K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 1 | /* |
| 2 | * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility, |
| 3 | * using the CPU's debug registers. Derived from |
| 4 | * "arch/x86/kernel/hw_breakpoint.c" |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
| 19 | * |
| 20 | * Copyright 2010 IBM Corporation |
| 21 | * Author: K.Prasad <prasad@linux.vnet.ibm.com> |
| 22 | * |
| 23 | */ |
| 24 | |
| 25 | #include <linux/hw_breakpoint.h> |
| 26 | #include <linux/notifier.h> |
| 27 | #include <linux/kprobes.h> |
| 28 | #include <linux/percpu.h> |
| 29 | #include <linux/kernel.h> |
| 30 | #include <linux/module.h> |
| 31 | #include <linux/sched.h> |
| 32 | #include <linux/init.h> |
| 33 | #include <linux/smp.h> |
| 34 | |
| 35 | #include <asm/hw_breakpoint.h> |
| 36 | #include <asm/processor.h> |
| 37 | #include <asm/sstep.h> |
| 38 | #include <asm/uaccess.h> |
| 39 | |
| 40 | /* |
| 41 | * Stores the breakpoints currently in use on each breakpoint address |
| 42 | * register for every cpu |
| 43 | */ |
| 44 | static DEFINE_PER_CPU(struct perf_event *, bp_per_reg); |
| 45 | |
| 46 | /* |
| 47 | * Install a perf counter breakpoint. |
| 48 | * |
| 49 | * We seek a free debug address register and use it for this |
| 50 | * breakpoint. |
| 51 | * |
| 52 | * Atomic: we hold the counter->ctx->lock and we only handle variables |
| 53 | * and registers local to this cpu. |
| 54 | */ |
| 55 | int arch_install_hw_breakpoint(struct perf_event *bp) |
| 56 | { |
| 57 | struct arch_hw_breakpoint *info = counter_arch_bp(bp); |
| 58 | struct perf_event **slot = &__get_cpu_var(bp_per_reg); |
| 59 | |
| 60 | *slot = bp; |
| 61 | |
| 62 | /* |
| 63 | * Do not install DABR values if the instruction must be single-stepped. |
| 64 | * If so, DABR will be populated in single_step_dabr_instruction(). |
| 65 | */ |
| 66 | if (current->thread.last_hit_ubp != bp) |
| 67 | set_dabr(info->address | info->type | DABR_TRANSLATION); |
| 68 | |
| 69 | return 0; |
| 70 | } |
| 71 | |
| 72 | /* |
| 73 | * Uninstall the breakpoint contained in the given counter. |
| 74 | * |
| 75 | * First we search the debug address register it uses and then we disable |
| 76 | * it. |
| 77 | * |
| 78 | * Atomic: we hold the counter->ctx->lock and we only handle variables |
| 79 | * and registers local to this cpu. |
| 80 | */ |
| 81 | void arch_uninstall_hw_breakpoint(struct perf_event *bp) |
| 82 | { |
| 83 | struct perf_event **slot = &__get_cpu_var(bp_per_reg); |
| 84 | |
| 85 | if (*slot != bp) { |
| 86 | WARN_ONCE(1, "Can't find the breakpoint"); |
| 87 | return; |
| 88 | } |
| 89 | |
| 90 | *slot = NULL; |
| 91 | set_dabr(0); |
| 92 | } |
| 93 | |
| 94 | /* |
| 95 | * Perform cleanup of arch-specific counters during unregistration |
| 96 | * of the perf-event |
| 97 | */ |
| 98 | void arch_unregister_hw_breakpoint(struct perf_event *bp) |
| 99 | { |
| 100 | /* |
| 101 | * If the breakpoint is unregistered between a hw_breakpoint_handler() |
| 102 | * and the single_step_dabr_instruction(), then cleanup the breakpoint |
