Merge branch 'tracing/core' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[linux-2.6-block.git] / arch / x86 / kernel / hw_breakpoint.c
CommitLineData
0067f129
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) 2009 IBM Corporation
24f1e32c 18 * Copyright (C) 2009 Frederic Weisbecker <fweisbec@gmail.com>
ba6909b7
P
19 *
20 * Authors: Alan Stern <stern@rowland.harvard.edu>
21 * K.Prasad <prasad@linux.vnet.ibm.com>
22 * Frederic Weisbecker <fweisbec@gmail.com>
0067f129
P
23 */
24
25/*
26 * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility,
27 * using the CPU's debug registers.
28 */
29
24f1e32c
FW
30#include <linux/perf_event.h>
31#include <linux/hw_breakpoint.h>
0067f129
P
32#include <linux/irqflags.h>
33#include <linux/notifier.h>
34#include <linux/kallsyms.h>
35#include <linux/kprobes.h>
36#include <linux/percpu.h>
37#include <linux/kdebug.h>
38#include <linux/kernel.h>
39#include <linux/module.h>
40#include <linux/sched.h>
41#include <linux/init.h>
42#include <linux/smp.h>
43
44#include <asm/hw_breakpoint.h>
45#include <asm/processor.h>
46#include <asm/debugreg.h>
47
24f1e32c 48/* Per cpu debug control register value */
28b4e0d8
TH
49DEFINE_PER_CPU(unsigned long, cpu_dr7);
50EXPORT_PER_CPU_SYMBOL(cpu_dr7);
24f1e32c
FW
51
52/* Per cpu debug address registers values */
53static DEFINE_PER_CPU(unsigned long, cpu_debugreg[HBP_NUM]);
0067f129
P
54
55/*
24f1e32c
FW
56 * Stores the breakpoints currently in use on each breakpoint address
57 * register for each cpus
0067f129 58 */
24f1e32c 59static DEFINE_PER_CPU(struct perf_event *, bp_per_reg[HBP_NUM]);
0067f129
P
60
61
2c31b795
FW
62static inline unsigned long
63__encode_dr7(int drnum, unsigned int len, unsigned int type)
0067f129
P
64{
65 unsigned long bp_info;
66
67 bp_info = (len | type) & 0xf;
68 bp_info <<= (DR_CONTROL_SHIFT + drnum * DR_CONTROL_SIZE);
2c31b795
FW
69 bp_info |= (DR_GLOBAL_ENABLE << (drnum * DR_ENABLE_SIZE));
70
0067f129
P
71 return bp_info;
72}
73
2c31b795
FW
74/*
75 * Encode the length, type, Exact, and Enable bits for a particular breakpoint
76 * as stored in debug register 7.
77 */
78unsigned long encode_dr7(int drnum, unsigned int len, unsigned int type)
79{
80 return __encode_dr7(drnum, len, type) | DR_GLOBAL_SLOWDOWN;
81}
82
24f1e32c
FW
83/*
84 * Decode the length and type bits for a particular breakpoint as
85 * stored in debug register 7. Return the "enabled" status.
86 */
87int decode_dr7(unsigned long dr7, int bpnum, unsigned *len, unsigned *type)
0067f129 88{
24f1e32c 89 int bp_info = dr7 >> (DR_CONTROL_SHIFT + bpnum * DR_CONTROL_SIZE);
0067f129 90
24f1e32c
FW
91 *len = (bp_info & 0xc) | 0x40;
92 *type = (bp_info & 0x3) | 0x80;
0067f129 93
24f1e32c 94 return (dr7 >> (bpnum * DR_ENABLE_SIZE)) & 0x3;
0067f129
P
95}
96
97/*
24f1e32c
FW
98 * Install a perf counter breakpoint.
99 *
100 * We seek a free debug address register and use it for this
101 * breakpoint. Eventually we enable it in the debug control register.
