned[idx]--;
203 if (old_count > 0)
204 tsk_pinned[old_idx]++;
209 * Add/remove the given breakpoint in our constraint table
211 static void
212 toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type,
213 int weight)
215 int cpu = bp->cpu;
216 struct task_struct *tsk = bp->hw.bp_target;
218 /* Pinned counter cpu profiling */
219 if (!tsk) {
221 if (enable)
222 per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight;
223 else
224 per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight;
225 return;
228 /* Pinned counter task profiling */
230 if (!enable)
231 list_del(&bp->hw.bp_list);
233 if (cpu >= 0) {
234 toggle_bp_task_slot(bp, cpu, enable, type, weight);
235 } else {
236 for_each_online_cpu(cpu)
237 toggle_bp_task_slot(bp, cpu, enable, type, weight);
240 if (enable)
241 list_add_tail(&bp->hw.bp_list, &bp_task_head);
245 * Function to perform processor-specific cleanup during unregistration
247 __weak void arch_unregister_hw_breakpoint(struct perf_event *bp)
250 * A weak stub function here for those archs that don't define
251 * it inside arch/.../kernel/hw_breakpoint.c
256 * Contraints to check before allowing this new breakpoint counter:
258 * == Non-pinned counter == (Considered as pinned for now)
260 * - If attached to a single cpu, check:
262 * (per_cpu(nr_bp_flexible, cpu) || (per_cpu(nr_cpu_bp_pinned, cpu)
263 * + max(per_cpu(nr_task_bp_pinned, cpu)))) < HBP_NUM
265 * -> If there are already non-pinned counters in this cpu, it means
266 * there is already a free slot for them.
267 * Otherwise, we check that the maximum number of per task
268 * breakpoints (for this cpu) plus the number of per cpu breakpoint
269 * (for this cpu) doesn't cover every registers.
271 * - If attached to every cpus, check:
273 * (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *))
274 * + max(per_cpu(nr_task_bp_pinned, *)))) < HBP_NUM
276 * -> This is roughly the same, except we check the number of per cpu
277 * bp for every cpu and we keep the max one. Same for the per tasks
278 * breakpoints.
281 * == Pinned counter ==
283 * - If attached to a single cpu, check:
285 * ((per_cpu(nr_bp_flexible, cpu) > 1) + per_cpu(nr_cpu_bp_pinned, cpu)
286 * + max(per_cpu(nr_task_bp_pinned, cpu))) < HBP_NUM
288 * -> Same checks as before. But now the nr_bp_flexible, if any, must keep
289 * one register at least (or they will never be fed).
291 * - If attached to every cpus, check:
293 * ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *))
294 * + max(per_cpu(nr_task_bp_pinned, *))) < HBP_NUM
296 static int __reserve_bp_slot(struct perf_event *bp)
298 struct bp_busy_slots slots = {0};
299 enum bp_type_idx type;
300 int weight;
302 /* We couldn't initialize breakpoint constraints on boot */
303 if (!constraints_initialized)
304 return -ENOMEM;
306 /* Basic checks */
307 if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY ||
308 bp->attr.bp_type == HW_BREAKPOINT_INVALID)
309 return -EINVAL;
311 type = find_slot_idx(bp);
312 weight = hw_breakpoint_weight(bp);
314 fetch_bp_busy_slots(&slots, bp, type);
316 * Simulate the addition of this breakpoint to the constraints
317 * and see the result.
319 fetch_this_slot(&slots, weight);
321 /* Flexible counters need to keep at least one slot */
322 if (slots.pinned + (!!slots.flexible) > nr_slots[type])
323 return -ENOSPC;
325 toggle_bp_slot(bp, true, type, weight);
327 return 0;
330 int reserve_bp_slot(struct perf_event *bp)
332 int ret;
334 mutex_lock(&nr_bp_mutex);
336 ret = __reserve_bp_slot(bp);
338 mutex_unlock(&nr_bp_mutex);
340 return ret;
343 static void __release_bp_slot(struct perf_event *bp)
345 enum bp_type_idx type;
346 int weight;
348 type = find_slot_idx(bp);
349 weight = hw_breakpoint_weight(bp);
350 toggle_bp_slot(bp, false, type, weight);
353 void release_bp_slot(struct perf_event *bp)
355 mutex_lock(&nr_bp_mutex);
357 arch_unregister_hw_breakpoint(bp);
358 __release_bp_slot(bp);
360 mutex_unlock(&nr_bp_mutex);
364 * Allow the kernel debugger to reserve breakpoint slots without
365 * taking a lock using the dbg_* variant of for the reserve and
366 * release breakpoint slots.