| 103 | * restoration variables to prevent dangling pointers. |
| 104 | */ |
| 105 | if (bp->ctx->task) |
| 106 | bp->ctx->task->thread.last_hit_ubp = NULL; |
| 107 | } |
| 108 | |
| 109 | /* |
| 110 | * Check for virtual address in kernel space. |
| 111 | */ |
| 112 | int arch_check_bp_in_kernelspace(struct perf_event *bp) |
| 113 | { |
| 114 | struct arch_hw_breakpoint *info = counter_arch_bp(bp); |
| 115 | |
| 116 | return is_kernel_addr(info->address); |
| 117 | } |
| 118 | |
| 119 | int arch_bp_generic_fields(int type, int *gen_bp_type) |
| 120 | { |
| 121 | switch (type) { |
| 122 | case DABR_DATA_READ: |
| 123 | *gen_bp_type = HW_BREAKPOINT_R; |
| 124 | break; |
| 125 | case DABR_DATA_WRITE: |
| 126 | *gen_bp_type = HW_BREAKPOINT_W; |
| 127 | break; |
| 128 | case (DABR_DATA_WRITE | DABR_DATA_READ): |
| 129 | *gen_bp_type = (HW_BREAKPOINT_W | HW_BREAKPOINT_R); |
| 130 | break; |
| 131 | default: |
| 132 | return -EINVAL; |
| 133 | } |
| 134 | return 0; |
| 135 | } |
| 136 | |
| 137 | /* |
| 138 | * Validate the arch-specific HW Breakpoint register settings |
| 139 | */ |
| 140 | int arch_validate_hwbkpt_settings(struct perf_event *bp) |
| 141 | { |
| 142 | int ret = -EINVAL; |
| 143 | struct arch_hw_breakpoint *info = counter_arch_bp(bp); |
| 144 | |
| 145 | if (!bp) |
| 146 | return ret; |
| 147 | |
| 148 | switch (bp->attr.bp_type) { |
| 149 | case HW_BREAKPOINT_R: |
| 150 | info->type = DABR_DATA_READ; |
| 151 | break; |
| 152 | case HW_BREAKPOINT_W: |
| 153 | info->type = DABR_DATA_WRITE; |
| 154 | break; |
| 155 | case HW_BREAKPOINT_R | HW_BREAKPOINT_W: |
| 156 | info->type = (DABR_DATA_READ | DABR_DATA_WRITE); |
| 157 | break; |
| 158 | default: |
| 159 | return ret; |
| 160 | } |
| 161 | |
| 162 | info->address = bp->attr.bp_addr; |
| 163 | info->len = bp->attr.bp_len; |
| 164 | |
| 165 | /* |
| 166 | * Since breakpoint length can be a maximum of HW_BREAKPOINT_LEN(8) |
| 167 | * and breakpoint addresses are aligned to nearest double-word |
| 168 | * HW_BREAKPOINT_ALIGN by rounding off to the lower address, the |
| 169 | * 'symbolsize' should satisfy the check below. |
| 170 | */ |
| 171 | if (info->len > |
| 172 | (HW_BREAKPOINT_LEN - (info->address & HW_BREAKPOINT_ALIGN))) |
| 173 | return -EINVAL; |
| 174 | return 0; |
| 175 | } |
| 176 | |
| 177 | /* |
K.Prasad | 06532a6 | 2010-06-15 11:35:41 +0530 | [diff] [blame] | 178 | * Restores the breakpoint on the debug registers. |
| 179 | * Invoke this function if it is known that the execution context is |
| 180 | * about to change to cause loss of MSR_SE settings. |
| 181 | */ |
| 182 | void thread_change_pc(struct task_struct *tsk, struct pt_regs *regs) |
| 183 | { |
| 184 | struct arch_hw_breakpoint *info; |
| 185 | |
| 186 | if (likely(!tsk->thread.last_hit_ubp)) |
| 187 | return; |
| 188 | |
| 189 | info = counter_arch_bp(tsk->thread.last_hit_ubp); |
| 190 | regs->msr &= ~MSR_SE; |
| 191 | set_dabr(info->address | info->type | DABR_TRANSLATION); |
| 192 | tsk->thread.last_hit_ubp = NULL; |
| 193 | } |
| 194 | |
| 195 | /* |
