Commit 068e35eee9ef98eb4cab55181977e24995d273be
Committed by
Frederic Weisbecker
1 parent
89e45aac42
Exists in
master
and in
7 other branches
hw breakpoints: Fix pid namespace bug
Hardware breakpoints can't be registered within pid namespaces because tsk->pid is passed rather than the pid in the current namespace. (See https://bugzilla.kernel.org/show_bug.cgi?id=17281 ) This is a quick fix demonstrating the problem but is not the best method of solving the problem since passing pids internally is not the best way to avoid pid namespace bugs. Subsequent patches will show a better solution. Much thanks to Frederic Weisbecker <fweisbec@gmail.com> for doing the bulk of the work finding this bug. Reported-by: Robin Green <greenrd@greenrd.org> Signed-off-by: Matt Helsley <matthltc@us.ibm.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Prasad <prasad@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Will Deacon <will.deacon@arm.com> Cc: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> Cc: 2.6.33-2.6.35 <stable@kernel.org> LKML-Reference: <f63454af09fb1915717251570423eb9ddd338340.1284407762.git.matthltc@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Showing 1 changed file with 2 additions and 1 deletions Inline Diff
kernel/hw_breakpoint.c
1 | /* | 1 | /* |
2 | * This program is free software; you can redistribute it and/or modify | 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 | 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 | 4 | * the Free Software Foundation; either version 2 of the License, or |
5 | * (at your option) any later version. | 5 | * (at your option) any later version. |
6 | * | 6 | * |
7 | * This program is distributed in the hope that it will be useful, | 7 | * This program is distributed in the hope that it will be useful, |
8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | 8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | 9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
10 | * GNU General Public License for more details. | 10 | * GNU General Public License for more details. |
11 | * | 11 | * |
12 | * You should have received a copy of the GNU General Public License | 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 | 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. | 14 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
15 | * | 15 | * |
16 | * Copyright (C) 2007 Alan Stern | 16 | * Copyright (C) 2007 Alan Stern |
17 | * Copyright (C) IBM Corporation, 2009 | 17 | * Copyright (C) IBM Corporation, 2009 |
18 | * Copyright (C) 2009, Frederic Weisbecker <fweisbec@gmail.com> | 18 | * Copyright (C) 2009, Frederic Weisbecker <fweisbec@gmail.com> |
19 | * | 19 | * |
20 | * Thanks to Ingo Molnar for his many suggestions. | 20 | * Thanks to Ingo Molnar for his many suggestions. |
21 | * | 21 | * |
22 | * Authors: Alan Stern <stern@rowland.harvard.edu> | 22 | * Authors: Alan Stern <stern@rowland.harvard.edu> |
23 | * K.Prasad <prasad@linux.vnet.ibm.com> | 23 | * K.Prasad <prasad@linux.vnet.ibm.com> |
24 | * Frederic Weisbecker <fweisbec@gmail.com> | 24 | * Frederic Weisbecker <fweisbec@gmail.com> |
25 | */ | 25 | */ |
26 | 26 | ||
27 | /* | 27 | /* |
28 | * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility, | 28 | * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility, |
29 | * using the CPU's debug registers. | 29 | * using the CPU's debug registers. |
30 | * This file contains the arch-independent routines. | 30 | * This file contains the arch-independent routines. |
31 | */ | 31 | */ |
32 | 32 | ||
33 | #include <linux/irqflags.h> | 33 | #include <linux/irqflags.h> |
34 | #include <linux/kallsyms.h> | 34 | #include <linux/kallsyms.h> |
35 | #include <linux/notifier.h> | 35 | #include <linux/notifier.h> |
36 | #include <linux/kprobes.h> | 36 | #include <linux/kprobes.h> |
37 | #include <linux/kdebug.h> | 37 | #include <linux/kdebug.h> |
38 | #include <linux/kernel.h> | 38 | #include <linux/kernel.h> |
39 | #include <linux/module.h> | 39 | #include <linux/module.h> |
40 | #include <linux/percpu.h> | 40 | #include <linux/percpu.h> |
41 | #include <linux/sched.h> | 41 | #include <linux/sched.h> |
42 | #include <linux/init.h> | 42 | #include <linux/init.h> |
43 | #include <linux/slab.h> | 43 | #include <linux/slab.h> |
44 | #include <linux/list.h> | 44 | #include <linux/list.h> |
45 | #include <linux/cpu.h> | 45 | #include <linux/cpu.h> |
46 | #include <linux/smp.h> | 46 | #include <linux/smp.h> |
47 | 47 | ||
48 | #include <linux/hw_breakpoint.h> | 48 | #include <linux/hw_breakpoint.h> |
49 | 49 | ||
50 | 50 | ||
51 | /* | 51 | /* |
52 | * Constraints data | 52 | * Constraints data |
53 | */ | 53 | */ |
54 | 54 | ||
