Commit | Line | Data |
---|---|---|
62a038d3 P |
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. | |
6 | * | |
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. | |
11 | * | |
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. | |
15 | * | |
16 | * Copyright (C) 2007 Alan Stern | |
17 | * Copyright (C) IBM Corporation, 2009 | |
24f1e32c | 18 | * Copyright (C) 2009, Frederic Weisbecker <fweisbec@gmail.com> |
ba1c813a FW |
19 | * |
20 | * Thanks to Ingo Molnar for his many suggestions. | |
ba6909b7 P |
21 | * |
22 | * Authors: Alan Stern <stern@rowland.harvard.edu> | |
23 | * K.Prasad <prasad@linux.vnet.ibm.com> | |
24 | * Frederic Weisbecker <fweisbec@gmail.com> | |
62a038d3 P |
25 | */ |
26 | ||
27 | /* | |
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. | |
31 | */ | |
32 | ||
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> | |
feef47d0 | 43 | #include <linux/slab.h> |
45a73372 | 44 | #include <linux/list.h> |
88f7a890 | 45 | #include <linux/cpu.h> |
62a038d3 P |
46 | #include <linux/smp.h> |
47 | ||
24f1e32c | 48 | #include <linux/hw_breakpoint.h> |
ba1c813a FW |
49 | /* |
50 | * Constraints data | |
51 | */ | |
bde96030 ON |
52 | struct bp_cpuinfo { |
53 | /* Number of pinned cpu breakpoints in a cpu */ | |
54 | unsigned int cpu_pinned; | |
55 | /* tsk_pinned[n] is the number of tasks having n+1 breakpoints */ | |
56 | unsigned int *tsk_pinned; | |
57 | /* Number of non-pinned cpu/task breakpoints in a cpu */ | |
58 | unsigned int flexible; /* XXX: placeholder, see fetch_this_slot() */ | |
59 | }; | |
62a038d3 | 60 | |
bde96030 | 61 | static DEFINE_PER_CPU(struct bp_cpuinfo, bp_cpuinfo[TYPE_MAX]); |
feef47d0 FW |
62 | static int nr_slots[TYPE_MAX]; |
63 | ||
bde96030 ON |
64 | static struct bp_cpuinfo *get_bp_info(int cpu, enum bp_type_idx type) |
65 | { | |
66 | return per_cpu_ptr(bp_cpuinfo + type, cpu); | |
67 | } | |
68 | ||
45a73372 FW |
69 | /* Keep track of the breakpoints attached to tasks */ |
70 | static LIST_HEAD(bp_task_head); | |
71 | ||
feef47d0 FW |
72 | static int constraints_initialized; |
73 | ||
ba1c813a FW |
74 | /* Gather the number of total pinned and un-pinned bp in a cpuset */ |
75 | struct bp_busy_slots { | |
76 | unsigned int pinned; | |
77 | unsigned int flexible; | |
78 | }; | |
79 | ||
80 | /* Serialize accesses to the above constraints */ | |
81 | static DEFINE_MUTEX(nr_bp_mutex); | |
82 | ||
f93a2054 FW |
83 | __weak int hw_breakpoint_weight(struct perf_event *bp) |
84 | { | |
85 | return 1; | |
86 | } | |
87 | ||
0102752e FW |
88 | static inline enum bp_type_idx find_slot_idx(struct perf_event *bp) |
89 | { | |
90 | if (bp->attr.bp_type & HW_BREAKPOINT_RW) | |
91 | return TYPE_DATA; | |
92 | ||
93 | return TYPE_INST; | |
94 | } | |
95 | ||
ba1c813a FW |
96 | /* |
97 | * Report the maximum number of pinned breakpoints a task | |
98 | * have in this cpu | |
99 | */ | |
0102752e | 100 | static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type) |
62a038d3 | 101 | { |
bde96030 | 102 | unsigned int *tsk_pinned = get_bp_info(cpu, type)->tsk_pinned; |
ba1c813a | 103 | int i; |
62a038d3 | 104 | |
feef47d0 | 105 | for (i = nr_slots[type] - 1; i >= 0; i--) { |
ba1c813a FW |
106 | if (tsk_pinned[i] > 0) |
107 | return i + 1; | |
62a038d3 P |
