]>
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 FW |
48 | #include <linux/hw_breakpoint.h> |
49 | ||
0102752e | 50 | |
ba1c813a FW |
51 | /* |
52 | * Constraints data | |
53 | */ | |
62a038d3 | 54 | |
ba1c813a | 55 | /* Number of pinned cpu breakpoints in a cpu */ |
0102752e | 56 | static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]); |
ba1c813a FW |
57 | |
58 | /* Number of pinned task breakpoints in a cpu */ | |
777d0411 | 59 | static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]); |
ba1c813a FW |
60 | |
61 | /* Number of non-pinned cpu/task breakpoints in a cpu */ | |
0102752e | 62 | static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]); |
ba1c813a | 63 | |
feef47d0 FW |
64 | static int nr_slots[TYPE_MAX]; |
65 | ||
45a73372 FW |
66 | /* Keep track of the breakpoints attached to tasks */ |
67 | static LIST_HEAD(bp_task_head); | |
68 | ||
feef47d0 FW |
69 | static int constraints_initialized; |
70 | ||
ba1c813a FW |
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; | |
75 | }; | |
76 | ||
77 | /* Serialize accesses to the above constraints */ | |
78 | static DEFINE_MUTEX(nr_bp_mutex); | |
79 | ||
f93a2054 FW |
80 | __weak int hw_breakpoint_weight(struct perf_event *bp) |
81 | { | |
82 | return 1; | |
83 | } | |
84 | ||
0102752e FW |
85 | static inline enum bp_type_idx find_slot_idx(struct perf_event *bp) |
86 | { | |
87 | if (bp->attr.bp_type & HW_BREAKPOINT_RW) | |
88 | return TYPE_DATA; | |
89 | ||
90 | return TYPE_INST; | |
91 | } | |
92 | ||
ba1c813a FW |
93 | /* |
94 | * Report the maximum number of pinned breakpoints a task | |
95 | * have in this cpu | |
96 | */ | |
0102752e | 97 | static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type) |
62a038d3 | 98 | { |
ba1c813a | 99 | int i; |
0102752e | 100 | unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); |
62a038d3 | 101 | |
feef47d0 | 102 | for (i = nr_slots[type] - 1; i >= 0; i--) { |
ba1c813a FW |
103 | if (tsk_pinned[i] > 0) |
104 | return i + 1; | |
62a038d3 P |
105 | } |
106 | ||
24f1e32c | 107 | return 0; |
62a038d3 P |
108 | } |
109 | ||
45a73372 FW |
110 | /* |
111 | * Count the number of breakpoints of the same type and same task. | |
112 | * The given event must be not on the list. | |
113 | */ | |
0d855354 | 114 | static int task_bp_pinned(int cpu, struct perf_event *bp, enum bp_type_idx type) |
56053170 | 115 | { |
d580ff86 | 116 | struct task_struct *tsk = bp->hw.bp_target; |
45a73372 | 117 | struct perf_event *iter; |
56053170 FW |
118 | int count = 0; |
119 | ||
45a73372 | 120 | list_for_each_entry(iter, &bp_task_head, hw.bp_list) { |
0d855354 MN |
121 | if (iter->hw.bp_target == tsk && |
122 | find_slot_idx(iter) == type && | |
8b4d801b | 123 | (iter->cpu < 0 || cpu == iter->cpu)) |
45a73372 | 124 | count += hw_breakpoint_weight(iter); |
56053170 FW |
125 | } |
126 | ||
56053170 FW |
127 | return count; |
128 | } | |
129 | ||
ba1c813a FW |
130 | /* |
131 | * Report the number of pinned/un-pinned breakpoints we have in | |
132 | * a given cpu (cpu > -1) or in all of them (cpu = -1). | |
133 | */ | |
56053170 | 134 | static void |
0102752e FW |
135 | fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, |