368 int dbg_reserve_bp_slot(struct perf_event *bp)
370 if (mutex_is_locked(&nr_bp_mutex))
371 return -1;
373 return __reserve_bp_slot(bp);
376 int dbg_release_bp_slot(struct perf_event *bp)
378 if (mutex_is_locked(&nr_bp_mutex))
379 return -1;
381 __release_bp_slot(bp);
383 return 0;
386 static int validate_hw_breakpoint(struct perf_event *bp)
388 int ret;
390 ret = arch_validate_hwbkpt_settings(bp);
391 if (ret)
392 return ret;
394 if (arch_check_bp_in_kernelspace(bp)) {
395 if (bp->attr.exclude_kernel)
396 return -EINVAL;
398 * Don't let unprivileged users set a breakpoint in the trap
399 * path to avoid trap recursion attacks.
401 if (!capable(CAP_SYS_ADMIN))
402 return -EPERM;
405 return 0;
408 int register_perf_hw_breakpoint(struct perf_event *bp)
410 int ret;
412 ret = reserve_bp_slot(bp);
413 if (ret)
414 return ret;
416 ret = validate_hw_breakpoint(bp);
418 /* if arch_validate_hwbkpt_settings() fails then release bp slot */
419 if (ret)
420 release_bp_slot(bp);
422 return ret;
426 * register_user_hw_breakpoint - register a hardware breakpoint for user space
427 * @attr: breakpoint attributes
428 * @triggered: callback to trigger when we hit the breakpoint
429 * @tsk: pointer to 'task_struct' of the process to which the address belongs
431 struct perf_event *
432 register_user_hw_breakpoint(struct perf_event_attr *attr,
433 perf_overflow_handler_t triggered,
434 struct task_struct *tsk)
436 return perf_event_create_kernel_counter(attr, -1, tsk, triggered);
438 EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
441 * modify_user_hw_breakpoint - modify a user-space hardware breakpoint
442 * @bp: the breakpoint structure to modify
443 * @attr: new breakpoint attributes
444 * @triggered: callback to trigger when we hit the breakpoint
445 * @tsk: pointer to 'task_struct' of the process to which the address belongs
447 int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr)
449 u64 old_addr = bp->attr.bp_addr;
450 u64 old_len = bp->attr.bp_len;
451 int old_type = bp->attr.bp_type;
452 int err = 0;
454 perf_event_disable(bp);
456 bp->attr.bp_addr = attr->bp_addr;
457 bp->attr.bp_type = attr->bp_type;
458 bp->attr.bp_len = attr->bp_len;
460 if (attr->disabled)
461 goto end;
463 err = validate_hw_breakpoint(bp);
464 if (!err)
465 perf_event_enable(bp);
467 if (err) {
468 bp->attr.bp_addr = old_addr;
469 bp->attr.bp_type = old_type;
470 bp->attr.bp_len = old_len;
471 if (!bp->attr.disabled)
472 perf_event_enable(bp);
474 return err;
477 end:
478 bp->attr.disabled = attr->disabled;
480 return 0;
482 EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint);
485 * unregister_hw_breakpoint - unregister a user-space hardware breakpoint
486 * @bp: the breakpoint structure to unregister
488 void unregister_hw_breakpoint(struct perf_event *bp)
490 if (!bp)
491 return;
492 perf_event_release_kernel(bp);
494 EXPORT_SYMBOL_GPL(unregister_hw_breakpoint);
497 * register_wide_hw_breakpoint - register a wide breakpoint in the kernel
498 * @attr: breakpoint attributes
499 * @triggered: callback to trigger when we hit the breakpoint
501 * @return a set of per_cpu pointers to perf events
503 struct perf_event * __percpu *
504 register_wide_hw_breakpoint(struct perf_event_attr *attr,
505 perf_overflow_handler_t triggered)
507 struct perf_event * __percpu *cpu_events, **pevent, *bp;