108 | } |
109 | ||
24f1e32c | 110 | return 0; |
62a038d3 P |
111 | } |
112 | ||
45a73372 FW |
113 | /* |
114 | * Count the number of breakpoints of the same type and same task. | |
115 | * The given event must be not on the list. | |
116 | */ | |
0d855354 | 117 | static int task_bp_pinned(int cpu, struct perf_event *bp, enum bp_type_idx type) |
56053170 | 118 | { |
50f16a8b | 119 | struct task_struct *tsk = bp->hw.target; |
45a73372 | 120 | struct perf_event *iter; |
56053170 FW |
121 | int count = 0; |
122 | ||
45a73372 | 123 | list_for_each_entry(iter, &bp_task_head, hw.bp_list) { |
50f16a8b | 124 | if (iter->hw.target == tsk && |
0d855354 | 125 | find_slot_idx(iter) == type && |
8b4d801b | 126 | (iter->cpu < 0 || cpu == iter->cpu)) |
45a73372 | 127 | count += hw_breakpoint_weight(iter); |
56053170 FW |
128 | } |
129 | ||
56053170 FW |
130 | return count; |
131 | } | |
132 | ||
1c10adbb ON |
133 | static const struct cpumask *cpumask_of_bp(struct perf_event *bp) |
134 | { | |
135 | if (bp->cpu >= 0) | |
136 | return cpumask_of(bp->cpu); | |
137 | return cpu_possible_mask; | |
138 | } | |
139 | ||
ba1c813a FW |
140 | /* |
141 | * Report the number of pinned/un-pinned breakpoints we have in | |
142 | * a given cpu (cpu > -1) or in all of them (cpu = -1). | |
143 | */ | |
56053170 | 144 | static void |
0102752e FW |
145 | fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, |
146 | enum bp_type_idx type) | |
ba1c813a | 147 | { |
1c10adbb ON |
148 | const struct cpumask *cpumask = cpumask_of_bp(bp); |
149 | int cpu; | |
56053170 | 150 | |
1c10adbb | 151 | for_each_cpu(cpu, cpumask) { |
bde96030 ON |
152 | struct bp_cpuinfo *info = get_bp_info(cpu, type); |
153 | int nr; | |
ba1c813a | 154 | |
bde96030 | 155 | nr = info->cpu_pinned; |
50f16a8b | 156 | if (!bp->hw.target) |
0102752e | 157 | nr += max_task_bp_pinned(cpu, type); |
56053170 | 158 | else |
0d855354 | 159 | nr += task_bp_pinned(cpu, bp, type); |
ba1c813a FW |
160 | |
161 | if (nr > slots->pinned) | |
162 | slots->pinned = nr; | |
163 | ||
bde96030 | 164 | nr = info->flexible; |
ba1c813a FW |
165 | if (nr > slots->flexible) |
166 | slots->flexible = nr; | |
167 | } | |
168 | } | |
169 | ||
f93a2054 FW |
170 | /* |
171 | * For now, continue to consider flexible as pinned, until we can | |
172 | * ensure no flexible event can ever be scheduled before a pinned event | |
173 | * in a same cpu. | |
174 | */ | |
175 | static void | |
176 | fetch_this_slot(struct bp_busy_slots *slots, int weight) | |
177 | { | |
178 | slots->pinned += weight; | |
179 | } | |
180 | ||
ba1c813a FW |
181 | /* |
182 | * Add a pinned breakpoint for the given task in our constraint table | |
183 | */ | |
7ab71f32 | 184 | static void toggle_bp_task_slot(struct perf_event *bp, int cpu, |
f93a2054 | 185 | enum bp_type_idx type, int weight) |
ba1c813a | 186 | { |
bde96030 | 187 | unsigned int *tsk_pinned = get_bp_info(cpu, type)->tsk_pinned; |
e1ebe862 ON |
188 | int old_idx, new_idx; |
189 | ||
190 | old_idx = task_bp_pinned(cpu, bp, type) - 1; | |
7ab71f32 | 191 | new_idx = old_idx + weight; |
e1ebe862 ON |
192 | |
193 | if (old_idx >= 0) | |
194 | tsk_pinned[old_idx]--; | |
195 | if (new_idx >= 0) | |
196 | tsk_pinned[new_idx]++; | |
ba1c813a FW |
197 | } |
198 | ||
