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> | |
88f7a890 | 43 | #include <linux/cpu.h> |
62a038d3 P |
44 | #include <linux/smp.h> |
45 | ||
24f1e32c FW |
46 | #include <linux/hw_breakpoint.h> |
47 | ||
ba1c813a FW |
48 | /* |
49 | * Constraints data | |
50 | */ | |
62a038d3 | 51 | |
ba1c813a FW |
52 | /* Number of pinned cpu breakpoints in a cpu */ |
53 | static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned); | |
54 | ||
55 | /* Number of pinned task breakpoints in a cpu */ | |
6ab88863 | 56 | static DEFINE_PER_CPU(unsigned int, nr_task_bp_pinned[HBP_NUM]); |
ba1c813a FW |
57 | |
58 | /* Number of non-pinned cpu/task breakpoints in a cpu */ | |
59 | static DEFINE_PER_CPU(unsigned int, nr_bp_flexible); | |
60 | ||
61 | /* Gather the number of total pinned and un-pinned bp in a cpuset */ | |
62 | struct bp_busy_slots { | |
63 | unsigned int pinned; | |
64 | unsigned int flexible; | |
65 | }; | |
66 | ||
67 | /* Serialize accesses to the above constraints */ | |
68 | static DEFINE_MUTEX(nr_bp_mutex); | |
69 | ||
70 | /* | |
71 | * Report the maximum number of pinned breakpoints a task | |
72 | * have in this cpu | |
73 | */ | |
74 | static unsigned int max_task_bp_pinned(int cpu) | |
62a038d3 | 75 | { |
ba1c813a | 76 | int i; |
6ab88863 | 77 | unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned, cpu); |
62a038d3 | 78 | |
ba1c813a FW |
79 | for (i = HBP_NUM -1; i >= 0; i--) { |
80 | if (tsk_pinned[i] > 0) | |
81 | return i + 1; | |
62a038d3 P |
82 | } |
83 | ||
24f1e32c | 84 | return 0; |
62a038d3 P |
85 | } |
86 | ||
56053170 FW |
87 | static int task_bp_pinned(struct task_struct *tsk) |
88 | { | |
89 | struct perf_event_context *ctx = tsk->perf_event_ctxp; | |
90 | struct list_head *list; | |
91 | struct perf_event *bp; | |
92 | unsigned long flags; | |
93 | int count = 0; | |
94 | ||
95 | if (WARN_ONCE(!ctx, "No perf context for this task")) | |
96 | return 0; | |
97 | ||
98 | list = &ctx->event_list; | |
99 | ||
e625cce1 | 100 | raw_spin_lock_irqsave(&ctx->lock, flags); |
56053170 FW |
101 | |
102 | /* | |
103 | * The current breakpoint counter is not included in the list | |
104 | * at the open() callback time | |
105 | */ | |
106 | list_for_each_entry(bp, list, event_entry) { | |
107 | if (bp->attr.type == PERF_TYPE_BREAKPOINT) | |
108 | count++; | |
109 | } | |
110 | ||
e625cce1 | 111 | raw_spin_unlock_irqrestore(&ctx->lock, flags); |
56053170 FW |
112 | |
113 | return count; | |
114 | } | |
115 | ||
ba1c813a FW |
116 | /* |
117 | * Report the number of pinned/un-pinned breakpoints we have in | |
118 | * a given cpu (cpu > -1) or in all of them (cpu = -1). | |
119 | */ | |
56053170 FW |
120 | static void |
121 | fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp) | |
ba1c813a | 122 | { |
56053170 FW |
123 | int cpu = bp->cpu; |
124 | struct task_struct *tsk = bp->ctx->task; | |
125 | ||
ba1c813a FW |
126 | if (cpu >= 0) { |
127 | slots->pinned = per_cpu(nr_cpu_bp_pinned, cpu); | |
56053170 FW |
128 | if (!tsk) |
129 | slots->pinned += max_task_bp_pinned(cpu); | |
130 | else | |
131 | slots->pinned += task_bp_pinned(tsk); | |
ba1c813a FW |
132 | slots->flexible = per_cpu(nr_bp_flexible, cpu); |
133 | ||
134 | return; | |
135 | } | |