55 | /* Number of pinned cpu breakpoints in a cpu */ | 55 | /* Number of pinned cpu breakpoints in a cpu */ |
56 | static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]); | 56 | static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]); |
57 | 57 | ||
58 | /* Number of pinned task breakpoints in a cpu */ | 58 | /* Number of pinned task breakpoints in a cpu */ |
59 | static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]); | 59 | static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]); |
60 | 60 | ||
61 | /* Number of non-pinned cpu/task breakpoints in a cpu */ | 61 | /* Number of non-pinned cpu/task breakpoints in a cpu */ |
62 | static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]); | 62 | static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]); |
63 | 63 | ||
64 | static int nr_slots[TYPE_MAX]; | 64 | static int nr_slots[TYPE_MAX]; |
65 | 65 | ||
66 | /* Keep track of the breakpoints attached to tasks */ | 66 | /* Keep track of the breakpoints attached to tasks */ |
67 | static LIST_HEAD(bp_task_head); | 67 | static LIST_HEAD(bp_task_head); |
68 | 68 | ||
69 | static int constraints_initialized; | 69 | static int constraints_initialized; |
70 | 70 | ||
71 | /* Gather the number of total pinned and un-pinned bp in a cpuset */ | 71 | /* Gather the number of total pinned and un-pinned bp in a cpuset */ |
72 | struct bp_busy_slots { | 72 | struct bp_busy_slots { |
73 | unsigned int pinned; | 73 | unsigned int pinned; |
74 | unsigned int flexible; | 74 | unsigned int flexible; |
75 | }; | 75 | }; |
76 | 76 | ||
77 | /* Serialize accesses to the above constraints */ | 77 | /* Serialize accesses to the above constraints */ |
78 | static DEFINE_MUTEX(nr_bp_mutex); | 78 | static DEFINE_MUTEX(nr_bp_mutex); |
79 | 79 | ||
80 | __weak int hw_breakpoint_weight(struct perf_event *bp) | 80 | __weak int hw_breakpoint_weight(struct perf_event *bp) |
81 | { | 81 | { |
82 | return 1; | 82 | return 1; |
83 | } | 83 | } |
84 | 84 | ||
85 | static inline enum bp_type_idx find_slot_idx(struct perf_event *bp) | 85 | static inline enum bp_type_idx find_slot_idx(struct perf_event *bp) |
86 | { | 86 | { |
87 | if (bp->attr.bp_type & HW_BREAKPOINT_RW) | 87 | if (bp->attr.bp_type & HW_BREAKPOINT_RW) |
88 | return TYPE_DATA; | 88 | return TYPE_DATA; |
89 | 89 | ||
90 | return TYPE_INST; | 90 | return TYPE_INST; |
91 | } | 91 | } |
92 | 92 | ||
93 | /* | 93 | /* |
94 | * Report the maximum number of pinned breakpoints a task | 94 | * Report the maximum number of pinned breakpoints a task |
95 | * have in this cpu | 95 | * have in this cpu |
96 | */ | 96 | */ |
97 | static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type) | 97 | static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type) |
98 | { | 98 | { |
99 | int i; | 99 | int i; |
100 | unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); | 100 | unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); |
101 | 101 | ||
102 | for (i = nr_slots[type] - 1; i >= 0; i--) { | 102 | for (i = nr_slots[type] - 1; i >= 0; i--) { |
103 | if (tsk_pinned[i] > 0) | 103 | if (tsk_pinned[i] > 0) |
104 | return i + 1; | 104 | return i + 1; |
105 | } | 105 | } |
106 | 106 | ||
107 | return 0; | 107 | return 0; |
108 | } | 108 | } |
109 | 109 | ||
110 | /* | 110 | /* |
111 | * Count the number of breakpoints of the same type and same task. | 111 | * Count the number of breakpoints of the same type and same task. |
112 | * The given event must be not on the list. | 112 | * The given event must be not on the list. |
113 | */ | 113 | */ |
114 | static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type) | 114 | static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type) |
115 | { | 115 | { |
116 | struct perf_event_context *ctx = bp->ctx; | 116 | struct perf_event_context *ctx = bp->ctx; |
117 | struct perf_event *iter; | 117 | struct perf_event *iter; |
118 | int count = 0; | 118 | int count = 0; |
119 | 119 | ||
120 | list_for_each_entry(iter, &bp_task_head, hw.bp_list) { | 120 | list_for_each_entry(iter, &bp_task_head, hw.bp_list) { |
121 | if (iter->ctx == ctx && find_slot_idx(iter) == type) | 121 | if (iter->ctx == ctx && find_slot_idx(iter) == type) |
122 | count += hw_breakpoint_weight(iter); | 122 | count += hw_breakpoint_weight(iter); |
123 | } | 123 | } |
124 | 124 | ||
125 | return count; | 125 | return count; |
126 | } | 126 | } |
127 | 127 | ||
128 | /* | 128 | /* |
129 | * Report the number of pinned/un-pinned breakpoints we have in | 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). | 130 | * a given cpu (cpu > -1) or in all of them (cpu = -1). |
131 | */ | 131 | */ |
132 | static void | 132 | static void |