199 | /* | |
200 | * Add/remove the given breakpoint in our constraint table | |
201 | */ | |
0102752e | 202 | static void |
f93a2054 FW |
203 | toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type, |
204 | int weight) | |
ba1c813a | 205 | { |
1c10adbb ON |
206 | const struct cpumask *cpumask = cpumask_of_bp(bp); |
207 | int cpu; | |
ba1c813a | 208 | |
7ab71f32 ON |
209 | if (!enable) |
210 | weight = -weight; | |
211 | ||
45a73372 | 212 | /* Pinned counter cpu profiling */ |
50f16a8b | 213 | if (!bp->hw.target) { |
bde96030 | 214 | get_bp_info(bp->cpu, type)->cpu_pinned += weight; |
45a73372 FW |
215 | return; |
216 | } | |
217 | ||
ba1c813a | 218 | /* Pinned counter task profiling */ |
1c10adbb | 219 | for_each_cpu(cpu, cpumask) |
7ab71f32 | 220 | toggle_bp_task_slot(bp, cpu, type, weight); |
ba1c813a | 221 | |
ba1c813a | 222 | if (enable) |
45a73372 | 223 | list_add_tail(&bp->hw.bp_list, &bp_task_head); |
e1ebe862 ON |
224 | else |
225 | list_del(&bp->hw.bp_list); | |
ba1c813a FW |
226 | } |
227 | ||
f7136c51 P |
228 | /* |
229 | * Function to perform processor-specific cleanup during unregistration | |
230 | */ | |
231 | __weak void arch_unregister_hw_breakpoint(struct perf_event *bp) | |
232 | { | |
233 | /* | |
234 | * A weak stub function here for those archs that don't define | |
235 | * it inside arch/.../kernel/hw_breakpoint.c | |
236 | */ | |
237 | } | |
238 | ||
ba1c813a FW |
239 | /* |
240 | * Contraints to check before allowing this new breakpoint counter: | |
241 | * | |
242 | * == Non-pinned counter == (Considered as pinned for now) | |
243 | * | |
244 | * - If attached to a single cpu, check: | |
245 | * | |
bde96030 ON |
246 | * (per_cpu(info->flexible, cpu) || (per_cpu(info->cpu_pinned, cpu) |
247 | * + max(per_cpu(info->tsk_pinned, cpu)))) < HBP_NUM | |
ba1c813a FW |
248 | * |
249 | * -> If there are already non-pinned counters in this cpu, it means | |
250 | * there is already a free slot for them. | |
251 | * Otherwise, we check that the maximum number of per task | |
252 | * breakpoints (for this cpu) plus the number of per cpu breakpoint | |
253 | * (for this cpu) doesn't cover every registers. | |
254 | * | |
255 | * - If attached to every cpus, check: | |
256 | * | |
bde96030 ON |
257 | * (per_cpu(info->flexible, *) || (max(per_cpu(info->cpu_pinned, *)) |
258 | * + max(per_cpu(info->tsk_pinned, *)))) < HBP_NUM | |
ba1c813a FW |
259 | * |
260 | * -> This is roughly the same, except we check the number of per cpu | |
261 | * bp for every cpu and we keep the max one. Same for the per tasks | |
262 | * breakpoints. | |
263 | * | |
264 | * | |
265 | * == Pinned counter == | |
266 | * | |
267 | * - If attached to a single cpu, check: | |
268 | * | |
bde96030 ON |
269 | * ((per_cpu(info->flexible, cpu) > 1) + per_cpu(info->cpu_pinned, cpu) |
270 | * + max(per_cpu(info->tsk_pinned, cpu))) < HBP_NUM | |
ba1c813a | 271 | * |
bde96030 | 272 | * -> Same checks as before. But now the info->flexible, if any, must keep |
ba1c813a FW |
273 | * one register at least (or they will never be fed). |
274 | * | |
275 | * - If attached to every cpus, check: | |
276 | * | |
bde96030 ON |
277 | * ((per_cpu(info->flexible, *) > 1) + max(per_cpu(info->cpu_pinned, *)) |
