| 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) IBM Corporation, 2005 |
| 17 | * Jeff Muizelaar, 2006, 2007 |
| 18 | * Pekka Paalanen, 2008 <pq@iki.fi> |
| 19 | * |
| 20 | * Derived from the read-mod example from relay-examples by Tom Zanussi. |
| 21 | */ |
| 22 | |
| 23 | #define pr_fmt(fmt) "mmiotrace: " fmt |
| 24 | |
| 25 | #define DEBUG 1 |
| 26 | |
| 27 | #include <linux/module.h> |
| 28 | #include <linux/debugfs.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/uaccess.h> |
| 31 | #include <linux/io.h> |
| 32 | #include <linux/kallsyms.h> |
| 33 | #include <asm/pgtable.h> |
| 34 | #include <linux/mmiotrace.h> |
| 35 | #include <asm/e820.h> /* for ISA_START_ADDRESS */ |
| 36 | #include <linux/atomic.h> |
| 37 | #include <linux/percpu.h> |
| 38 | #include <linux/cpu.h> |
| 39 | |
| 40 | #include "pf_in.h" |
| 41 | |
| 42 | struct trap_reason { |
| 43 | unsigned long addr; |
| 44 | unsigned long ip; |
| 45 | enum reason_type type; |
| 46 | int active_traces; |
| 47 | }; |
| 48 | |
| 49 | struct remap_trace { |
| 50 | struct list_head list; |
| 51 | struct kmmio_probe probe; |
| 52 | resource_size_t phys; |
| 53 | unsigned long id; |
| 54 | }; |
| 55 | |
| 56 | /* Accessed per-cpu. */ |
| 57 | static DEFINE_PER_CPU(struct trap_reason, pf_reason); |
| 58 | static DEFINE_PER_CPU(struct mmiotrace_rw, cpu_trace); |
| 59 | |
| 60 | static DEFINE_MUTEX(mmiotrace_mutex); |
| 61 | static DEFINE_SPINLOCK(trace_lock); |
| 62 | static atomic_t mmiotrace_enabled; |
| 63 | static LIST_HEAD(trace_list); /* struct remap_trace */ |
| 64 | |
| 65 | /* |
| 66 | * Locking in this file: |
| 67 | * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections. |
| 68 | * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex |
| 69 | * and trace_lock. |
| 70 | * - Routines depending on is_enabled() must take trace_lock. |
| 71 | * - trace_list users must hold trace_lock. |
| 72 | * - is_enabled() guarantees that mmio_trace_{rw,mapping} are allowed. |
| 73 | * - pre/post callbacks assume the effect of is_enabled() being true. |
| 74 | */ |
| 75 | |
| 76 | /* module parameters */ |
| 77 | static unsigned long filter_offset; |
| 78 | static bool nommiotrace; |
| 79 | static bool trace_pc; |
| 80 | |
| 81 | module_param(filter_offset, ulong, 0); |
| 82 | module_param(nommiotrace, bool, 0); |
| 83 | module_param(trace_pc, bool, 0); |
| 84 | |
| 85 | MODULE_PARM_DESC(filter_offset, "Start address of traced mappings."); |
| 86 | MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing."); |
| 87 | MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions."); |
| 88 | |
| 89 | static bool is_enabled(void) |
| 90 | { |
| 91 | return atomic_read(&mmiotrace_enabled); |
| 92 | } |
| 93 | |
| 94 | static void print_pte(unsigned long address) |
| 95 | { |
| 96 | unsigned int level; |
| 97 | pte_t *pte = lookup_address(address, &level); |
| 98 | |
| 99 | if (!pte) { |
| 100 | pr_err("Error in %s: no pte for page 0x%08lx\n", |
| 101 | __func__, address); |
| 102 | return; |
| 103 | } |
| 104 | |
| 105 | if (level == PG_LEVEL_2M) { |