133 | fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, | 133 | fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, |
134 | enum bp_type_idx type) | 134 | enum bp_type_idx type) |
135 | { | 135 | { |
136 | int cpu = bp->cpu; | 136 | int cpu = bp->cpu; |
137 | struct task_struct *tsk = bp->ctx->task; | 137 | struct task_struct *tsk = bp->ctx->task; |
138 | 138 | ||
139 | if (cpu >= 0) { | 139 | if (cpu >= 0) { |
140 | slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu); | 140 | slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu); |
141 | if (!tsk) | 141 | if (!tsk) |
142 | slots->pinned += max_task_bp_pinned(cpu, type); | 142 | slots->pinned += max_task_bp_pinned(cpu, type); |
143 | else | 143 | else |
144 | slots->pinned += task_bp_pinned(bp, type); | 144 | slots->pinned += task_bp_pinned(bp, type); |
145 | slots->flexible = per_cpu(nr_bp_flexible[type], cpu); | 145 | slots->flexible = per_cpu(nr_bp_flexible[type], cpu); |
146 | 146 | ||
147 | return; | 147 | return; |
148 | } | 148 | } |
149 | 149 | ||
150 | for_each_online_cpu(cpu) { | 150 | for_each_online_cpu(cpu) { |
151 | unsigned int nr; | 151 | unsigned int nr; |
152 | 152 | ||
153 | nr = per_cpu(nr_cpu_bp_pinned[type], cpu); | 153 | nr = per_cpu(nr_cpu_bp_pinned[type], cpu); |
154 | if (!tsk) | 154 | if (!tsk) |
155 | nr += max_task_bp_pinned(cpu, type); | 155 | nr += max_task_bp_pinned(cpu, type); |
156 | else | 156 | else |
157 | nr += task_bp_pinned(bp, type); | 157 | nr += task_bp_pinned(bp, type); |
158 | 158 | ||
159 | if (nr > slots->pinned) | 159 | if (nr > slots->pinned) |
160 | slots->pinned = nr; | 160 | slots->pinned = nr; |
161 | 161 | ||
162 | nr = per_cpu(nr_bp_flexible[type], cpu); | 162 | nr = per_cpu(nr_bp_flexible[type], cpu); |
163 | 163 | ||
164 | if (nr > slots->flexible) | 164 | if (nr > slots->flexible) |
165 | slots->flexible = nr; | 165 | slots->flexible = nr; |
166 | } | 166 | } |
167 | } | 167 | } |
168 | 168 | ||
169 | /* | 169 | /* |
170 | * For now, continue to consider flexible as pinned, until we can | 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 | 171 | * ensure no flexible event can ever be scheduled before a pinned event |
172 | * in a same cpu. | 172 | * in a same cpu. |
173 | */ | 173 | */ |
174 | static void | 174 | static void |
175 | fetch_this_slot(struct bp_busy_slots *slots, int weight) | 175 | fetch_this_slot(struct bp_busy_slots *slots, int weight) |
176 | { | 176 | { |
177 | slots->pinned += weight; | 177 | slots->pinned += weight; |
178 | } | 178 | } |
179 | 179 | ||
180 | /* | 180 | /* |
181 | * Add a pinned breakpoint for the given task in our constraint table | 181 | * Add a pinned breakpoint for the given task in our constraint table |
182 | */ | 182 | */ |
183 | static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable, | 183 | static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable, |
184 | enum bp_type_idx type, int weight) | 184 | enum bp_type_idx type, int weight) |
185 | { | 185 | { |
186 | unsigned int *tsk_pinned; | 186 | unsigned int *tsk_pinned; |
187 | int old_count = 0; | 187 | int old_count = 0; |
188 | int old_idx = 0; | 188 | int old_idx = 0; |
189 | int idx = 0; | 189 | int idx = 0; |
190 | 190 | ||
191 | old_count = task_bp_pinned(bp, type); | 191 | old_count = task_bp_pinned(bp, type); |
192 | old_idx = old_count - 1; | 192 | old_idx = old_count - 1; |
193 | idx = old_idx + weight; | 193 | idx = old_idx + weight; |
194 | 194 | ||
195 | /* tsk_pinned[n] is the number of tasks having n breakpoints */ | 195 | /* tsk_pinned[n] is the number of tasks having n breakpoints */ |
196 | tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); | 196 | tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); |
197 | if (enable) { | 197 | if (enable) { |
198 | tsk_pinned[idx]++; | 198 | tsk_pinned[idx]++; |
199 | if (old_count > 0) | 199 | if (old_count > 0) |
200 | tsk_pinned[old_idx]--; | 200 | tsk_pinned[old_idx]--; |
201 | } else { | 201 | } else { |
202 | tsk_pinned[idx]--; | 202 | tsk_pinned[idx]--; |
203 | if (old_count > 0) | 203 | if (old_count > 0) |
204 | tsk_pinned[old_idx]++; | 204 | tsk_pinned[old_idx]++; |
205 | } | 205 | } |
206 | } | 206 | } |
207 | 207 | ||
208 | /* | 208 | /* |
209 | * Add/remove the given breakpoint in our constraint table | 209 | * Add/remove the given breakpoint in our constraint table |
210 | */ | 210 | */ |
211 | static void | 211 | static void |
212 | toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type, | 212 | toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type, |
213 | int weight) | 213 | int weight) |
214 | { | 214 | { |
215 | int cpu = bp->cpu; | 215 | int cpu = bp->cpu; |