102 *
103 * Atomic: we hold the counter->ctx->lock and we only handle variables
104 * and registers local to this cpu.
0067f129 105 */
24f1e32c 106int arch_install_hw_breakpoint(struct perf_event *bp)
0067f129 107{
24f1e32c
FW
108 struct arch_hw_breakpoint *info = counter_arch_bp(bp);
109 unsigned long *dr7;
110 int i;
111
112 for (i = 0; i < HBP_NUM; i++) {
113 struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]);
114
115 if (!*slot) {
116 *slot = bp;
117 break;
118 }
0067f129
P
119 }
120
24f1e32c
FW
121 if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
122 return -EBUSY;
123
124 set_debugreg(info->address, i);
125 __get_cpu_var(cpu_debugreg[i]) = info->address;
126
28b4e0d8 127 dr7 = &__get_cpu_var(cpu_dr7);
24f1e32c
FW
128 *dr7 |= encode_dr7(i, info->len, info->type);
129
130 set_debugreg(*dr7, 7);
131
132 return 0;
0067f129
P
133}
134
135/*
24f1e32c
FW
136 * Uninstall the breakpoint contained in the given counter.
137 *
138 * First we search the debug address register it uses and then we disable
139 * it.
140 *
141 * Atomic: we hold the counter->ctx->lock and we only handle variables
142 * and registers local to this cpu.
0067f129 143 */
24f1e32c 144void arch_uninstall_hw_breakpoint(struct perf_event *bp)
0067f129 145{
24f1e32c
FW
146 struct arch_hw_breakpoint *info = counter_arch_bp(bp);
147 unsigned long *dr7;
148 int i;
149
150 for (i = 0; i < HBP_NUM; i++) {
151 struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]);
152
153 if (*slot == bp) {
154 *slot = NULL;
155 break;
156 }
157 }
158
159 if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
160 return;
0067f129 161
28b4e0d8 162 dr7 = &__get_cpu_var(cpu_dr7);
2c31b795 163 *dr7 &= ~__encode_dr7(i, info->len, info->type);
24f1e32c
FW
164
165 set_debugreg(*dr7, 7);
0067f129
P
166}
167
168static int get_hbp_len(u8 hbp_len)
169{
170 unsigned int len_in_bytes = 0;
171
172 switch (hbp_len) {
24f1e32c 173 case X86_BREAKPOINT_LEN_1:
0067f129
P
174 len_in_bytes = 1;
175 break;
24f1e32c 176 case X86_BREAKPOINT_LEN_2:
0067f129
P
177 len_in_bytes = 2;
178 break;
24f1e32c 179 case X86_BREAKPOINT_LEN_4:
0067f129
P
180 len_in_bytes = 4;
181 break;
182#ifdef CONFIG_X86_64
24f1e32c 183 case X86_BREAKPOINT_LEN_8:
0067f129
P
184 len_in_bytes = 8;
185 break;
186#endif
187 }
188 return len_in_bytes;
189}
190
191/*
192 * Check for virtual address in user space.
193 */
194int arch_check_va_in_userspace(unsigned long va, u8 hbp_len)
195{
196 unsigned int len;
197
198 len = get_hbp_len(hbp_len);
199
200 return (va <= TASK_SIZE - len);
201}
202
203/*
204 * Check for virtual address in kernel space.