| 106 | pr_emerg("4MB pages are not currently supported: 0x%08lx\n", |
| 107 | address); |
| 108 | BUG(); |
| 109 | } |
| 110 | pr_info("pte for 0x%lx: 0x%llx 0x%llx\n", |
| 111 | address, |
| 112 | (unsigned long long)pte_val(*pte), |
| 113 | (unsigned long long)pte_val(*pte) & _PAGE_PRESENT); |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * For some reason the pre/post pairs have been called in an |
| 118 | * unmatched order. Report and die. |
| 119 | */ |
| 120 | static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr) |
| 121 | { |
| 122 | const struct trap_reason *my_reason = &get_cpu_var(pf_reason); |
| 123 | pr_emerg("unexpected fault for address: 0x%08lx, last fault for address: 0x%08lx\n", |
| 124 | addr, my_reason->addr); |
| 125 | print_pte(addr); |
| 126 | print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip); |
| 127 | print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip); |
| 128 | #ifdef __i386__ |
| 129 | pr_emerg("eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n", |
| 130 | regs->ax, regs->bx, regs->cx, regs->dx); |
| 131 | pr_emerg("esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n", |
| 132 | regs->si, regs->di, regs->bp, regs->sp); |
| 133 | #else |
| 134 | pr_emerg("rax: %016lx rcx: %016lx rdx: %016lx\n", |
| 135 | regs->ax, regs->cx, regs->dx); |
| 136 | pr_emerg("rsi: %016lx rdi: %016lx rbp: %016lx rsp: %016lx\n", |
| 137 | regs->si, regs->di, regs->bp, regs->sp); |
| 138 | #endif |
| 139 | put_cpu_var(pf_reason); |
| 140 | BUG(); |
| 141 | } |
| 142 | |
| 143 | static void pre(struct kmmio_probe *p, struct pt_regs *regs, |
| 144 | unsigned long addr) |
| 145 | { |
| 146 | struct trap_reason *my_reason = &get_cpu_var(pf_reason); |
| 147 | struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace); |
| 148 | const unsigned long instptr = instruction_pointer(regs); |
| 149 | const enum reason_type type = get_ins_type(instptr); |
| 150 | struct remap_trace *trace = p->private; |
| 151 | |
| 152 | /* it doesn't make sense to have more than one active trace per cpu */ |
| 153 | if (my_reason->active_traces) |
| 154 | die_kmmio_nesting_error(regs, addr); |
| 155 | else |
| 156 | my_reason->active_traces++; |
| 157 | |
| 158 | my_reason->type = type; |
| 159 | my_reason->addr = addr; |
| 160 | my_reason->ip = instptr; |
| 161 | |
| 162 | my_trace->phys = addr - trace->probe.addr + trace->phys; |
| 163 | my_trace->map_id = trace->id; |
| 164 | |
| 165 | /* |
| 166 | * Only record the program counter when requested. |
| 167 | * It may taint clean-room reverse engineering. |
| 168 | */ |
| 169 | if (trace_pc) |
| 170 | my_trace->pc = instptr; |
| 171 | else |
| 172 | my_trace->pc = 0; |
| 173 | |
| 174 | /* |
| 175 | * XXX: the timestamp recorded will be *after* the tracing has been |
| 176 | * done, not at the time we hit the instruction. SMP implications |
| 177 | * on event ordering? |
| 178 | */ |
| 179 | |
| 180 | switch (type) { |
| 181 | case REG_READ: |
| 182 | my_trace->opcode = MMIO_READ; |
| 183 | my_trace->width = get_ins_mem_width(instptr); |
| 184 | break; |