216 | struct task_struct *tsk = bp->ctx->task; | 216 | struct task_struct *tsk = bp->ctx->task; |
217 | 217 | ||
218 | /* Pinned counter cpu profiling */ | 218 | /* Pinned counter cpu profiling */ |
219 | if (!tsk) { | 219 | if (!tsk) { |
220 | 220 | ||
221 | if (enable) | 221 | if (enable) |
222 | per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight; | 222 | per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight; |
223 | else | 223 | else |
224 | per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight; | 224 | per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight; |
225 | return; | 225 | return; |
226 | } | 226 | } |
227 | 227 | ||
228 | /* Pinned counter task profiling */ | 228 | /* Pinned counter task profiling */ |
229 | 229 | ||
230 | if (!enable) | 230 | if (!enable) |
231 | list_del(&bp->hw.bp_list); | 231 | list_del(&bp->hw.bp_list); |
232 | 232 | ||
233 | if (cpu >= 0) { | 233 | if (cpu >= 0) { |
234 | toggle_bp_task_slot(bp, cpu, enable, type, weight); | 234 | toggle_bp_task_slot(bp, cpu, enable, type, weight); |
235 | } else { | 235 | } else { |
236 | for_each_online_cpu(cpu) | 236 | for_each_online_cpu(cpu) |
237 | toggle_bp_task_slot(bp, cpu, enable, type, weight); | 237 | toggle_bp_task_slot(bp, cpu, enable, type, weight); |
238 | } | 238 | } |
239 | 239 | ||
240 | if (enable) | 240 | if (enable) |
241 | list_add_tail(&bp->hw.bp_list, &bp_task_head); | 241 | list_add_tail(&bp->hw.bp_list, &bp_task_head); |
242 | } | 242 | } |
243 | 243 | ||
244 | /* | 244 | /* |
245 | * Function to perform processor-specific cleanup during unregistration | 245 | * Function to perform processor-specific cleanup during unregistration |
246 | */ | 246 | */ |
247 | __weak void arch_unregister_hw_breakpoint(struct perf_event *bp) | 247 | __weak void arch_unregister_hw_breakpoint(struct perf_event *bp) |
248 | { | 248 | { |
249 | /* | 249 | /* |
250 | * A weak stub function here for those archs that don't define | 250 | * A weak stub function here for those archs that don't define |
251 | * it inside arch/.../kernel/hw_breakpoint.c | 251 | * it inside arch/.../kernel/hw_breakpoint.c |
252 | */ | 252 | */ |
253 | } | 253 | } |
254 | 254 | ||
255 | /* | 255 | /* |
256 | * Contraints to check before allowing this new breakpoint counter: | 256 | * Contraints to check before allowing this new breakpoint counter: |
257 | * | 257 | * |
258 | * == Non-pinned counter == (Considered as pinned for now) | 258 | * == Non-pinned counter == (Considered as pinned for now) |
259 | * | 259 | * |
260 | * - If attached to a single cpu, check: | 260 | * - If attached to a single cpu, check: |
261 | * | 261 | * |
262 | * (per_cpu(nr_bp_flexible, cpu) || (per_cpu(nr_cpu_bp_pinned, cpu) | 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 | 263 | * + max(per_cpu(nr_task_bp_pinned, cpu)))) < HBP_NUM |
264 | * | 264 | * |
265 | * -> If there are already non-pinned counters in this cpu, it means | 265 | * -> If there are already non-pinned counters in this cpu, it means |
266 | * there is already a free slot for them. | 266 | * there is already a free slot for them. |
267 | * Otherwise, we check that the maximum number of per task | 267 | * Otherwise, we check that the maximum number of per task |
268 | * breakpoints (for this cpu) plus the number of per cpu breakpoint | 268 | * breakpoints (for this cpu) plus the number of per cpu breakpoint |
269 | * (for this cpu) doesn't cover every registers. | 269 | * (for this cpu) doesn't cover every registers. |
270 | * | 270 | * |
271 | * - If attached to every cpus, check: | 271 | * - If attached to every cpus, check: |
272 | * | 272 | * |
273 | * (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *)) | 273 | * (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *)) |
274 | * + max(per_cpu(nr_task_bp_pinned, *)))) < HBP_NUM | 274 | * + max(per_cpu(nr_task_bp_pinned, *)))) < HBP_NUM |
275 | * | 275 | * |
276 | * -> This is roughly the same, except we check the number of per cpu | 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 | 277 | * bp for every cpu and we keep the max one. Same for the per tasks |
278 | * breakpoints. | 278 | * breakpoints. |
279 | * | 279 | * |
280 | * | 280 | * |
281 | * == Pinned counter == | 281 | * == Pinned counter == |
282 | * | 282 | * |
283 | * - If attached to a single cpu, check: | 283 | * - If attached to a single cpu, check: |
284 | * | 284 | * |
285 | * ((per_cpu(nr_bp_flexible, cpu) > 1) + per_cpu(nr_cpu_bp_pinned, cpu) | 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 | 286 | * + max(per_cpu(nr_task_bp_pinned, cpu))) < HBP_NUM |
287 | * | 287 | * |