K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 196 | * Handle debug exception notifications. |
| 197 | */ |
| 198 | int __kprobes hw_breakpoint_handler(struct die_args *args) |
| 199 | { |
| 200 | bool is_ptrace_bp = false; |
| 201 | int rc = NOTIFY_STOP; |
| 202 | struct perf_event *bp; |
| 203 | struct pt_regs *regs = args->regs; |
| 204 | int stepped = 1; |
| 205 | struct arch_hw_breakpoint *info; |
| 206 | unsigned int instr; |
K.Prasad | e3e9408 | 2010-06-15 11:36:12 +0530 | [diff] [blame^] | 207 | unsigned long dar = regs->dar; |
K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 208 | |
| 209 | /* Disable breakpoints during exception handling */ |
| 210 | set_dabr(0); |
| 211 | /* |
| 212 | * The counter may be concurrently released but that can only |
| 213 | * occur from a call_rcu() path. We can then safely fetch |
| 214 | * the breakpoint, use its callback, touch its counter |
| 215 | * while we are in an rcu_read_lock() path. |
| 216 | */ |
| 217 | rcu_read_lock(); |
| 218 | |
| 219 | bp = __get_cpu_var(bp_per_reg); |
| 220 | if (!bp) |
| 221 | goto out; |
| 222 | info = counter_arch_bp(bp); |
| 223 | is_ptrace_bp = (bp->overflow_handler == ptrace_triggered) ? |
| 224 | true : false; |
| 225 | |
| 226 | /* |
| 227 | * Return early after invoking user-callback function without restoring |
| 228 | * DABR if the breakpoint is from ptrace which always operates in |
| 229 | * one-shot mode. The ptrace-ed process will receive the SIGTRAP signal |
| 230 | * generated in do_dabr(). |
| 231 | */ |
| 232 | if (is_ptrace_bp) { |
| 233 | perf_bp_event(bp, regs); |
| 234 | rc = NOTIFY_DONE; |
| 235 | goto out; |
| 236 | } |
| 237 | |
K.Prasad | e3e9408 | 2010-06-15 11:36:12 +0530 | [diff] [blame^] | 238 | /* |
| 239 | * Verify if dar lies within the address range occupied by the symbol |
| 240 | * being watched to filter extraneous exceptions. |
| 241 | */ |
| 242 | if (!((bp->attr.bp_addr <= dar) && |
| 243 | (dar <= (bp->attr.bp_addr + bp->attr.bp_len)))) { |
| 244 | /* |
| 245 | * This exception is triggered not because of a memory access |
| 246 | * on the monitored variable but in the double-word address |
| 247 | * range in which it is contained. We will consume this |
| 248 | * exception, considering it as 'noise'. |
| 249 | */ |
| 250 | info->extraneous_interrupt = true; |
| 251 | } else |
| 252 | info->extraneous_interrupt = false; |
| 253 | |
K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 254 | /* Do not emulate user-space instructions, instead single-step them */ |
| 255 | if (user_mode(regs)) { |
| 256 | bp->ctx->task->thread.last_hit_ubp = bp; |
| 257 | regs->msr |= MSR_SE; |
| 258 | goto out; |
| 259 | } |
| 260 | |
| 261 | stepped = 0; |
| 262 | instr = 0; |
| 263 | if (!__get_user_inatomic(instr, (unsigned int *) regs->nip)) |
| 264 | stepped = emulate_step(regs, instr); |
| 265 | |
| 266 | /* |
| 267 | * emulate_step() could not execute it. We've failed in reliably |
| 268 | * handling the hw-breakpoint. Unregister it and throw a warning |
| 269 | * message to let the user know about it. |
| 270 | */ |