| 185 | case REG_WRITE: |
| 186 | my_trace->opcode = MMIO_WRITE; |
| 187 | my_trace->width = get_ins_mem_width(instptr); |
| 188 | my_trace->value = get_ins_reg_val(instptr, regs); |
| 189 | break; |
| 190 | case IMM_WRITE: |
| 191 | my_trace->opcode = MMIO_WRITE; |
| 192 | my_trace->width = get_ins_mem_width(instptr); |
| 193 | my_trace->value = get_ins_imm_val(instptr); |
| 194 | break; |
| 195 | default: |
| 196 | { |
| 197 | unsigned char *ip = (unsigned char *)instptr; |
| 198 | my_trace->opcode = MMIO_UNKNOWN_OP; |
| 199 | my_trace->width = 0; |
| 200 | my_trace->value = (*ip) << 16 | *(ip + 1) << 8 | |
| 201 | *(ip + 2); |
| 202 | } |
| 203 | } |
| 204 | put_cpu_var(cpu_trace); |
| 205 | put_cpu_var(pf_reason); |
| 206 | } |
| 207 | |
| 208 | static void post(struct kmmio_probe *p, unsigned long condition, |
| 209 | struct pt_regs *regs) |
| 210 | { |
| 211 | struct trap_reason *my_reason = &get_cpu_var(pf_reason); |
| 212 | struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace); |
| 213 | |
| 214 | /* this should always return the active_trace count to 0 */ |
| 215 | my_reason->active_traces--; |
| 216 | if (my_reason->active_traces) { |
| 217 | pr_emerg("unexpected post handler"); |
| 218 | BUG(); |
| 219 | } |
| 220 | |
| 221 | switch (my_reason->type) { |
| 222 | case REG_READ: |
| 223 | my_trace->value = get_ins_reg_val(my_reason->ip, regs); |
| 224 | break; |
| 225 | default: |
| 226 | break; |
| 227 | } |
| 228 | |
| 229 | mmio_trace_rw(my_trace); |
| 230 | put_cpu_var(cpu_trace); |
| 231 | put_cpu_var(pf_reason); |
| 232 | } |
| 233 | |
| 234 | static void ioremap_trace_core(resource_size_t offset, unsigned long size, |
| 235 | void __iomem *addr) |
| 236 | { |
| 237 | static atomic_t next_id; |
| 238 | struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL); |
| 239 | /* These are page-unaligned. */ |
| 240 | struct mmiotrace_map map = { |
| 241 | .phys = offset, |
| 242 | .virt = (unsigned long)addr, |
| 243 | .len = size, |
| 244 | .opcode = MMIO_PROBE |
| 245 | }; |
| 246 | |
| 247 | if (!trace) { |
| 248 | pr_err("kmalloc failed in ioremap\n"); |
| 249 | return; |
| 250 | } |
| 251 | |
| 252 | *trace = (struct remap_trace) { |
| 253 | .probe = { |
| 254 | .addr = (unsigned long)addr, |
| 255 | .len = size, |
| 256 | .pre_handler = pre, |
| 257 | .post_handler = post, |
| 258 | .private = trace |
| 259 | }, |
| 260 | .phys = offset, |
| 261 | .id = atomic_inc_return(&next_id) |
| 262 | }; |
| 263 | map.map_id = trace->id; |
| 264 | |
| 265 | spin_lock_irq(&trace_lock); |
| 266 | if (!is_enabled()) { |
| 267 | kfree(trace); |
| 268 | goto not_enabled; |
| 269 | } |
| 270 | |
| 271 | mmio_trace_mapping(&map); |
| 272 | list_add_tail(&trace->list, &trace_list); |
| 273 | if (!nommiotrace) |
| 274 | register_kmmio_probe(&trace->probe); |
| 275 | |
| 276 | not_enabled: |
| 277 | spin_unlock_irq(&trace_lock); |
| 278 | } |
| 279 | |
| 280 | void mmiotrace_ioremap(resource_size_t offset, unsigned long size, |
| 281 | void __iomem *addr) |
| 282 | { |
| 283 | if (!is_enabled()) /* recheck and proper locking in *_core() */ |