205 */
4555835b 206static int arch_check_va_in_kernelspace(unsigned long va, u8 hbp_len)
0067f129
P
207{
208 unsigned int len;
209
210 len = get_hbp_len(hbp_len);
211
212 return (va >= TASK_SIZE) && ((va + len - 1) >= TASK_SIZE);
213}
214
24f1e32c
FW
215int arch_bp_generic_fields(int x86_len, int x86_type,
216 int *gen_len, int *gen_type)
0067f129 217{
24f1e32c
FW
218 /* Len */
219 switch (x86_len) {
220 case X86_BREAKPOINT_LEN_1:
221 *gen_len = HW_BREAKPOINT_LEN_1;
222 break;
223 case X86_BREAKPOINT_LEN_2:
224 *gen_len = HW_BREAKPOINT_LEN_2;
225 break;
226 case X86_BREAKPOINT_LEN_4:
227 *gen_len = HW_BREAKPOINT_LEN_4;
228 break;
229#ifdef CONFIG_X86_64
230 case X86_BREAKPOINT_LEN_8:
231 *gen_len = HW_BREAKPOINT_LEN_8;
232 break;
233#endif
234 default:
235 return -EINVAL;
236 }
0067f129 237
24f1e32c
FW
238 /* Type */
239 switch (x86_type) {
240 case X86_BREAKPOINT_EXECUTE:
241 *gen_type = HW_BREAKPOINT_X;
0067f129 242 break;
24f1e32c
FW
243 case X86_BREAKPOINT_WRITE:
244 *gen_type = HW_BREAKPOINT_W;
0067f129 245 break;
24f1e32c
FW
246 case X86_BREAKPOINT_RW:
247 *gen_type = HW_BREAKPOINT_W | HW_BREAKPOINT_R;
0067f129
P
248 break;
249 default:
24f1e32c 250 return -EINVAL;
0067f129
P
251 }
252
24f1e32c
FW
253 return 0;
254}
255
256
257static int arch_build_bp_info(struct perf_event *bp)
258{
259 struct arch_hw_breakpoint *info = counter_arch_bp(bp);
260
261 info->address = bp->attr.bp_addr;
262
263 /* Len */
264 switch (bp->attr.bp_len) {
0067f129 265 case HW_BREAKPOINT_LEN_1:
24f1e32c 266 info->len = X86_BREAKPOINT_LEN_1;
0067f129
P
267 break;
268 case HW_BREAKPOINT_LEN_2:
24f1e32c 269 info->len = X86_BREAKPOINT_LEN_2;
0067f129
P
270 break;
271 case HW_BREAKPOINT_LEN_4:
24f1e32c 272 info->len = X86_BREAKPOINT_LEN_4;
0067f129
P
273 break;
274#ifdef CONFIG_X86_64
275 case HW_BREAKPOINT_LEN_8:
24f1e32c
FW
276 info->len = X86_BREAKPOINT_LEN_8;
277 break;
278#endif
279 default:
280 return -EINVAL;
281 }
282
283 /* Type */
284 switch (bp->attr.bp_type) {
285 case HW_BREAKPOINT_W:
286 info->type = X86_BREAKPOINT_WRITE;
287 break;
288 case HW_BREAKPOINT_W | HW_BREAKPOINT_R:
289 info->type = X86_BREAKPOINT_RW;
290 break;
291 case HW_BREAKPOINT_X:
292 info->type = X86_BREAKPOINT_EXECUTE;
293 break;
294 default:
295 return -EINVAL;
296 }
297
298 return 0;
299}
300/*
301 * Validate the arch-specific HW Breakpoint register settings
302 */
303int arch_validate_hwbkpt_settings(struct perf_event *bp,
304 struct task_struct *tsk)
305{
306 struct arch_hw_breakpoint *info = counter_arch_bp(bp);
307 unsigned int align;
308 int ret;
309
310
311 ret = arch_build_bp_info(bp);
312 if (ret)
313 return ret;
314
315 ret = -EINVAL;
316
317 if (info->type == X86_BREAKPOINT_EXECUTE)
318 /*
319 * Ptrace-refactoring code
320 * For now, we'll allow instruction breakpoint only for user-space
321 * addresses
322 */
323 if ((!arch_check_va_in_userspace(info->address, info->len)) &&
324 info->len != X86_BREAKPOINT_EXECUTE)
325 return ret;
326
327 switch (info->len) {
328 case X86_BREAKPOINT_LEN_1:
329 align = 0;
330 break;
331 case X86_BREAKPOINT_LEN_2:
332 align = 1;
333 break;
334 case X86_BREAKPOINT_LEN_4:
335 align = 3;
336 break;
337#ifdef CONFIG_X86_64
338 case X86_BREAKPOINT_LEN_8:
0067f129
P
339 align = 7;
340 break;
341#endif
342 default:
343 return ret;
344 }
345
84d71092
FW
346 /*
347 * For kernel-addresses, either the address or symbol name can be
348 * specified.