136 | enum bp_type_idx type) | |
ba1c813a | 137 | { |
56053170 | 138 | int cpu = bp->cpu; |
d580ff86 | 139 | struct task_struct *tsk = bp->hw.bp_target; |
56053170 | 140 | |
ba1c813a | 141 | if (cpu >= 0) { |
0102752e | 142 | slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu); |
56053170 | 143 | if (!tsk) |
0102752e | 144 | slots->pinned += max_task_bp_pinned(cpu, type); |
56053170 | 145 | else |
0d855354 | 146 | slots->pinned += task_bp_pinned(cpu, bp, type); |
0102752e | 147 | slots->flexible = per_cpu(nr_bp_flexible[type], cpu); |
ba1c813a FW |
148 | |
149 | return; | |
150 | } | |
151 | ||
c790b0ad | 152 | for_each_possible_cpu(cpu) { |
ba1c813a FW |
153 | unsigned int nr; |
154 | ||
0102752e | 155 | nr = per_cpu(nr_cpu_bp_pinned[type], cpu); |
56053170 | 156 | if (!tsk) |
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 | ||
0102752e | 164 | nr = per_cpu(nr_bp_flexible[type], cpu); |
ba1c813a FW |
165 | |
166 | if (nr > slots->flexible) | |
167 | slots->flexible = nr; | |
168 | } | |
169 | } | |
170 | ||
f93a2054 FW |
171 | /* |
172 | * For now, continue to consider flexible as pinned, until we can | |
173 | * ensure no flexible event can ever be scheduled before a pinned event | |
174 | * in a same cpu. | |
175 | */ | |
176 | static void | |
177 | fetch_this_slot(struct bp_busy_slots *slots, int weight) | |
178 | { | |
179 | slots->pinned += weight; | |
180 | } | |
181 | ||
ba1c813a FW |
182 | /* |
183 | * Add a pinned breakpoint for the given task in our constraint table | |
184 | */ | |
7ab71f32 | 185 | static void toggle_bp_task_slot(struct perf_event *bp, int cpu, |
f93a2054 | 186 | enum bp_type_idx type, int weight) |
ba1c813a | 187 | { |
e1ebe862 ON |
188 | /* tsk_pinned[n-1] is the number of tasks having n>0 breakpoints */ |
189 | unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); | |
190 | int old_idx, new_idx; | |
191 | ||
192 | old_idx = task_bp_pinned(cpu, bp, type) - 1; | |
7ab71f32 | 193 | new_idx = old_idx + weight; |
e1ebe862 ON |
194 | |
195 | if (old_idx >= 0) | |
196 | tsk_pinned[old_idx]--; | |
197 | if (new_idx >= 0) | |
198 | tsk_pinned[new_idx]++; | |
ba1c813a FW |
199 | } |
200 | ||
201 | /* | |
202 | * Add/remove the given breakpoint in our constraint table | |
203 | */ | |
0102752e | 204 | static void |
f93a2054 FW |
205 | toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type, |
206 | int weight) | |
ba1c813a FW |
207 | { |
208 | int cpu = bp->cpu; | |
d580ff86 | 209 | struct task_struct *tsk = bp->hw.bp_target; |
ba1c813a | 210 | |
7ab71f32 ON |
211 | if (!enable) |
212 | weight = -weight; | |
213 | ||
45a73372 FW |
214 | /* Pinned counter cpu profiling */ |
215 | if (!tsk) { | |
7ab71f32 | 216 | per_cpu(nr_cpu_bp_pinned[type], cpu) += weight; |
45a73372 FW |
217 | return; |
218 | } | |
219 | ||
ba1c813a | 220 | /* Pinned counter task profiling */ |
45a73372 | 221 | if (cpu >= 0) { |
7ab71f32 | 222 | toggle_bp_task_slot(bp, cpu, type, weight); |
45a73372 | 223 | } else { |
c790b0ad | 224 | for_each_possible_cpu(cpu) |
7ab71f32 | 225 | toggle_bp_task_slot(bp, cpu, type, weight); |
ba1c813a FW |
226 | } |
227 | ||
ba1c813a | 228 | if (enable) |
45a73372 | 229 | list_add_tail(&bp->hw.bp_list, &bp_task_head); |