| 284 | return; |
| 285 | |
| 286 | pr_debug("ioremap_*(0x%llx, 0x%lx) = %p\n", |
| 287 | (unsigned long long)offset, size, addr); |
| 288 | if ((filter_offset) && (offset != filter_offset)) |
| 289 | return; |
| 290 | ioremap_trace_core(offset, size, addr); |
| 291 | } |
| 292 | |
| 293 | static void iounmap_trace_core(volatile void __iomem *addr) |
| 294 | { |
| 295 | struct mmiotrace_map map = { |
| 296 | .phys = 0, |
| 297 | .virt = (unsigned long)addr, |
| 298 | .len = 0, |
| 299 | .opcode = MMIO_UNPROBE |
| 300 | }; |
| 301 | struct remap_trace *trace; |
| 302 | struct remap_trace *tmp; |
| 303 | struct remap_trace *found_trace = NULL; |
| 304 | |
| 305 | pr_debug("Unmapping %p.\n", addr); |
| 306 | |
| 307 | spin_lock_irq(&trace_lock); |
| 308 | if (!is_enabled()) |
| 309 | goto not_enabled; |
| 310 | |
| 311 | list_for_each_entry_safe(trace, tmp, &trace_list, list) { |
| 312 | if ((unsigned long)addr == trace->probe.addr) { |
| 313 | if (!nommiotrace) |
| 314 | unregister_kmmio_probe(&trace->probe); |
| 315 | list_del(&trace->list); |
| 316 | found_trace = trace; |
| 317 | break; |
| 318 | } |
| 319 | } |
| 320 | map.map_id = (found_trace) ? found_trace->id : -1; |
| 321 | mmio_trace_mapping(&map); |
| 322 | |
| 323 | not_enabled: |
| 324 | spin_unlock_irq(&trace_lock); |
| 325 | if (found_trace) { |
| 326 | synchronize_rcu(); /* unregister_kmmio_probe() requirement */ |
| 327 | kfree(found_trace); |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | void mmiotrace_iounmap(volatile void __iomem *addr) |
| 332 | { |
| 333 | might_sleep(); |
| 334 | if (is_enabled()) /* recheck and proper locking in *_core() */ |
| 335 | iounmap_trace_core(addr); |
| 336 | } |
| 337 | |
| 338 | int mmiotrace_printk(const char *fmt, ...) |
| 339 | { |
| 340 | int ret = 0; |
| 341 | va_list args; |
| 342 | unsigned long flags; |
| 343 | va_start(args, fmt); |
| 344 | |
| 345 | spin_lock_irqsave(&trace_lock, flags); |
| 346 | if (is_enabled()) |
| 347 | ret = mmio_trace_printk(fmt, args); |
| 348 | spin_unlock_irqrestore(&trace_lock, flags); |
| 349 | |
| 350 | va_end(args); |
| 351 | return ret; |
| 352 | } |
| 353 | EXPORT_SYMBOL(mmiotrace_printk); |
| 354 | |
| 355 | static void clear_trace_list(void) |
| 356 | { |
| 357 | struct remap_trace *trace; |
| 358 | struct remap_trace *tmp; |
| 359 | |
| 360 | /* |
| 361 | * No locking required, because the caller ensures we are in a |
| 362 | * critical section via mutex, and is_enabled() is false, |
| 363 | * i.e. nothing can traverse or modify this list. |
| 364 | * Caller also ensures is_enabled() cannot change. |
| 365 | */ |
| 366 | list_for_each_entry(trace, &trace_list, list) { |
| 367 | pr_notice("purging non-iounmapped trace @0x%08lx, size 0x%lx.\n", |
| 368 | trace->probe.addr, trace->probe.len); |
| 369 | if (!nommiotrace) |
| 370 | unregister_kmmio_probe(&trace->probe); |
| 371 | } |
| 372 | synchronize_rcu(); /* unregister_kmmio_probe() requirement */ |
| 373 | |
| 374 | list_for_each_entry_safe(trace, tmp, &trace_list, list) { |