136 | ||
137 | for_each_online_cpu(cpu) { | |
138 | unsigned int nr; | |
139 | ||
140 | nr = per_cpu(nr_cpu_bp_pinned, cpu); | |
56053170 FW |
141 | if (!tsk) |
142 | nr += max_task_bp_pinned(cpu); | |
143 | else | |
144 | nr += task_bp_pinned(tsk); | |
ba1c813a FW |
145 | |
146 | if (nr > slots->pinned) | |
147 | slots->pinned = nr; | |
148 | ||
149 | nr = per_cpu(nr_bp_flexible, cpu); | |
150 | ||
151 | if (nr > slots->flexible) | |
152 | slots->flexible = nr; | |
153 | } | |
154 | } | |
155 | ||
156 | /* | |
157 | * Add a pinned breakpoint for the given task in our constraint table | |
158 | */ | |
159 | static void toggle_bp_task_slot(struct task_struct *tsk, int cpu, bool enable) | |
160 | { | |
11e66357 | 161 | unsigned int *tsk_pinned; |
56053170 | 162 | int count = 0; |
ba1c813a | 163 | |
56053170 | 164 | count = task_bp_pinned(tsk); |
ba1c813a | 165 | |
6ab88863 | 166 | tsk_pinned = per_cpu(nr_task_bp_pinned, cpu); |
ba1c813a | 167 | if (enable) { |
11e66357 | 168 | tsk_pinned[count]++; |
ba1c813a | 169 | if (count > 0) |
11e66357 | 170 | tsk_pinned[count-1]--; |
ba1c813a | 171 | } else { |
11e66357 | 172 | tsk_pinned[count]--; |
ba1c813a | 173 | if (count > 0) |
11e66357 | 174 | tsk_pinned[count-1]++; |
ba1c813a FW |
175 | } |
176 | } | |
177 | ||
178 | /* | |
179 | * Add/remove the given breakpoint in our constraint table | |
180 | */ | |
181 | static void toggle_bp_slot(struct perf_event *bp, bool enable) | |
182 | { | |
183 | int cpu = bp->cpu; | |
184 | struct task_struct *tsk = bp->ctx->task; | |
185 | ||
186 | /* Pinned counter task profiling */ | |
187 | if (tsk) { | |
188 | if (cpu >= 0) { | |
189 | toggle_bp_task_slot(tsk, cpu, enable); | |
190 | return; | |
191 | } | |
192 | ||
193 | for_each_online_cpu(cpu) | |
194 | toggle_bp_task_slot(tsk, cpu, enable); | |
195 | return; | |
196 | } | |
197 | ||
198 | /* Pinned counter cpu profiling */ | |
199 | if (enable) | |
200 | per_cpu(nr_cpu_bp_pinned, bp->cpu)++; | |
201 | else | |
202 | per_cpu(nr_cpu_bp_pinned, bp->cpu)--; | |
203 | } | |
204 | ||
205 | /* | |
206 | * Contraints to check before allowing this new breakpoint counter: | |
207 | * | |
208 | * == Non-pinned counter == (Considered as pinned for now) | |
209 | * | |
210 | * - If attached to a single cpu, check: | |
211 | * | |
212 | * (per_cpu(nr_bp_flexible, cpu) || (per_cpu(nr_cpu_bp_pinned, cpu) | |
6ab88863 | 213 | * + max(per_cpu(nr_task_bp_pinned, cpu)))) < HBP_NUM |
ba1c813a FW |
214 | * |
215 | * -> If there are already non-pinned counters in this cpu, it means | |
216 | * there is already a free slot for them. | |
217 | * Otherwise, we check that the maximum number of per task | |
218 | * breakpoints (for this cpu) plus the number of per cpu breakpoint | |
219 | * (for this cpu) doesn't cover every registers. | |
220 | * | |
221 | * - If attached to every cpus, check: | |
222 | * | |
223 | * (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *)) | |
6ab88863 | 224 | * + max(per_cpu(nr_task_bp_pinned, *)))) < HBP_NUM |
ba1c813a FW |
225 | * |
226 | * -> This is roughly the same, except we check the number of per cpu | |
227 | * bp for every cpu and we keep the max one. Same for the per tasks | |
228 | * breakpoints. | |