288 | * -> Same checks as before. But now the nr_bp_flexible, if any, must keep | 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). | 289 | * one register at least (or they will never be fed). |
290 | * | 290 | * |
291 | * - If attached to every cpus, check: | 291 | * - If attached to every cpus, check: |
292 | * | 292 | * |
293 | * ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *)) | 293 | * ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *)) |
294 | * + max(per_cpu(nr_task_bp_pinned, *))) < HBP_NUM | 294 | * + max(per_cpu(nr_task_bp_pinned, *))) < HBP_NUM |
295 | */ | 295 | */ |
296 | static int __reserve_bp_slot(struct perf_event *bp) | 296 | static int __reserve_bp_slot(struct perf_event *bp) |
297 | { | 297 | { |
298 | struct bp_busy_slots slots = {0}; | 298 | struct bp_busy_slots slots = {0}; |
299 | enum bp_type_idx type; | 299 | enum bp_type_idx type; |
300 | int weight; | 300 | int weight; |
301 | 301 | ||
302 | /* We couldn't initialize breakpoint constraints on boot */ | 302 | /* We couldn't initialize breakpoint constraints on boot */ |
303 | if (!constraints_initialized) | 303 | if (!constraints_initialized) |
304 | return -ENOMEM; | 304 | return -ENOMEM; |
305 | 305 | ||
306 | /* Basic checks */ | 306 | /* Basic checks */ |
307 | if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY || | 307 | if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY || |
308 | bp->attr.bp_type == HW_BREAKPOINT_INVALID) | 308 | bp->attr.bp_type == HW_BREAKPOINT_INVALID) |
309 | return -EINVAL; | 309 | return -EINVAL; |
310 | 310 | ||
311 | type = find_slot_idx(bp); | 311 | type = find_slot_idx(bp); |
312 | weight = hw_breakpoint_weight(bp); | 312 | weight = hw_breakpoint_weight(bp); |
313 | 313 | ||
314 | fetch_bp_busy_slots(&slots, bp, type); | 314 | fetch_bp_busy_slots(&slots, bp, type); |
315 | /* | 315 | /* |
316 | * Simulate the addition of this breakpoint to the constraints | 316 | * Simulate the addition of this breakpoint to the constraints |
317 | * and see the result. | 317 | * and see the result. |
318 | */ | 318 | */ |
319 | fetch_this_slot(&slots, weight); | 319 | fetch_this_slot(&slots, weight); |
320 | 320 | ||
321 | /* Flexible counters need to keep at least one slot */ | 321 | /* Flexible counters need to keep at least one slot */ |
322 | if (slots.pinned + (!!slots.flexible) > nr_slots[type]) | 322 | if (slots.pinned + (!!slots.flexible) > nr_slots[type]) |
323 | return -ENOSPC; | 323 | return -ENOSPC; |
324 | 324 | ||
325 | toggle_bp_slot(bp, true, type, weight); | 325 | toggle_bp_slot(bp, true, type, weight); |
326 | 326 | ||
327 | return 0; | 327 | return 0; |
328 | } | 328 | } |
329 | 329 | ||
330 | int reserve_bp_slot(struct perf_event *bp) | 330 | int reserve_bp_slot(struct perf_event *bp) |
331 | { | 331 | { |
332 | int ret; | 332 | int ret; |
333 | 333 | ||
334 | mutex_lock(&nr_bp_mutex); | 334 | mutex_lock(&nr_bp_mutex); |
335 | 335 | ||
336 | ret = __reserve_bp_slot(bp); | 336 | ret = __reserve_bp_slot(bp); |
337 | 337 | ||
338 | mutex_unlock(&nr_bp_mutex); | 338 | mutex_unlock(&nr_bp_mutex); |
339 | 339 | ||
340 | return ret; | 340 | return ret; |
341 | } | 341 | } |
342 | 342 | ||
343 | static void __release_bp_slot(struct perf_event *bp) | 343 | static void __release_bp_slot(struct perf_event *bp) |
344 | { | 344 | { |
345 | enum bp_type_idx type; | 345 | enum bp_type_idx type; |
346 | int weight; | 346 | int weight; |
347 | 347 | ||
348 | type = find_slot_idx(bp); | 348 | type = find_slot_idx(bp); |
349 | weight = hw_breakpoint_weight(bp); | 349 | weight = hw_breakpoint_weight(bp); |
350 | toggle_bp_slot(bp, false, type, weight); | 350 | toggle_bp_slot(bp, false, type, weight); |
351 | } | 351 | } |
352 | 352 | ||
353 | void release_bp_slot(struct perf_event *bp) | 353 | void release_bp_slot(struct perf_event *bp) |
354 | { | 354 | { |
355 | mutex_lock(&nr_bp_mutex); | 355 | mutex_lock(&nr_bp_mutex); |
356 | 356 | ||
357 | arch_unregister_hw_breakpoint(bp); | 357 | arch_unregister_hw_breakpoint(bp); |
358 | __release_bp_slot(bp); | 358 | __release_bp_slot(bp); |
359 | 359 | ||
360 | mutex_unlock(&nr_bp_mutex); | 360 | mutex_unlock(&nr_bp_mutex); |
361 | } | 361 | } |
362 | 362 | ||
363 | /* | 363 | /* |
364 | * Allow the kernel debugger to reserve breakpoint slots without | 364 | * Allow the kernel debugger to reserve breakpoint slots without |
365 | * taking a lock using the dbg_* variant of for the reserve and | 365 | * taking a lock using the dbg_* variant of for the reserve and |
366 | * release breakpoint slots. | 366 | * release breakpoint slots. |
367 | */ | 367 | */ |
368 | int dbg_reserve_bp_slot(struct perf_event *bp) | 368 | int dbg_reserve_bp_slot(struct perf_event *bp) |