278 | * + max(per_cpu(info->tsk_pinned, *))) < HBP_NUM | |
ba1c813a | 279 | */ |
5352ae63 | 280 | static int __reserve_bp_slot(struct perf_event *bp) |
ba1c813a FW |
281 | { |
282 | struct bp_busy_slots slots = {0}; | |
0102752e | 283 | enum bp_type_idx type; |
f93a2054 | 284 | int weight; |
ba1c813a | 285 | |
feef47d0 FW |
286 | /* We couldn't initialize breakpoint constraints on boot */ |
287 | if (!constraints_initialized) | |
288 | return -ENOMEM; | |
289 | ||
0102752e FW |
290 | /* Basic checks */ |
291 | if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY || | |
292 | bp->attr.bp_type == HW_BREAKPOINT_INVALID) | |
293 | return -EINVAL; | |
294 | ||
295 | type = find_slot_idx(bp); | |
f93a2054 FW |
296 | weight = hw_breakpoint_weight(bp); |
297 | ||
0102752e | 298 | fetch_bp_busy_slots(&slots, bp, type); |
45a73372 FW |
299 | /* |
300 | * Simulate the addition of this breakpoint to the constraints | |
301 | * and see the result. | |
302 | */ | |
f93a2054 | 303 | fetch_this_slot(&slots, weight); |
ba1c813a FW |
304 | |
305 | /* Flexible counters need to keep at least one slot */ | |
feef47d0 | 306 | if (slots.pinned + (!!slots.flexible) > nr_slots[type]) |
5352ae63 | 307 | return -ENOSPC; |
ba1c813a | 308 | |
f93a2054 | 309 | toggle_bp_slot(bp, true, type, weight); |
ba1c813a | 310 | |
5352ae63 JW |
311 | return 0; |
312 | } | |
313 | ||
314 | int reserve_bp_slot(struct perf_event *bp) | |
315 | { | |
316 | int ret; | |
317 | ||
318 | mutex_lock(&nr_bp_mutex); | |
319 | ||
320 | ret = __reserve_bp_slot(bp); | |
321 | ||
ba1c813a FW |
322 | mutex_unlock(&nr_bp_mutex); |
323 | ||
324 | return ret; | |
325 | } | |
326 | ||
5352ae63 JW |
327 | static void __release_bp_slot(struct perf_event *bp) |
328 | { | |
0102752e | 329 | enum bp_type_idx type; |
f93a2054 | 330 | int weight; |
0102752e FW |
331 | |
332 | type = find_slot_idx(bp); | |
f93a2054 FW |
333 | weight = hw_breakpoint_weight(bp); |
334 | toggle_bp_slot(bp, false, type, weight); | |
5352ae63 JW |
335 | } |
336 | ||
24f1e32c | 337 | void release_bp_slot(struct perf_event *bp) |
62a038d3 | 338 | { |
ba1c813a FW |
339 | mutex_lock(&nr_bp_mutex); |
340 | ||
f7136c51 | 341 | arch_unregister_hw_breakpoint(bp); |
5352ae63 | 342 | __release_bp_slot(bp); |
ba1c813a FW |
343 | |
344 | mutex_unlock(&nr_bp_mutex); | |
62a038d3 P |
345 | } |
346 | ||
5352ae63 JW |
347 | /* |
348 | * Allow the kernel debugger to reserve breakpoint slots without | |
349 | * taking a lock using the dbg_* variant of for the reserve and | |
350 | * release breakpoint slots. | |
351 | */ | |
352 | int dbg_reserve_bp_slot(struct perf_event *bp) | |
353 | { | |
354 | if (mutex_is_locked(&nr_bp_mutex)) | |
355 | return -1; | |
356 | ||
357 | return __reserve_bp_slot(bp); | |
358 | } | |
359 | ||
360 | int dbg_release_bp_slot(struct perf_event *bp) | |
361 | { | |
362 | if (mutex_is_locked(&nr_bp_mutex)) | |
363 | return -1; | |
364 | ||
365 | __release_bp_slot(bp); | |
366 | ||
367 | return 0; | |
368 | } | |
ba1c813a | 369 | |
b2812d03 FW |
370 | static int validate_hw_breakpoint(struct perf_event *bp) |
371 | { | |
372 | int ret; | |
373 | ||
374 | ret = arch_validate_hwbkpt_settings(bp); | |
375 | if (ret) | |
376 | return ret; | |
377 | ||
378 | if (arch_check_bp_in_kernelspace(bp)) { | |