229 | * | |
230 | * | |
231 | * == Pinned counter == | |
232 | * | |
233 | * - If attached to a single cpu, check: | |
234 | * | |
235 | * ((per_cpu(nr_bp_flexible, cpu) > 1) + per_cpu(nr_cpu_bp_pinned, cpu) | |
6ab88863 | 236 | * + max(per_cpu(nr_task_bp_pinned, cpu))) < HBP_NUM |
ba1c813a FW |
237 | * |
238 | * -> Same checks as before. But now the nr_bp_flexible, if any, must keep | |
239 | * one register at least (or they will never be fed). | |
240 | * | |
241 | * - If attached to every cpus, check: | |
242 | * | |
243 | * ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *)) | |
6ab88863 | 244 | * + max(per_cpu(nr_task_bp_pinned, *))) < HBP_NUM |
ba1c813a | 245 | */ |
5352ae63 | 246 | static int __reserve_bp_slot(struct perf_event *bp) |
ba1c813a FW |
247 | { |
248 | struct bp_busy_slots slots = {0}; | |
ba1c813a | 249 | |
56053170 | 250 | fetch_bp_busy_slots(&slots, bp); |
ba1c813a FW |
251 | |
252 | /* Flexible counters need to keep at least one slot */ | |
5352ae63 JW |
253 | if (slots.pinned + (!!slots.flexible) == HBP_NUM) |
254 | return -ENOSPC; | |
ba1c813a FW |
255 | |
256 | toggle_bp_slot(bp, true); | |
257 | ||
5352ae63 JW |
258 | return 0; |
259 | } | |
260 | ||
261 | int reserve_bp_slot(struct perf_event *bp) | |
262 | { | |
263 | int ret; | |
264 | ||
265 | mutex_lock(&nr_bp_mutex); | |
266 | ||
267 | ret = __reserve_bp_slot(bp); | |
268 | ||
ba1c813a FW |
269 | mutex_unlock(&nr_bp_mutex); |
270 | ||
271 | return ret; | |
272 | } | |
273 | ||
5352ae63 JW |
274 | static void __release_bp_slot(struct perf_event *bp) |
275 | { | |
276 | toggle_bp_slot(bp, false); | |
277 | } | |
278 | ||
24f1e32c | 279 | void release_bp_slot(struct perf_event *bp) |
62a038d3 | 280 | { |
ba1c813a FW |
281 | mutex_lock(&nr_bp_mutex); |
282 | ||
5352ae63 | 283 | __release_bp_slot(bp); |
ba1c813a FW |
284 | |
285 | mutex_unlock(&nr_bp_mutex); | |
62a038d3 P |
286 | } |
287 | ||
5352ae63 JW |
288 | /* |
289 | * Allow the kernel debugger to reserve breakpoint slots without | |
290 | * taking a lock using the dbg_* variant of for the reserve and | |
291 | * release breakpoint slots. | |
292 | */ | |
293 | int dbg_reserve_bp_slot(struct perf_event *bp) | |
294 | { | |
295 | if (mutex_is_locked(&nr_bp_mutex)) | |
296 | return -1; | |
297 | ||
298 | return __reserve_bp_slot(bp); | |
299 | } | |
300 | ||
301 | int dbg_release_bp_slot(struct perf_event *bp) | |
302 | { | |
303 | if (mutex_is_locked(&nr_bp_mutex)) | |
304 | return -1; | |
305 | ||
306 | __release_bp_slot(bp); | |
307 | ||
308 | return 0; | |
309 | } | |
ba1c813a | 310 | |
b2812d03 FW |
311 | static int validate_hw_breakpoint(struct perf_event *bp) |
312 | { | |
313 | int ret; | |
314 | ||
315 | ret = arch_validate_hwbkpt_settings(bp); | |
316 | if (ret) | |
317 | return ret; | |
318 | ||
319 | if (arch_check_bp_in_kernelspace(bp)) { | |
320 | if (bp->attr.exclude_kernel) | |
321 | return -EINVAL; | |
322 | /* | |
323 | * Don't let unprivileged users set a breakpoint in the trap | |
324 | * path to avoid trap recursion attacks. | |
325 | */ | |
326 | if (!capable(CAP_SYS_ADMIN)) | |
327 | return -EPERM; | |
328 | } | |
329 | ||
330 | return 0; | |
331 | } | |
332 | ||