e1ebe862 ON |
230 | else |
231 | list_del(&bp->hw.bp_list); | |
ba1c813a FW |
232 | } |
233 | ||
f7136c51 P |
234 | /* |
235 | * Function to perform processor-specific cleanup during unregistration | |
236 | */ | |
237 | __weak void arch_unregister_hw_breakpoint(struct perf_event *bp) | |
238 | { | |
239 | /* | |
240 | * A weak stub function here for those archs that don't define | |
241 | * it inside arch/.../kernel/hw_breakpoint.c | |
242 | */ | |
243 | } | |
244 | ||
ba1c813a FW |
245 | /* |
246 | * Contraints to check before allowing this new breakpoint counter: | |
247 | * | |
248 | * == Non-pinned counter == (Considered as pinned for now) | |
249 | * | |
250 | * - If attached to a single cpu, check: | |
251 | * | |
252 | * (per_cpu(nr_bp_flexible, cpu) || (per_cpu(nr_cpu_bp_pinned, cpu) | |
6ab88863 | 253 | * + max(per_cpu(nr_task_bp_pinned, cpu)))) < HBP_NUM |
ba1c813a FW |
254 | * |
255 | * -> If there are already non-pinned counters in this cpu, it means | |
256 | * there is already a free slot for them. | |
257 | * Otherwise, we check that the maximum number of per task | |
258 | * breakpoints (for this cpu) plus the number of per cpu breakpoint | |
259 | * (for this cpu) doesn't cover every registers. | |
260 | * | |
261 | * - If attached to every cpus, check: | |
262 | * | |
263 | * (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *)) | |
6ab88863 | 264 | * + max(per_cpu(nr_task_bp_pinned, *)))) < HBP_NUM |
ba1c813a FW |
265 | * |
266 | * -> This is roughly the same, except we check the number of per cpu | |
267 | * bp for every cpu and we keep the max one. Same for the per tasks | |
268 | * breakpoints. | |
269 | * | |
270 | * | |
271 | * == Pinned counter == | |
272 | * | |
273 | * - If attached to a single cpu, check: | |
274 | * | |
275 | * ((per_cpu(nr_bp_flexible, cpu) > 1) + per_cpu(nr_cpu_bp_pinned, cpu) | |
6ab88863 | 276 | * + max(per_cpu(nr_task_bp_pinned, cpu))) < HBP_NUM |
ba1c813a FW |
277 | * |
278 | * -> Same checks as before. But now the nr_bp_flexible, if any, must keep | |
279 | * one register at least (or they will never be fed). | |
280 | * | |
281 | * - If attached to every cpus, check: | |
282 | * | |
283 | * ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *)) | |
6ab88863 | 284 | * + max(per_cpu(nr_task_bp_pinned, *))) < HBP_NUM |
ba1c813a | 285 | */ |
5352ae63 | 286 | static int __reserve_bp_slot(struct perf_event *bp) |
ba1c813a FW |
287 | { |
288 | struct bp_busy_slots slots = {0}; | |
0102752e | 289 | enum bp_type_idx type; |
f93a2054 | 290 | int weight; |
ba1c813a | 291 | |
feef47d0 FW |
292 | /* We couldn't initialize breakpoint constraints on boot */ |
293 | if (!constraints_initialized) | |
294 | return -ENOMEM; | |
295 | ||
0102752e FW |
296 | /* Basic checks */ |
297 | if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY || | |
298 | bp->attr.bp_type == HW_BREAKPOINT_INVALID) | |
299 | return -EINVAL; | |
300 | ||
301 | type = find_slot_idx(bp); | |
f93a2054 FW |
302 | weight = hw_breakpoint_weight(bp); |
303 | ||
0102752e | 304 | fetch_bp_busy_slots(&slots, bp, type); |
45a73372 FW |
305 | /* |
306 | * Simulate the addition of this breakpoint to the constraints | |