| 375 | list_del(&trace->list); |
| 376 | kfree(trace); |
| 377 | } |
| 378 | } |
| 379 | |
| 380 | #ifdef CONFIG_HOTPLUG_CPU |
| 381 | static cpumask_var_t downed_cpus; |
| 382 | |
| 383 | static void enter_uniprocessor(void) |
| 384 | { |
| 385 | int cpu; |
| 386 | int err; |
| 387 | |
| 388 | if (downed_cpus == NULL && |
| 389 | !alloc_cpumask_var(&downed_cpus, GFP_KERNEL)) { |
| 390 | pr_notice("Failed to allocate mask\n"); |
| 391 | goto out; |
| 392 | } |
| 393 | |
| 394 | get_online_cpus(); |
| 395 | cpumask_copy(downed_cpus, cpu_online_mask); |
| 396 | cpumask_clear_cpu(cpumask_first(cpu_online_mask), downed_cpus); |
| 397 | if (num_online_cpus() > 1) |
| 398 | pr_notice("Disabling non-boot CPUs...\n"); |
| 399 | put_online_cpus(); |
| 400 | |
| 401 | for_each_cpu(cpu, downed_cpus) { |
| 402 | err = cpu_down(cpu); |
| 403 | if (!err) |
| 404 | pr_info("CPU%d is down.\n", cpu); |
| 405 | else |
| 406 | pr_err("Error taking CPU%d down: %d\n", cpu, err); |
| 407 | } |
| 408 | out: |
| 409 | if (num_online_cpus() > 1) |
| 410 | pr_warning("multiple CPUs still online, may miss events.\n"); |
| 411 | } |
| 412 | |
| 413 | static void leave_uniprocessor(void) |
| 414 | { |
| 415 | int cpu; |
| 416 | int err; |
| 417 | |
| 418 | if (downed_cpus == NULL || cpumask_weight(downed_cpus) == 0) |
| 419 | return; |
| 420 | pr_notice("Re-enabling CPUs...\n"); |
| 421 | for_each_cpu(cpu, downed_cpus) { |
| 422 | err = cpu_up(cpu); |
| 423 | if (!err) |
| 424 | pr_info("enabled CPU%d.\n", cpu); |
| 425 | else |
| 426 | pr_err("cannot re-enable CPU%d: %d\n", cpu, err); |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | #else /* !CONFIG_HOTPLUG_CPU */ |
| 431 | static void enter_uniprocessor(void) |
| 432 | { |
| 433 | if (num_online_cpus() > 1) |
| 434 | pr_warning("multiple CPUs are online, may miss events. " |
| 435 | "Suggest booting with maxcpus=1 kernel argument.\n"); |
| 436 | } |
| 437 | |
| 438 | static void leave_uniprocessor(void) |
| 439 | { |
| 440 | } |
| 441 | #endif |
| 442 | |
| 443 | void enable_mmiotrace(void) |
| 444 | { |
| 445 | mutex_lock(&mmiotrace_mutex); |
| 446 | if (is_enabled()) |
| 447 | goto out; |
| 448 | |
| 449 | if (nommiotrace) |
| 450 | pr_info("MMIO tracing disabled.\n"); |
| 451 | kmmio_init(); |
| 452 | enter_uniprocessor(); |
| 453 | spin_lock_irq(&trace_lock); |
| 454 | atomic_inc(&mmiotrace_enabled); |
| 455 | spin_unlock_irq(&trace_lock); |
| 456 | pr_info("enabled.\n"); |
| 457 | out: |
| 458 | mutex_unlock(&mmiotrace_mutex); |
| 459 | } |
| 460 | |
| 461 | void disable_mmiotrace(void) |
| 462 | { |
| 463 | mutex_lock(&mmiotrace_mutex); |
| 464 | if (!is_enabled()) |
| 465 | goto out; |
| 466 | |
| 467 | spin_lock_irq(&trace_lock); |
| 468 | atomic_dec(&mmiotrace_enabled); |
| 469 | BUG_ON(is_enabled()); |
| 470 | spin_unlock_irq(&trace_lock); |
| 471 | |
| 472 | clear_trace_list(); /* guarantees: no more kmmio callbacks */ |
| 473 | leave_uniprocessor(); |
| 474 | kmmio_cleanup(); |
| 475 | pr_info("disabled.\n"); |
| 476 | out: |
| 477 | mutex_unlock(&mmiotrace_mutex); |
| 478 | } |