b326e956 | 333 | int register_perf_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 334 | { |
24f1e32c | 335 | int ret; |
62a038d3 | 336 | |
24f1e32c FW |
337 | ret = reserve_bp_slot(bp); |
338 | if (ret) | |
339 | return ret; | |
62a038d3 | 340 | |
b2812d03 | 341 | ret = validate_hw_breakpoint(bp); |
62a038d3 | 342 | |
b23ff0e9 MS |
343 | /* if arch_validate_hwbkpt_settings() fails then release bp slot */ |
344 | if (ret) | |
345 | release_bp_slot(bp); | |
346 | ||
24f1e32c FW |
347 | return ret; |
348 | } | |
62a038d3 | 349 | |
62a038d3 P |
350 | /** |
351 | * register_user_hw_breakpoint - register a hardware breakpoint for user space | |
5fa10b28 | 352 | * @attr: breakpoint attributes |
24f1e32c | 353 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 354 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 355 | */ |
24f1e32c | 356 | struct perf_event * |
5fa10b28 | 357 | register_user_hw_breakpoint(struct perf_event_attr *attr, |
b326e956 | 358 | perf_overflow_handler_t triggered, |
5fa10b28 | 359 | struct task_struct *tsk) |
62a038d3 | 360 | { |
5fa10b28 | 361 | return perf_event_create_kernel_counter(attr, -1, tsk->pid, triggered); |
62a038d3 P |
362 | } |
363 | EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); | |
364 | ||
365 | /** | |
366 | * modify_user_hw_breakpoint - modify a user-space hardware breakpoint | |
24f1e32c | 367 | * @bp: the breakpoint structure to modify |
5fa10b28 | 368 | * @attr: new breakpoint attributes |
24f1e32c | 369 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 370 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
62a038d3 | 371 | */ |
44234adc | 372 | int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) |
62a038d3 | 373 | { |
44234adc | 374 | u64 old_addr = bp->attr.bp_addr; |
cd757645 | 375 | u64 old_len = bp->attr.bp_len; |
44234adc | 376 | int old_type = bp->attr.bp_type; |
44234adc FW |
377 | int err = 0; |
378 | ||
379 | perf_event_disable(bp); | |
380 | ||
381 | bp->attr.bp_addr = attr->bp_addr; | |
382 | bp->attr.bp_type = attr->bp_type; | |
383 | bp->attr.bp_len = attr->bp_len; | |
384 | ||
385 | if (attr->disabled) | |
386 | goto end; | |
62a038d3 | 387 | |
b2812d03 | 388 | err = validate_hw_breakpoint(bp); |
44234adc FW |
389 | if (!err) |
390 | perf_event_enable(bp); | |
391 | ||
392 | if (err) { | |
393 | bp->attr.bp_addr = old_addr; | |
394 | bp->attr.bp_type = old_type; | |
395 | bp->attr.bp_len = old_len; | |
396 | if (!bp->attr.disabled) | |
397 | perf_event_enable(bp); | |
398 | ||
399 | return err; | |
400 | } | |
401 | ||
402 | end: | |
403 | bp->attr.disabled = attr->disabled; | |
404 | ||
405 | return 0; | |
62a038d3 P |
406 | } |
407 | EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); | |
408 | ||
409 | /** | |
24f1e32c | 410 | * unregister_hw_breakpoint - unregister a user-space hardware breakpoint |
62a038d3 | 411 | * @bp: the breakpoint structure to unregister |
62a038d3 | 412 | */ |
24f1e32c | 413 | void unregister_hw_breakpoint(struct perf_event *bp) |
62a038d3 | 414 | { |
24f1e32c FW |
415 | if (!bp) |
416 | return; | |
417 | perf_event_release_kernel(bp); | |
418 | } | |
419 | EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); | |
420 | ||
62a038d3 | 421 | /** |