349 */
350 if (info->name)
351 info->address = (unsigned long)
352 kallsyms_lookup_name(info->name);
0067f129
P
353 /*
354 * Check that the low-order bits of the address are appropriate
355 * for the alignment implied by len.
356 */
24f1e32c 357 if (info->address & align)
0067f129
P
358 return -EINVAL;
359
360 /* Check that the virtual address is in the proper range */
361 if (tsk) {
24f1e32c 362 if (!arch_check_va_in_userspace(info->address, info->len))
0067f129
P
363 return -EFAULT;
364 } else {
24f1e32c 365 if (!arch_check_va_in_kernelspace(info->address, info->len))
0067f129
P
366 return -EFAULT;
367 }
24f1e32c 368
0067f129
P
369 return 0;
370}
371
9f6b3c2c
FW
372/*
373 * Dump the debug register contents to the user.
374 * We can't dump our per cpu values because it
375 * may contain cpu wide breakpoint, something that
376 * doesn't belong to the current task.
377 *
378 * TODO: include non-ptrace user breakpoints (perf)
379 */
380void aout_dump_debugregs(struct user *dump)
381{
382 int i;
383 int dr7 = 0;
384 struct perf_event *bp;
385 struct arch_hw_breakpoint *info;
386 struct thread_struct *thread = &current->thread;
387
388 for (i = 0; i < HBP_NUM; i++) {
389 bp = thread->ptrace_bps[i];
390
391 if (bp && !bp->attr.disabled) {
392 dump->u_debugreg[i] = bp->attr.bp_addr;
393 info = counter_arch_bp(bp);
394 dr7 |= encode_dr7(i, info->len, info->type);
395 } else {
396 dump->u_debugreg[i] = 0;
397 }
398 }
399
400 dump->u_debugreg[4] = 0;
401 dump->u_debugreg[5] = 0;
402 dump->u_debugreg[6] = current->thread.debugreg6;
403
404 dump->u_debugreg[7] = dr7;
405}
68efa37d 406EXPORT_SYMBOL_GPL(aout_dump_debugregs);
9f6b3c2c 407
24f1e32c
FW
408/*
409 * Release the user breakpoints used by ptrace
410 */
411void flush_ptrace_hw_breakpoint(struct task_struct *tsk)
0067f129 412{
24f1e32c
FW
413 int i;
414 struct thread_struct *t = &tsk->thread;
415
416 for (i = 0; i < HBP_NUM; i++) {
417 unregister_hw_breakpoint(t->ptrace_bps[i]);
418 t->ptrace_bps[i] = NULL;
419 }
0067f129
P
420}
421
24f1e32c 422void hw_breakpoint_restore(void)
0067f129 423{
24f1e32c
FW
424 set_debugreg(__get_cpu_var(cpu_debugreg[0]), 0);
425 set_debugreg(__get_cpu_var(cpu_debugreg[1]), 1);
426 set_debugreg(__get_cpu_var(cpu_debugreg[2]), 2);
427 set_debugreg(__get_cpu_var(cpu_debugreg[3]), 3);
428 set_debugreg(current->thread.debugreg6, 6);
28b4e0d8 429 set_debugreg(__get_cpu_var(cpu_dr7), 7);
0067f129 430}
24f1e32c 431EXPORT_SYMBOL_GPL(hw_breakpoint_restore);
0067f129
P
432
433/*
434 * Handle debug exception notifications.