369 | { | 369 | { |
370 | if (mutex_is_locked(&nr_bp_mutex)) | 370 | if (mutex_is_locked(&nr_bp_mutex)) |
371 | return -1; | 371 | return -1; |
372 | 372 | ||
373 | return __reserve_bp_slot(bp); | 373 | return __reserve_bp_slot(bp); |
374 | } | 374 | } |
375 | 375 | ||
376 | int dbg_release_bp_slot(struct perf_event *bp) | 376 | int dbg_release_bp_slot(struct perf_event *bp) |
377 | { | 377 | { |
378 | if (mutex_is_locked(&nr_bp_mutex)) | 378 | if (mutex_is_locked(&nr_bp_mutex)) |
379 | return -1; | 379 | return -1; |
380 | 380 | ||
381 | __release_bp_slot(bp); | 381 | __release_bp_slot(bp); |
382 | 382 | ||
383 | return 0; | 383 | return 0; |
384 | } | 384 | } |
385 | 385 | ||
386 | static int validate_hw_breakpoint(struct perf_event *bp) | 386 | static int validate_hw_breakpoint(struct perf_event *bp) |
387 | { | 387 | { |
388 | int ret; | 388 | int ret; |
389 | 389 | ||
390 | ret = arch_validate_hwbkpt_settings(bp); | 390 | ret = arch_validate_hwbkpt_settings(bp); |
391 | if (ret) | 391 | if (ret) |
392 | return ret; | 392 | return ret; |
393 | 393 | ||
394 | if (arch_check_bp_in_kernelspace(bp)) { | 394 | if (arch_check_bp_in_kernelspace(bp)) { |
395 | if (bp->attr.exclude_kernel) | 395 | if (bp->attr.exclude_kernel) |
396 | return -EINVAL; | 396 | return -EINVAL; |
397 | /* | 397 | /* |
398 | * Don't let unprivileged users set a breakpoint in the trap | 398 | * Don't let unprivileged users set a breakpoint in the trap |
399 | * path to avoid trap recursion attacks. | 399 | * path to avoid trap recursion attacks. |
400 | */ | 400 | */ |
401 | if (!capable(CAP_SYS_ADMIN)) | 401 | if (!capable(CAP_SYS_ADMIN)) |
402 | return -EPERM; | 402 | return -EPERM; |
403 | } | 403 | } |
404 | 404 | ||
405 | return 0; | 405 | return 0; |
406 | } | 406 | } |
407 | 407 | ||
408 | int register_perf_hw_breakpoint(struct perf_event *bp) | 408 | int register_perf_hw_breakpoint(struct perf_event *bp) |
409 | { | 409 | { |
410 | int ret; | 410 | int ret; |
411 | 411 | ||
412 | ret = reserve_bp_slot(bp); | 412 | ret = reserve_bp_slot(bp); |
413 | if (ret) | 413 | if (ret) |
414 | return ret; | 414 | return ret; |
415 | 415 | ||
416 | ret = validate_hw_breakpoint(bp); | 416 | ret = validate_hw_breakpoint(bp); |
417 | 417 | ||
418 | /* if arch_validate_hwbkpt_settings() fails then release bp slot */ | 418 | /* if arch_validate_hwbkpt_settings() fails then release bp slot */ |
419 | if (ret) | 419 | if (ret) |
420 | release_bp_slot(bp); | 420 | release_bp_slot(bp); |
421 | 421 | ||
422 | return ret; | 422 | return ret; |
423 | } | 423 | } |
424 | 424 | ||
425 | /** | 425 | /** |
426 | * register_user_hw_breakpoint - register a hardware breakpoint for user space | 426 | * register_user_hw_breakpoint - register a hardware breakpoint for user space |
427 | * @attr: breakpoint attributes | 427 | * @attr: breakpoint attributes |
428 | * @triggered: callback to trigger when we hit the breakpoint | 428 | * @triggered: callback to trigger when we hit the breakpoint |
429 | * @tsk: pointer to 'task_struct' of the process to which the address belongs | 429 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
430 | */ | 430 | */ |
431 | struct perf_event * | 431 | struct perf_event * |
432 | register_user_hw_breakpoint(struct perf_event_attr *attr, | 432 | register_user_hw_breakpoint(struct perf_event_attr *attr, |
433 | perf_overflow_handler_t triggered, | 433 | perf_overflow_handler_t triggered, |
434 | struct task_struct *tsk) | 434 | struct task_struct *tsk) |
435 | { | 435 | { |
436 | return perf_event_create_kernel_counter(attr, -1, tsk->pid, triggered); | 436 | return perf_event_create_kernel_counter(attr, -1, task_pid_vnr(tsk), |
437 | triggered); | ||
437 | } | 438 | } |
438 | EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); | 439 | EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); |
439 | 440 | ||
440 | /** | 441 | /** |
441 | * modify_user_hw_breakpoint - modify a user-space hardware breakpoint | 442 | * modify_user_hw_breakpoint - modify a user-space hardware breakpoint |
442 | * @bp: the breakpoint structure to modify | 443 | * @bp: the breakpoint structure to modify |
443 | * @attr: new breakpoint attributes | 444 | * @attr: new breakpoint attributes |
444 | * @triggered: callback to trigger when we hit the breakpoint | 445 | * @triggered: callback to trigger when we hit the breakpoint |
445 | * @tsk: pointer to 'task_struct' of the process to which the address belongs | 446 | * @tsk: pointer to 'task_struct' of the process to which the address belongs |
446 | */ | 447 | */ |
447 | int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) | 448 | int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr) |