508 long err;
509 int cpu;
511 cpu_events = alloc_percpu(typeof(*cpu_events));
512 if (!cpu_events)
513 return (void __percpu __force *)ERR_PTR(-ENOMEM);
515 get_online_cpus();
516 for_each_online_cpu(cpu) {
517 pevent = per_cpu_ptr(cpu_events, cpu);
518 bp = perf_event_create_kernel_counter(attr, cpu, NULL, triggered);
520 *pevent = bp;
522 if (IS_ERR(bp)) {
523 err = PTR_ERR(bp);
524 goto fail;
527 put_online_cpus();
529 return cpu_events;
531 fail:
532 for_each_online_cpu(cpu) {
533 pevent = per_cpu_ptr(cpu_events, cpu);
534 if (IS_ERR(*pevent))
535 break;
536 unregister_hw_breakpoint(*pevent);
538 put_online_cpus();
540 free_percpu(cpu_events);
541 return (void __percpu __force *)ERR_PTR(err);
543 EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
546 * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel
547 * @cpu_events: the per cpu set of events to unregister
549 void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)
551 int cpu;
552 struct perf_event **pevent;
554 for_each_possible_cpu(cpu) {
555 pevent = per_cpu_ptr(cpu_events, cpu);
556 unregister_hw_breakpoint(*pevent);
558 free_percpu(cpu_events);
560 EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint);
562 static struct notifier_block hw_breakpoint_exceptions_nb = {
563 .notifier_call = hw_breakpoint_exceptions_notify,
564 /* we need to be notified first */
565 .priority = 0x7fffffff
568 static void bp_perf_event_destroy(struct perf_event *event)
570 release_bp_slot(event);
573 static int hw_breakpoint_event_init(struct perf_event *bp)
575 int err;
577 if (bp->attr.type != PERF_TYPE_BREAKPOINT)
578 return -ENOENT;
580 err = register_perf_hw_breakpoint(bp);
581 if (err)
582 return err;
584 bp->destroy = bp_perf_event_destroy;
586 return 0;
589 static int hw_breakpoint_add(struct perf_event *bp, int flags)
591 if (!(flags & PERF_EF_START))
592 bp->hw.state = PERF_HES_STOPPED;
594 return arch_install_hw_breakpoint(bp);
597 static void hw_breakpoint_del(struct perf_event *bp, int flags)
599 arch_uninstall_hw_breakpoint(bp);
602 static void hw_breakpoint_start(struct perf_event *bp, int flags)
604 bp->hw.state = 0;
607 static void hw_breakpoint_stop(struct perf_event *bp, int flags)
609 bp->hw.state = PERF_HES_STOPPED;
612 static struct pmu perf_breakpoint = {
613 .task_ctx_nr = perf_sw_context, /* could eventually get its own */
615 .event_init = hw_breakpoint_event_init,
616 .add = hw_breakpoint_add,
617 .del = hw_breakpoint_del,
618 .start = hw_breakpoint_start,
619 .stop = hw_breakpoint_stop,
620 .read = hw_breakpoint_pmu_read,
623 int __init init_hw_breakpoint(void)
625 unsigned int **task_bp_pinned;
626 int cpu, err_cpu;
627 int i;
629 for (i = 0; i < TYPE_MAX; i++)
630 nr_slots[i] = hw_breakpoint_slots(i);
632 for_each_possible_cpu(cpu) {
633 for (i = 0; i < TYPE_MAX; i++) {
634 task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu);
635 *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i],
636 GFP_KERNEL);
637 if (!*task_bp_pinned)
638 goto err_alloc;
642 constraints_initialized = 1;
644 perf_pmu_register(&perf_breakpoint);
646 return register_die_notifier(&hw_breakpoint_exceptions_nb);
648 err_alloc:
649 for_each_possible_cpu(err_cpu) {
650 if (err_cpu == cpu)
651 break;
652 for (i = 0; i < TYPE_MAX; i++)
653 kfree(per_cpu(nr_task_bp_pinned[i], cpu));
656 return -ENOMEM;