307 | * and see the result. | |
308 | */ | |
f93a2054 | 309 | fetch_this_slot(&slots, weight); |
ba1c813a FW |
310 | |
311 | /* Flexible counters need to keep at least one slot */ | |
feef47d0 | 312 | if (slots.pinned + (!!slots.flexible) > nr_slots[type]) |
5352ae63 | 313 | return -ENOSPC; |
ba1c813a | 314 | |
f93a2054 | 315 | toggle_bp_slot(bp, true, type, weight); |
ba1c813a | 316 | |
5352ae63 JW |
317 | return 0; |
318 | } | |
319 | ||
320 | int reserve_bp_slot(struct perf_event *bp) | |
321 | { | |
322 | int ret; | |
323 | ||
324 | mutex_lock(&nr_bp_mutex); | |
325 | ||
326 | ret = __reserve_bp_slot(bp); | |
327 | ||
ba1c813a FW |
328 | mutex_unlock(&nr_bp_mutex); |
329 | ||
330 | return ret; | |
331 | } | |
332 | ||
5352ae63 JW |
333 | static void __release_bp_slot(struct perf_event *bp) |
334 | { | |
0102752e | 335 | enum bp_type_idx type; |
f93a2054 | 336 | int weight; |
0102752e FW |
337 | |
338 | type = find_slot_idx(bp); | |
f93a2054 FW |
339 | weight = hw_breakpoint_weight(bp); |
340 | toggle_bp_slot(bp, false, type, weight); | |
5352ae63 JW |
341 | } |
342 | ||
24f1e32c | 343 | void release_bp_slot(struct perf_event *bp) |
62a038d3 | 344 | { |
ba1c813a FW |
345 | mutex_lock(&nr_bp_mutex); |
346 | ||
f7136c51 | 347 | arch_unregister_hw_breakpoint(bp); |
5352ae63 | 348 | __release_bp_slot(bp); |
ba1c813a FW |
349 | |
350 | mutex_unlock(&nr_bp_mutex); | |
62a038d3 P |
351 | } |
352 | ||
5352ae63 JW |
353 | /* |
354 | * Allow the kernel debugger to reserve breakpoint slots without | |
355 | * taking a lock using the dbg_* variant of for the reserve and | |
356 | * release breakpoint slots. | |
357 | */ | |
358 | int dbg_reserve_bp_slot(struct perf_event *bp) | |
359 | { | |
360 | if (mutex_is_locked(&nr_bp_mutex)) | |
361 | return -1; | |
362 | ||
363 | return __reserve_bp_slot(bp); | |
364 | } | |
365 | ||
366 | int dbg_release_bp_slot(struct perf_event *bp) | |
367 | { | |
368 | if (mutex_is_locked(&nr_bp_mutex)) | |
369 | return -1; | |
370 | ||
371 | __release_bp_slot(bp); | |
372 | ||
373 | return 0; | |
374 | } | |
ba1c813a | 375 | |
b2812d03 FW |
376 | static int validate_hw_breakpoint(struct perf_event *bp) |
377 | { | |
378 | int ret; | |
379 | ||
380 | ret = arch_validate_hwbkpt_settings(bp); | |
381 | if (ret) | |
382 | return ret; | |
383 | ||
384 | if (arch_check_bp_in_kernelspace(bp)) { | |
385 | if (bp->attr.exclude_kernel) | |
386 | return -EINVAL; | |
387 | /* | |
388 | * Don't let unprivileged users set a breakpoint in the trap | |
389 | * path to avoid trap recursion attacks. | |
390 | */ | |
391 | if (!capable(CAP_SYS_ADMIN)) | |
392 | return -EPERM; | |
393 | } | |
394 | ||
395 | return 0; | |
396 | } | |
397 | ||
b326e956 | 398 | int register_perf_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 399 | { |
24f1e32c | 400 | int ret; |
62a038d3 | 401 | |
24f1e32c FW |
402 | ret = reserve_bp_slot(bp); |
403 | if (ret) | |
404 | return ret; | |
62a038d3 | 405 | |
b2812d03 | 406 | ret = validate_hw_breakpoint(bp); |
62a038d3 | 407 | |
b23ff0e9 MS |
408 | /* if arch_validate_hwbkpt_settings() fails then release bp slot */ |
409 | if (ret) | |
410 | release_bp_slot(bp); | |
411 | ||