448 | { | 449 | { |
449 | u64 old_addr = bp->attr.bp_addr; | 450 | u64 old_addr = bp->attr.bp_addr; |
450 | u64 old_len = bp->attr.bp_len; | 451 | u64 old_len = bp->attr.bp_len; |
451 | int old_type = bp->attr.bp_type; | 452 | int old_type = bp->attr.bp_type; |
452 | int err = 0; | 453 | int err = 0; |
453 | 454 | ||
454 | perf_event_disable(bp); | 455 | perf_event_disable(bp); |
455 | 456 | ||
456 | bp->attr.bp_addr = attr->bp_addr; | 457 | bp->attr.bp_addr = attr->bp_addr; |
457 | bp->attr.bp_type = attr->bp_type; | 458 | bp->attr.bp_type = attr->bp_type; |
458 | bp->attr.bp_len = attr->bp_len; | 459 | bp->attr.bp_len = attr->bp_len; |
459 | 460 | ||
460 | if (attr->disabled) | 461 | if (attr->disabled) |
461 | goto end; | 462 | goto end; |
462 | 463 | ||
463 | err = validate_hw_breakpoint(bp); | 464 | err = validate_hw_breakpoint(bp); |
464 | if (!err) | 465 | if (!err) |
465 | perf_event_enable(bp); | 466 | perf_event_enable(bp); |
466 | 467 | ||
467 | if (err) { | 468 | if (err) { |
468 | bp->attr.bp_addr = old_addr; | 469 | bp->attr.bp_addr = old_addr; |
469 | bp->attr.bp_type = old_type; | 470 | bp->attr.bp_type = old_type; |
470 | bp->attr.bp_len = old_len; | 471 | bp->attr.bp_len = old_len; |
471 | if (!bp->attr.disabled) | 472 | if (!bp->attr.disabled) |
472 | perf_event_enable(bp); | 473 | perf_event_enable(bp); |
473 | 474 | ||
474 | return err; | 475 | return err; |
475 | } | 476 | } |
476 | 477 | ||
477 | end: | 478 | end: |
478 | bp->attr.disabled = attr->disabled; | 479 | bp->attr.disabled = attr->disabled; |
479 | 480 | ||
480 | return 0; | 481 | return 0; |
481 | } | 482 | } |
482 | EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); | 483 | EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint); |
483 | 484 | ||
484 | /** | 485 | /** |
485 | * unregister_hw_breakpoint - unregister a user-space hardware breakpoint | 486 | * unregister_hw_breakpoint - unregister a user-space hardware breakpoint |
486 | * @bp: the breakpoint structure to unregister | 487 | * @bp: the breakpoint structure to unregister |
487 | */ | 488 | */ |
488 | void unregister_hw_breakpoint(struct perf_event *bp) | 489 | void unregister_hw_breakpoint(struct perf_event *bp) |
489 | { | 490 | { |
490 | if (!bp) | 491 | if (!bp) |
491 | return; | 492 | return; |
492 | perf_event_release_kernel(bp); | 493 | perf_event_release_kernel(bp); |
493 | } | 494 | } |
494 | EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); | 495 | EXPORT_SYMBOL_GPL(unregister_hw_breakpoint); |
495 | 496 | ||
496 | /** | 497 | /** |
497 | * register_wide_hw_breakpoint - register a wide breakpoint in the kernel | 498 | * register_wide_hw_breakpoint - register a wide breakpoint in the kernel |
498 | * @attr: breakpoint attributes | 499 | * @attr: breakpoint attributes |
499 | * @triggered: callback to trigger when we hit the breakpoint | 500 | * @triggered: callback to trigger when we hit the breakpoint |
500 | * | 501 | * |
501 | * @return a set of per_cpu pointers to perf events | 502 | * @return a set of per_cpu pointers to perf events |
502 | */ | 503 | */ |
503 | struct perf_event * __percpu * | 504 | struct perf_event * __percpu * |
504 | register_wide_hw_breakpoint(struct perf_event_attr *attr, | 505 | register_wide_hw_breakpoint(struct perf_event_attr *attr, |
505 | perf_overflow_handler_t triggered) | 506 | perf_overflow_handler_t triggered) |
506 | { | 507 | { |
507 | struct perf_event * __percpu *cpu_events, **pevent, *bp; | 508 | struct perf_event * __percpu *cpu_events, **pevent, *bp; |
508 | long err; | 509 | long err; |
509 | int cpu; | 510 | int cpu; |
510 | 511 | ||
511 | cpu_events = alloc_percpu(typeof(*cpu_events)); | 512 | cpu_events = alloc_percpu(typeof(*cpu_events)); |
512 | if (!cpu_events) | 513 | if (!cpu_events) |
513 | return (void __percpu __force *)ERR_PTR(-ENOMEM); | 514 | return (void __percpu __force *)ERR_PTR(-ENOMEM); |
514 | 515 | ||
515 | get_online_cpus(); | 516 | get_online_cpus(); |
516 | for_each_online_cpu(cpu) { | 517 | for_each_online_cpu(cpu) { |
517 | pevent = per_cpu_ptr(cpu_events, cpu); | 518 | pevent = per_cpu_ptr(cpu_events, cpu); |
518 | bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered); | 519 | bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered); |
519 | 520 | ||
520 | *pevent = bp; | 521 | *pevent = bp; |
521 | 522 | ||
522 | if (IS_ERR(bp)) { | 523 | if (IS_ERR(bp)) { |
523 | err = PTR_ERR(bp); | 524 | err = PTR_ERR(bp); |
524 | goto fail; | 525 | goto fail; |
525 | } | 526 | } |
526 | } | 527 | } |
527 | put_online_cpus(); | 528 | put_online_cpus(); |
528 | 529 | ||
529 | return cpu_events; | 530 | return cpu_events; |
530 | 531 | ||