| 271 | if (!stepped) { |
| 272 | WARN(1, "Unable to handle hardware breakpoint. Breakpoint at " |
| 273 | "0x%lx will be disabled.", info->address); |
| 274 | perf_event_disable(bp); |
| 275 | goto out; |
| 276 | } |
| 277 | /* |
| 278 | * As a policy, the callback is invoked in a 'trigger-after-execute' |
| 279 | * fashion |
| 280 | */ |
K.Prasad | e3e9408 | 2010-06-15 11:36:12 +0530 | [diff] [blame^] | 281 | if (!info->extraneous_interrupt) |
| 282 | perf_bp_event(bp, regs); |
K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 283 | |
| 284 | set_dabr(info->address | info->type | DABR_TRANSLATION); |
| 285 | out: |
| 286 | rcu_read_unlock(); |
| 287 | return rc; |
| 288 | } |
| 289 | |
| 290 | /* |
| 291 | * Handle single-step exceptions following a DABR hit. |
| 292 | */ |
| 293 | int __kprobes single_step_dabr_instruction(struct die_args *args) |
| 294 | { |
| 295 | struct pt_regs *regs = args->regs; |
| 296 | struct perf_event *bp = NULL; |
| 297 | struct arch_hw_breakpoint *bp_info; |
| 298 | |
| 299 | bp = current->thread.last_hit_ubp; |
| 300 | /* |
| 301 | * Check if we are single-stepping as a result of a |
| 302 | * previous HW Breakpoint exception |
| 303 | */ |
| 304 | if (!bp) |
| 305 | return NOTIFY_DONE; |
| 306 | |
| 307 | bp_info = counter_arch_bp(bp); |
| 308 | |
| 309 | /* |
| 310 | * We shall invoke the user-defined callback function in the single |
| 311 | * stepping handler to confirm to 'trigger-after-execute' semantics |
| 312 | */ |
K.Prasad | e3e9408 | 2010-06-15 11:36:12 +0530 | [diff] [blame^] | 313 | if (!bp_info->extraneous_interrupt) |
| 314 | perf_bp_event(bp, regs); |
K.Prasad | 5aae8a5 | 2010-06-15 11:35:19 +0530 | [diff] [blame] | 315 | |
| 316 | /* |
| 317 | * Do not disable MSR_SE if the process was already in |
| 318 | * single-stepping mode. |
| 319 | */ |
| 320 | if (!test_thread_flag(TIF_SINGLESTEP)) |
| 321 | regs->msr &= ~MSR_SE; |
| 322 | |
| 323 | set_dabr(bp_info->address | bp_info->type | DABR_TRANSLATION); |
| 324 | current->thread.last_hit_ubp = NULL; |
| 325 | return NOTIFY_STOP; |
| 326 | } |
| 327 | |
| 328 | /* |
| 329 | * Handle debug exception notifications. |
| 330 | */ |
| 331 | int __kprobes hw_breakpoint_exceptions_notify( |
| 332 | struct notifier_block *unused, unsigned long val, void *data) |
| 333 | { |
| 334 | int ret = NOTIFY_DONE; |
| 335 | |
| 336 | switch (val) { |
| 337 | case DIE_DABR_MATCH: |
| 338 | ret = hw_breakpoint_handler(data); |
| 339 | break; |
| 340 | case DIE_SSTEP: |
| 341 | ret = single_step_dabr_instruction(data); |
| 342 | break; |
| 343 | } |
| 344 | |
| 345 | return ret; |
| 346 | } |
| 347 | |
| 348 | /* |
| 349 | * Release the user breakpoints used by ptrace |
| 350 | */ |
| 351 | void flush_ptrace_hw_breakpoint(struct task_struct *tsk) |
| 352 | { |
| 353 | struct thread_struct *t = &tsk->thread; |
| 354 | |
| 355 | unregister_hw_breakpoint(t->ptrace_bps[0]); |
| 356 | t->ptrace_bps[0] = NULL; |
| 357 | } |
| 358 | |
| 359 | void hw_breakpoint_pmu_read(struct perf_event *bp) |
| 360 | { |
| 361 | /* TODO */ |
| 362 | } |