435 *
436 * Return value is either NOTIFY_STOP or NOTIFY_DONE as explained below.
437 *
438 * NOTIFY_DONE returned if one of the following conditions is true.
439 * i) When the causative address is from user-space and the exception
440 * is a valid one, i.e. not triggered as a result of lazy debug register
441 * switching
442 * ii) When there are more bits than trap<n> set in DR6 register (such
443 * as BD, BS or BT) indicating that more than one debug condition is
444 * met and requires some more action in do_debug().
445 *
446 * NOTIFY_STOP returned for all other cases
447 *
448 */
4555835b 449static int __kprobes hw_breakpoint_handler(struct die_args *args)
0067f129
P
450{
451 int i, cpu, rc = NOTIFY_STOP;
24f1e32c 452 struct perf_event *bp;
62edab90
P
453 unsigned long dr7, dr6;
454 unsigned long *dr6_p;
455
456 /* The DR6 value is pointed by args->err */
457 dr6_p = (unsigned long *)ERR_PTR(args->err);
458 dr6 = *dr6_p;
0067f129
P
459
460 /* Do an early return if no trap bits are set in DR6 */
461 if ((dr6 & DR_TRAP_BITS) == 0)
462 return NOTIFY_DONE;
463
0067f129
P
464 get_debugreg(dr7, 7);
465 /* Disable breakpoints during exception handling */
466 set_debugreg(0UL, 7);
467 /*
468 * Assert that local interrupts are disabled
469 * Reset the DRn bits in the virtualized register value.
470 * The ptrace trigger routine will add in whatever is needed.
471 */
472 current->thread.debugreg6 &= ~DR_TRAP_BITS;
473 cpu = get_cpu();
474
475 /* Handle all the breakpoints that were triggered */
476 for (i = 0; i < HBP_NUM; ++i) {
477 if (likely(!(dr6 & (DR_TRAP0 << i))))
478 continue;
24f1e32c 479
0067f129 480 /*
24f1e32c
FW
481 * The counter may be concurrently released but that can only
482 * occur from a call_rcu() path. We can then safely fetch
483 * the breakpoint, use its callback, touch its counter
484 * while we are in an rcu_read_lock() path.
0067f129 485 */
24f1e32c
FW
486 rcu_read_lock();
487
488 bp = per_cpu(bp_per_reg[i], cpu);
489 if (bp)
490 rc = NOTIFY_DONE;
62edab90
P
491 /*
492 * Reset the 'i'th TRAP bit in dr6 to denote completion of
493 * exception handling
494 */
495 (*dr6_p) &= ~(DR_TRAP0 << i);
0067f129
P
496 /*
497 * bp can be NULL due to lazy debug register switching
24f1e32c 498 * or due to concurrent perf counter removing.
0067f129 499 */
24f1e32c
FW
500 if (!bp) {
501 rcu_read_unlock();
502 break;
503 }
504
b326e956 505 perf_bp_event(bp, args->regs);
0067f129 506
24f1e32c 507 rcu_read_unlock();
0067f129
P
508 }
509 if (dr6 & (~DR_TRAP_BITS))
510 rc = NOTIFY_DONE;
511
512 set_debugreg(dr7, 7);
eadb8a09 513 put_cpu();
24f1e32c 514
0067f129
P
515 return rc;
516}
517
518/*
519 * Handle debug exception notifications.
520 */
521int __kprobes hw_breakpoint_exceptions_notify(
522 struct notifier_block *unused, unsigned long val, void *data)
523{
524 if (val != DIE_DEBUG)
525 return NOTIFY_DONE;
526
527 return hw_breakpoint_handler(data);
528}
24f1e32c
FW
529
530void hw_breakpoint_pmu_read(struct perf_event *bp)
531{
532 /* TODO */
533}
534
535void hw_breakpoint_pmu_unthrottle(struct perf_event *bp)
536{
537 /* TODO */
538}