379 | if (bp->attr.exclude_kernel) | |
380 | return -EINVAL; | |
381 | /* | |
382 | * Don't let unprivileged users set a breakpoint in the trap | |
383 | * path to avoid trap recursion attacks. | |
384 | */ | |
385 | if (!capable(CAP_SYS_ADMIN)) | |
386 | return -EPERM; | |
387 | } | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
b326e956 | 392 | int register_perf_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 393 | { |
24f1e32c | 394 | int ret; |
62a038d3 | 395 | |
24f1e32c FW |
396 | ret = reserve_bp_slot(bp); |
397 | if (ret) | |
398 | return ret; | |
62a038d3 | 399 | |
b2812d03 | 400 | ret = validate_hw_breakpoint(bp); |
62a038d3 | 401 | |
b23ff0e9 MS |
402 | /* if arch_validate_hwbkpt_settings() fails then release bp slot */ |
403 | if (ret) | |
404 | release_bp_slot(bp); | |
405 | ||
24f1e32c FW |
406 | return ret; |
407 | } | |
62a038d3 | 408 | |
62a038d3 P |
409 | /** |
410 | * register_user_hw_breakpoint - register a hardware breakpoint for user space | |
5fa10b28 | 411 | * @attr: breakpoint attributes |
24f1e32c | 412 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 413 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 414 | */ |
24f1e32c | 415 | struct perf_event * |
5fa10b28 | 416 | register_user_hw_breakpoint(struct perf_event_attr *attr, |
b326e956 | 417 | perf_overflow_handler_t triggered, |
4dc0da86 | 418 | void *context, |
5fa10b28 | 419 | struct task_struct *tsk) |
62a038d3 | 420 | { |
4dc0da86 AK |
421 | return perf_event_create_kernel_counter(attr, -1, tsk, triggered, |
422 | context); | |
62a038d3 P |
423 | } |
424 | EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); | |
425 | ||
426 | /** | |
427 | * modify_user_hw_breakpoint - modify a user-space hardware breakpoint | |
24f1e32c | 428 | * @bp: the breakpoint structure to modify |
5fa10b28 | 429 | * @attr: new breakpoint attributes |
24f1e32c | 430 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 431 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 432 | */ |
44234adc | 433 | int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) |
62a038d3 | 434 | { |
44234adc | 435 | u64 old_addr = bp->attr.bp_addr; |
cd757645 | 436 | u64 old_len = bp->attr.bp_len; |
44234adc | 437 | int old_type = bp->attr.bp_type; |
44234adc FW |
438 | int err = 0; |
439 | ||
500ad2d8 P |
440 | /* |
441 | * modify_user_hw_breakpoint can be invoked with IRQs disabled and hence it | |
442 | * will not be possible to raise IPIs that invoke __perf_event_disable. | |
443 | * So call the function directly after making sure we are targeting the | |
444 | * current task. | |
445 | */ | |
446 | if (irqs_disabled() && bp->ctx && bp->ctx->task == current) | |
fae3fde6 | 447 | perf_event_disable_local(bp); |
500ad2d8 P |
448 | else |
449 | perf_event_disable(bp); | |
44234adc FW |
450 | |
451 | bp->attr.bp_addr = attr->bp_addr; | |
452 | bp->attr.bp_type = attr->bp_type; | |
453 | bp->attr.bp_len = attr->bp_len; | |
454 | ||
455 | if (attr->disabled) | |
456 | goto end; | |
62a038d3 | 457 | |
b2812d03 | 458 | err = validate_hw_breakpoint(bp); |
44234adc FW |
459 | if (!err) |
460 | perf_event_enable(bp); | |
461 | ||
462 | if (err) { | |
463 | bp->attr.bp_addr = old_addr; | |
464 | bp->attr.bp_type = old_type; | |
465 | bp->attr.bp_len = old_len; | |