24f1e32c FW |
412 | return ret; |
413 | } | |
62a038d3 | 414 | |
62a038d3 P |
415 | /** |
416 | * register_user_hw_breakpoint - register a hardware breakpoint for user space | |
5fa10b28 | 417 | * @attr: breakpoint attributes |
24f1e32c | 418 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 419 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 420 | */ |
24f1e32c | 421 | struct perf_event * |
5fa10b28 | 422 | register_user_hw_breakpoint(struct perf_event_attr *attr, |
b326e956 | 423 | perf_overflow_handler_t triggered, |
4dc0da86 | 424 | void *context, |
5fa10b28 | 425 | struct task_struct *tsk) |
62a038d3 | 426 | { |
4dc0da86 AK |
427 | return perf_event_create_kernel_counter(attr, -1, tsk, triggered, |
428 | context); | |
62a038d3 P |
429 | } |
430 | EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); | |
431 | ||
432 | /** | |
433 | * modify_user_hw_breakpoint - modify a user-space hardware breakpoint | |
24f1e32c | 434 | * @bp: the breakpoint structure to modify |
5fa10b28 | 435 | * @attr: new breakpoint attributes |
24f1e32c | 436 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 437 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 438 | */ |
44234adc | 439 | int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) |
62a038d3 | 440 | { |
44234adc | 441 | u64 old_addr = bp->attr.bp_addr; |
cd757645 | 442 | u64 old_len = bp->attr.bp_len; |
44234adc | 443 | int old_type = bp->attr.bp_type; |
44234adc FW |
444 | int err = 0; |
445 | ||
500ad2d8 P |
446 | /* |
447 | * modify_user_hw_breakpoint can be invoked with IRQs disabled and hence it | |
448 | * will not be possible to raise IPIs that invoke __perf_event_disable. | |
449 | * So call the function directly after making sure we are targeting the | |
450 | * current task. | |
451 | */ | |
452 | if (irqs_disabled() && bp->ctx && bp->ctx->task == current) | |
453 | __perf_event_disable(bp); | |
454 | else | |
455 | perf_event_disable(bp); | |
44234adc FW |
456 | |
457 | bp->attr.bp_addr = attr->bp_addr; | |
458 | bp->attr.bp_type = attr->bp_type; | |
459 | bp->attr.bp_len = attr->bp_len; | |
460 | ||
461 | if (attr->disabled) | |
462 | goto end; | |
62a038d3 | 463 | |
b2812d03 | 464 | err = validate_hw_breakpoint(bp); |
44234adc FW |
465 | if (!err) |
466 | perf_event_enable(bp); | |
467 | ||
468 | if (err) { | |
469 | bp->attr.bp_addr = old_addr; | |
470 | bp->attr.bp_type = old_type; | |
471 | bp->attr.bp_len = old_len; | |
472 | if (!bp->attr.disabled) | |
473 | perf_event_enable(bp); | |
474 | ||
475 | return err; | |
476 | } | |
477 | ||
478 | end: | |
479 | bp->attr.disabled = attr->disabled; | |
480 | ||
481 | return 0; | |
62a038d3 P |
482 | } |
483 | EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); | |
484 | ||
485 | /** | |
24f1e32c | 486 | * unregister_hw_breakpoint - unregister a user-space hardware breakpoint |
62a038d3 | 487 | * @bp: the breakpoint structure to unregister |
62a038d3 | 488 | */ |
24f1e32c | 489 | void unregister_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 490 | { |
24f1e32c FW |
491 | if (!bp) |
492 | return; | |
493 | perf_event_release_kernel(bp); | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); | |