24f1e32c | 422 | * register_wide_hw_breakpoint - register a wide breakpoint in the kernel |
dd1853c3 | 423 | * @attr: breakpoint attributes |
24f1e32c | 424 | * @triggered: callback to trigger when we hit the breakpoint |
62a038d3 | 425 | * |
24f1e32c | 426 | * @return a set of per_cpu pointers to perf events |
62a038d3 | 427 | */ |
44ee6358 | 428 | struct perf_event * __percpu * |
dd1853c3 | 429 | register_wide_hw_breakpoint(struct perf_event_attr *attr, |
b326e956 | 430 | perf_overflow_handler_t triggered) |
62a038d3 | 431 | { |
44ee6358 | 432 | struct perf_event * __percpu *cpu_events, **pevent, *bp; |
24f1e32c FW |
433 | long err; |
434 | int cpu; | |
435 | ||
436 | cpu_events = alloc_percpu(typeof(*cpu_events)); | |
437 | if (!cpu_events) | |
44ee6358 | 438 | return (void __percpu __force *)ERR_PTR(-ENOMEM); |
62a038d3 | 439 | |
88f7a890 LZ |
440 | get_online_cpus(); |
441 | for_each_online_cpu(cpu) { | |
24f1e32c | 442 | pevent = per_cpu_ptr(cpu_events, cpu); |
dd1853c3 | 443 | bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered); |
62a038d3 | 444 | |
24f1e32c | 445 | *pevent = bp; |
62a038d3 | 446 | |
605bfaee | 447 | if (IS_ERR(bp)) { |
24f1e32c FW |
448 | err = PTR_ERR(bp); |
449 | goto fail; | |
450 | } | |
62a038d3 | 451 | } |
88f7a890 | 452 | put_online_cpus(); |
62a038d3 | 453 | |
24f1e32c FW |
454 | return cpu_events; |
455 | ||
456 | fail: | |
88f7a890 | 457 | for_each_online_cpu(cpu) { |
24f1e32c | 458 | pevent = per_cpu_ptr(cpu_events, cpu); |
605bfaee | 459 | if (IS_ERR(*pevent)) |
24f1e32c FW |
460 | break; |
461 | unregister_hw_breakpoint(*pevent); | |
462 | } | |
88f7a890 LZ |
463 | put_online_cpus(); |
464 | ||
24f1e32c | 465 | free_percpu(cpu_events); |
44ee6358 | 466 | return (void __percpu __force *)ERR_PTR(err); |
62a038d3 | 467 | } |
f60d24d2 | 468 | EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint); |
62a038d3 P |
469 | |
470 | /** | |
24f1e32c FW |
471 | * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel |
472 | * @cpu_events: the per cpu set of events to unregister | |
62a038d3 | 473 | */ |
44ee6358 | 474 | void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events) |
62a038d3 | 475 | { |
24f1e32c FW |
476 | int cpu; |
477 | struct perf_event **pevent; | |
62a038d3 | 478 | |
24f1e32c FW |
479 | for_each_possible_cpu(cpu) { |
480 | pevent = per_cpu_ptr(cpu_events, cpu); | |
481 | unregister_hw_breakpoint(*pevent); | |
62a038d3 | 482 | } |
24f1e32c | 483 | free_percpu(cpu_events); |
62a038d3 | 484 | } |
f60d24d2 | 485 | EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint); |
62a038d3 P |
486 | |
487 | static struct notifier_block hw_breakpoint_exceptions_nb = { | |
488 | .notifier_call = hw_breakpoint_exceptions_notify, | |
489 | /* we need to be notified first */ | |
490 | .priority = 0x7fffffff | |
491 | }; | |
492 | ||
493 | static int __init init_hw_breakpoint(void) | |
494 | { | |
495 | return register_die_notifier(&hw_breakpoint_exceptions_nb); | |
496 | } | |
62a038d3 | 497 | core_initcall(init_hw_breakpoint); |
24f1e32c FW |
498 | |
499 | ||
500 | struct pmu perf_ops_bp = { | |
501 | .enable = arch_install_hw_breakpoint, | |
502 | .disable = arch_uninstall_hw_breakpoint, | |
503 | .read = hw_breakpoint_pmu_read, | |
24f1e32c | 504 | }; |