466 | if (!bp->attr.disabled) | |
467 | perf_event_enable(bp); | |
468 | ||
469 | return err; | |
470 | } | |
471 | ||
472 | end: | |
473 | bp->attr.disabled = attr->disabled; | |
474 | ||
475 | return 0; | |
62a038d3 P |
476 | } |
477 | EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); | |
478 | ||
479 | /** | |
24f1e32c | 480 | * unregister_hw_breakpoint - unregister a user-space hardware breakpoint |
62a038d3 | 481 | * @bp: the breakpoint structure to unregister |
62a038d3 | 482 | */ |
24f1e32c | 483 | void unregister_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 484 | { |
24f1e32c FW |
485 | if (!bp) |
486 | return; | |
487 | perf_event_release_kernel(bp); | |
488 | } | |
489 | EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); | |
490 | ||
62a038d3 | 491 | /** |
24f1e32c | 492 | * register_wide_hw_breakpoint - register a wide breakpoint in the kernel |
dd1853c3 | 493 | * @attr: breakpoint attributes |
24f1e32c | 494 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 495 | * |
24f1e32c | 496 | * @return a set of per_cpu pointers to perf events |
62a038d3 | 497 | */ |
44ee6358 | 498 | struct perf_event * __percpu * |
dd1853c3 | 499 | register_wide_hw_breakpoint(struct perf_event_attr *attr, |
4dc0da86 AK |
500 | perf_overflow_handler_t triggered, |
501 | void *context) | |
62a038d3 | 502 | { |
e12cbc10 ON |
503 | struct perf_event * __percpu *cpu_events, *bp; |
504 | long err = 0; | |
24f1e32c FW |
505 | int cpu; |
506 | ||
507 | cpu_events = alloc_percpu(typeof(*cpu_events)); | |
508 | if (!cpu_events) | |
44ee6358 | 509 | return (void __percpu __force *)ERR_PTR(-ENOMEM); |
62a038d3 | 510 | |
88f7a890 LZ |
511 | get_online_cpus(); |
512 | for_each_online_cpu(cpu) { | |
4dc0da86 AK |
513 | bp = perf_event_create_kernel_counter(attr, cpu, NULL, |
514 | triggered, context); | |
605bfaee | 515 | if (IS_ERR(bp)) { |
24f1e32c | 516 | err = PTR_ERR(bp); |
e12cbc10 | 517 | break; |
24f1e32c | 518 | } |
24f1e32c | 519 | |
e12cbc10 | 520 | per_cpu(*cpu_events, cpu) = bp; |
24f1e32c | 521 | } |
88f7a890 LZ |
522 | put_online_cpus(); |
523 | ||
e12cbc10 ON |
524 | if (likely(!err)) |
525 | return cpu_events; | |
526 | ||
527 | unregister_wide_hw_breakpoint(cpu_events); | |
44ee6358 | 528 | return (void __percpu __force *)ERR_PTR(err); |
62a038d3 | 529 | } |
f60d24d2 | 530 | EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint); |
62a038d3 P |
531 | |
532 | /** | |
24f1e32c FW |
533 | * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel |
534 | * @cpu_events: the per cpu set of events to unregister | |
62a038d3 | 535 | */ |
44ee6358 | 536 | void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events) |
62a038d3 | 537 | { |
24f1e32c | 538 | int cpu; |
62a038d3 | 539 | |
e12cbc10 ON |
540 | for_each_possible_cpu(cpu) |
541 | unregister_hw_breakpoint(per_cpu(*cpu_events, cpu)); | |
542 | ||
24f1e32c | 543 | free_percpu(cpu_events); |
62a038d3 | 544 | } |
f60d24d2 | 545 | EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint); |
62a038d3 P |
546 | |
547 | static struct notifier_block hw_breakpoint_exceptions_nb = { | |
548 | .notifier_call = hw_breakpoint_exceptions_notify, | |
549 | /* we need to be notified first */ | |
550 | .priority = 0x7fffffff | |