496 | ||
62a038d3 | 497 | /** |
24f1e32c | 498 | * register_wide_hw_breakpoint - register a wide breakpoint in the kernel |
dd1853c3 | 499 | * @attr: breakpoint attributes |
24f1e32c | 500 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 501 | * |
24f1e32c | 502 | * @return a set of per_cpu pointers to perf events |
62a038d3 | 503 | */ |
44ee6358 | 504 | struct perf_event * __percpu * |
dd1853c3 | 505 | register_wide_hw_breakpoint(struct perf_event_attr *attr, |
4dc0da86 AK |
506 | perf_overflow_handler_t triggered, |
507 | void *context) | |
62a038d3 | 508 | { |
44ee6358 | 509 | struct perf_event * __percpu *cpu_events, **pevent, *bp; |
24f1e32c FW |
510 | long err; |
511 | int cpu; | |
512 | ||
513 | cpu_events = alloc_percpu(typeof(*cpu_events)); | |
514 | if (!cpu_events) | |
44ee6358 | 515 | return (void __percpu __force *)ERR_PTR(-ENOMEM); |
62a038d3 | 516 | |
88f7a890 LZ |
517 | get_online_cpus(); |
518 | for_each_online_cpu(cpu) { | |
24f1e32c | 519 | pevent = per_cpu_ptr(cpu_events, cpu); |
4dc0da86 AK |
520 | bp = perf_event_create_kernel_counter(attr, cpu, NULL, |
521 | triggered, context); | |
62a038d3 | 522 | |
24f1e32c | 523 | *pevent = bp; |
62a038d3 | 524 | |
605bfaee | 525 | if (IS_ERR(bp)) { |
24f1e32c FW |
526 | err = PTR_ERR(bp); |
527 | goto fail; | |
528 | } | |
62a038d3 | 529 | } |
88f7a890 | 530 | put_online_cpus(); |
62a038d3 | 531 | |
24f1e32c FW |
532 | return cpu_events; |
533 | ||
534 | fail: | |
88f7a890 | 535 | for_each_online_cpu(cpu) { |
24f1e32c | 536 | pevent = per_cpu_ptr(cpu_events, cpu); |
605bfaee | 537 | if (IS_ERR(*pevent)) |
24f1e32c FW |
538 | break; |
539 | unregister_hw_breakpoint(*pevent); | |
540 | } | |
88f7a890 LZ |
541 | put_online_cpus(); |
542 | ||
24f1e32c | 543 | free_percpu(cpu_events); |
44ee6358 | 544 | return (void __percpu __force *)ERR_PTR(err); |
62a038d3 | 545 | } |
f60d24d2 | 546 | EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint); |
62a038d3 P |
547 | |
548 | /** | |
24f1e32c FW |
549 | * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel |
550 | * @cpu_events: the per cpu set of events to unregister | |
62a038d3 | 551 | */ |
44ee6358 | 552 | void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events) |
62a038d3 | 553 | { |
24f1e32c FW |
554 | int cpu; |
555 | struct perf_event **pevent; | |
62a038d3 | 556 | |
24f1e32c FW |
557 | for_each_possible_cpu(cpu) { |
558 | pevent = per_cpu_ptr(cpu_events, cpu); | |
559 | unregister_hw_breakpoint(*pevent); | |
62a038d3 | 560 | } |
24f1e32c | 561 | free_percpu(cpu_events); |
62a038d3 | 562 | } |
f60d24d2 | 563 | EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint); |
62a038d3 P |
564 | |
565 | static struct notifier_block hw_breakpoint_exceptions_nb = { | |
566 | .notifier_call = hw_breakpoint_exceptions_notify, | |
567 | /* we need to be notified first */ | |
568 | .priority = 0x7fffffff | |
569 | }; | |
570 | ||
b0a873eb PZ |
571 | static void bp_perf_event_destroy(struct perf_event *event) |
572 | { | |
573 | release_bp_slot(event); | |
574 | } | |
575 | ||
576 | static int hw_breakpoint_event_init(struct perf_event *bp) | |
577 | { | |
578 | int err; | |
579 | ||
580 | if (bp->attr.type != PERF_TYPE_BREAKPOINT) | |