531 | fail: | 532 | fail: |
532 | for_each_online_cpu(cpu) { | 533 | for_each_online_cpu(cpu) { |
533 | pevent = per_cpu_ptr(cpu_events, cpu); | 534 | pevent = per_cpu_ptr(cpu_events, cpu); |
534 | if (IS_ERR(*pevent)) | 535 | if (IS_ERR(*pevent)) |
535 | break; | 536 | break; |
536 | unregister_hw_breakpoint(*pevent); | 537 | unregister_hw_breakpoint(*pevent); |
537 | } | 538 | } |
538 | put_online_cpus(); | 539 | put_online_cpus(); |
539 | 540 | ||
540 | free_percpu(cpu_events); | 541 | free_percpu(cpu_events); |
541 | return (void __percpu __force *)ERR_PTR(err); | 542 | return (void __percpu __force *)ERR_PTR(err); |
542 | } | 543 | } |
543 | EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint); | 544 | EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint); |
544 | 545 | ||
545 | /** | 546 | /** |
546 | * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel | 547 | * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel |
547 | * @cpu_events: the per cpu set of events to unregister | 548 | * @cpu_events: the per cpu set of events to unregister |
548 | */ | 549 | */ |
549 | void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events) | 550 | void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events) |
550 | { | 551 | { |
551 | int cpu; | 552 | int cpu; |
552 | struct perf_event **pevent; | 553 | struct perf_event **pevent; |
553 | 554 | ||
554 | for_each_possible_cpu(cpu) { | 555 | for_each_possible_cpu(cpu) { |
555 | pevent = per_cpu_ptr(cpu_events, cpu); | 556 | pevent = per_cpu_ptr(cpu_events, cpu); |
556 | unregister_hw_breakpoint(*pevent); | 557 | unregister_hw_breakpoint(*pevent); |
557 | } | 558 | } |
558 | free_percpu(cpu_events); | 559 | free_percpu(cpu_events); |
559 | } | 560 | } |
560 | EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint); | 561 | EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint); |
561 | 562 | ||
562 | static struct notifier_block hw_breakpoint_exceptions_nb = { | 563 | static struct notifier_block hw_breakpoint_exceptions_nb = { |
563 | .notifier_call = hw_breakpoint_exceptions_notify, | 564 | .notifier_call = hw_breakpoint_exceptions_notify, |
564 | /* we need to be notified first */ | 565 | /* we need to be notified first */ |
565 | .priority = 0x7fffffff | 566 | .priority = 0x7fffffff |
566 | }; | 567 | }; |
567 | 568 | ||
568 | static int __init init_hw_breakpoint(void) | 569 | static int __init init_hw_breakpoint(void) |
569 | { | 570 | { |
570 | unsigned int **task_bp_pinned; | 571 | unsigned int **task_bp_pinned; |
571 | int cpu, err_cpu; | 572 | int cpu, err_cpu; |
572 | int i; | 573 | int i; |
573 | 574 | ||
574 | for (i = 0; i < TYPE_MAX; i++) | 575 | for (i = 0; i < TYPE_MAX; i++) |
575 | nr_slots[i] = hw_breakpoint_slots(i); | 576 | nr_slots[i] = hw_breakpoint_slots(i); |
576 | 577 | ||
577 | for_each_possible_cpu(cpu) { | 578 | for_each_possible_cpu(cpu) { |
578 | for (i = 0; i < TYPE_MAX; i++) { | 579 | for (i = 0; i < TYPE_MAX; i++) { |
579 | task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu); | 580 | task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu); |
580 | *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i], | 581 | *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i], |
581 | GFP_KERNEL); | 582 | GFP_KERNEL); |
582 | if (!*task_bp_pinned) | 583 | if (!*task_bp_pinned) |
583 | goto err_alloc; | 584 | goto err_alloc; |
584 | } | 585 | } |
585 | } | 586 | } |
586 | 587 | ||
587 | constraints_initialized = 1; | 588 | constraints_initialized = 1; |
588 | 589 | ||
589 | return register_die_notifier(&hw_breakpoint_exceptions_nb); | 590 | return register_die_notifier(&hw_breakpoint_exceptions_nb); |
590 | 591 | ||
591 | err_alloc: | 592 | err_alloc: |
592 | for_each_possible_cpu(err_cpu) { | 593 | for_each_possible_cpu(err_cpu) { |
593 | if (err_cpu == cpu) | 594 | if (err_cpu == cpu) |
594 | break; | 595 | break; |
595 | for (i = 0; i < TYPE_MAX; i++) | 596 | for (i = 0; i < TYPE_MAX; i++) |
596 | kfree(per_cpu(nr_task_bp_pinned[i], cpu)); | 597 | kfree(per_cpu(nr_task_bp_pinned[i], cpu)); |
597 | } | 598 | } |
598 | 599 | ||
599 | return -ENOMEM; | 600 | return -ENOMEM; |
600 | } | 601 | } |
601 | core_initcall(init_hw_breakpoint); | 602 | core_initcall(init_hw_breakpoint); |
602 | 603 | ||
603 | 604 | ||
604 | struct pmu perf_ops_bp = { | 605 | struct pmu perf_ops_bp = { |
605 | .enable = arch_install_hw_breakpoint, | 606 | .enable = arch_install_hw_breakpoint, |
606 | .disable = arch_uninstall_hw_breakpoint, | 607 | .disable = arch_uninstall_hw_breakpoint, |
607 | .read = hw_breakpoint_pmu_read, | 608 | .read = hw_breakpoint_pmu_read, |
608 | }; | 609 | }; |
609 | 610 |