551 | }; | |
552 | ||
b0a873eb PZ |
553 | static void bp_perf_event_destroy(struct perf_event *event) |
554 | { | |
555 | release_bp_slot(event); | |
556 | } | |
557 | ||
558 | static int hw_breakpoint_event_init(struct perf_event *bp) | |
559 | { | |
560 | int err; | |
561 | ||
562 | if (bp->attr.type != PERF_TYPE_BREAKPOINT) | |
563 | return -ENOENT; | |
564 | ||
2481c5fa SE |
565 | /* |
566 | * no branch sampling for breakpoint events | |
567 | */ | |
568 | if (has_branch_stack(bp)) | |
569 | return -EOPNOTSUPP; | |
570 | ||
b0a873eb PZ |
571 | err = register_perf_hw_breakpoint(bp); |
572 | if (err) | |
573 | return err; | |
574 | ||
575 | bp->destroy = bp_perf_event_destroy; | |
576 | ||
577 | return 0; | |
578 | } | |
579 | ||
a4eaf7f1 PZ |
580 | static int hw_breakpoint_add(struct perf_event *bp, int flags) |
581 | { | |
582 | if (!(flags & PERF_EF_START)) | |
583 | bp->hw.state = PERF_HES_STOPPED; | |
584 | ||
ab573844 JO |
585 | if (is_sampling_event(bp)) { |
586 | bp->hw.last_period = bp->hw.sample_period; | |
587 | perf_swevent_set_period(bp); | |
588 | } | |
589 | ||
a4eaf7f1 PZ |
590 | return arch_install_hw_breakpoint(bp); |
591 | } | |
592 | ||
593 | static void hw_breakpoint_del(struct perf_event *bp, int flags) | |
594 | { | |
595 | arch_uninstall_hw_breakpoint(bp); | |
596 | } | |
597 | ||
598 | static void hw_breakpoint_start(struct perf_event *bp, int flags) | |
599 | { | |
600 | bp->hw.state = 0; | |
601 | } | |
602 | ||
603 | static void hw_breakpoint_stop(struct perf_event *bp, int flags) | |
604 | { | |
605 | bp->hw.state = PERF_HES_STOPPED; | |
606 | } | |
607 | ||
b0a873eb | 608 | static struct pmu perf_breakpoint = { |
89a1e187 PZ |
609 | .task_ctx_nr = perf_sw_context, /* could eventually get its own */ |
610 | ||
b0a873eb | 611 | .event_init = hw_breakpoint_event_init, |
a4eaf7f1 PZ |
612 | .add = hw_breakpoint_add, |
613 | .del = hw_breakpoint_del, | |
614 | .start = hw_breakpoint_start, | |
615 | .stop = hw_breakpoint_stop, | |
b0a873eb PZ |
616 | .read = hw_breakpoint_pmu_read, |
617 | }; | |
618 | ||
3c502e7a | 619 | int __init init_hw_breakpoint(void) |
62a038d3 | 620 | { |
feef47d0 FW |
621 | int cpu, err_cpu; |
622 | int i; | |
623 | ||
624 | for (i = 0; i < TYPE_MAX; i++) | |
625 | nr_slots[i] = hw_breakpoint_slots(i); | |
626 | ||
627 | for_each_possible_cpu(cpu) { | |
628 | for (i = 0; i < TYPE_MAX; i++) { | |
bde96030 ON |
629 | struct bp_cpuinfo *info = get_bp_info(cpu, i); |
630 | ||
631 | info->tsk_pinned = kcalloc(nr_slots[i], sizeof(int), | |
632 | GFP_KERNEL); | |
633 | if (!info->tsk_pinned) | |
feef47d0 FW |
634 | goto err_alloc; |
635 | } | |
636 | } | |
637 | ||
638 | constraints_initialized = 1; | |
639 | ||
2e80a82a | 640 | perf_pmu_register(&perf_breakpoint, "breakpoint", PERF_TYPE_BREAKPOINT); |
b0a873eb | 641 | |
62a038d3 | 642 | return register_die_notifier(&hw_breakpoint_exceptions_nb); |
feef47d0 FW |
643 | |
644 | err_alloc: | |
645 | for_each_possible_cpu(err_cpu) { | |
feef47d0 | 646 | for (i = 0; i < TYPE_MAX; i++) |
bde96030 | 647 | kfree(get_bp_info(err_cpu, i)->tsk_pinned); |
30ce2f7e NK |
648 | if (err_cpu == cpu) |
649 | break; | |
feef47d0 FW |
650 | } |
651 | ||
652 | return -ENOMEM; | |
62a038d3 | 653 | } |
24f1e32c FW |
654 | |
655 |