581 | return -ENOENT; | |
582 | ||
2481c5fa SE |
583 | /* |
584 | * no branch sampling for breakpoint events | |
585 | */ | |
586 | if (has_branch_stack(bp)) | |
587 | return -EOPNOTSUPP; | |
588 | ||
b0a873eb PZ |
589 | err = register_perf_hw_breakpoint(bp); |
590 | if (err) | |
591 | return err; | |
592 | ||
593 | bp->destroy = bp_perf_event_destroy; | |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
a4eaf7f1 PZ |
598 | static int hw_breakpoint_add(struct perf_event *bp, int flags) |
599 | { | |
600 | if (!(flags & PERF_EF_START)) | |
601 | bp->hw.state = PERF_HES_STOPPED; | |
602 | ||
ab573844 JO |
603 | if (is_sampling_event(bp)) { |
604 | bp->hw.last_period = bp->hw.sample_period; | |
605 | perf_swevent_set_period(bp); | |
606 | } | |
607 | ||
a4eaf7f1 PZ |
608 | return arch_install_hw_breakpoint(bp); |
609 | } | |
610 | ||
611 | static void hw_breakpoint_del(struct perf_event *bp, int flags) | |
612 | { | |
613 | arch_uninstall_hw_breakpoint(bp); | |
614 | } | |
615 | ||
616 | static void hw_breakpoint_start(struct perf_event *bp, int flags) | |
617 | { | |
618 | bp->hw.state = 0; | |
619 | } | |
620 | ||
621 | static void hw_breakpoint_stop(struct perf_event *bp, int flags) | |
622 | { | |
623 | bp->hw.state = PERF_HES_STOPPED; | |
624 | } | |
625 | ||
35edc2a5 PZ |
626 | static int hw_breakpoint_event_idx(struct perf_event *bp) |
627 | { | |
628 | return 0; | |
629 | } | |
630 | ||
b0a873eb | 631 | static struct pmu perf_breakpoint = { |
89a1e187 PZ |
632 | .task_ctx_nr = perf_sw_context, /* could eventually get its own */ |
633 | ||
b0a873eb | 634 | .event_init = hw_breakpoint_event_init, |
a4eaf7f1 PZ |
635 | .add = hw_breakpoint_add, |
636 | .del = hw_breakpoint_del, | |
637 | .start = hw_breakpoint_start, | |
638 | .stop = hw_breakpoint_stop, | |
b0a873eb | 639 | .read = hw_breakpoint_pmu_read, |
35edc2a5 PZ |
640 | |
641 | .event_idx = hw_breakpoint_event_idx, | |
b0a873eb PZ |
642 | }; |
643 | ||
3c502e7a | 644 | int __init init_hw_breakpoint(void) |
62a038d3 | 645 | { |
feef47d0 FW |
646 | unsigned int **task_bp_pinned; |
647 | int cpu, err_cpu; | |
648 | int i; | |
649 | ||
650 | for (i = 0; i < TYPE_MAX; i++) | |
651 | nr_slots[i] = hw_breakpoint_slots(i); | |
652 | ||
653 | for_each_possible_cpu(cpu) { | |
654 | for (i = 0; i < TYPE_MAX; i++) { | |
655 | task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu); | |
656 | *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i], | |
657 | GFP_KERNEL); | |
658 | if (!*task_bp_pinned) | |
659 | goto err_alloc; | |
660 | } | |
661 | } | |
662 | ||
663 | constraints_initialized = 1; | |
664 | ||
2e80a82a | 665 | perf_pmu_register(&perf_breakpoint, "breakpoint", PERF_TYPE_BREAKPOINT); |
b0a873eb | 666 | |
62a038d3 | 667 | return register_die_notifier(&hw_breakpoint_exceptions_nb); |
feef47d0 FW |
668 | |
669 | err_alloc: | |
670 | for_each_possible_cpu(err_cpu) { | |
feef47d0 | 671 | for (i = 0; i < TYPE_MAX; i++) |
02e176af | 672 | kfree(per_cpu(nr_task_bp_pinned[i], err_cpu)); |
30ce2f7e NK |
673 | if (err_cpu == cpu) |
674 | break; | |
feef47d0 FW |
675 | } |
676 | ||
677 | return -ENOMEM; | |
62a038d3 | 678 | } |
24f1e32c FW |
679 | |
680 |