arch/x86/mm/mmio-mod.c

   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/version.h>
  33 #include <linux/kallsyms.h>
  34 #include <asm/pgtable.h>
  35 #include <linux/mmiotrace.h>
  36 #include <asm/e820.h> /* for ISA_START_ADDRESS */
  37 #include <asm/atomic.h>
  38 #include <linux/percpu.h>
  39 #include <linux/cpu.h>
  40
  41 #include "pf_in.h"
  42
  43 struct trap_reason {
  44         unsigned long addr;
  45         unsigned long ip;
  46         enum reason_type type;
  47         int active_traces;
  48 };
  49
  50 struct remap_trace {
  51         struct list_head list;
  52         struct kmmio_probe probe;
  53         resource_size_t phys;
  54         unsigned long id;
  55 };
  56
  57 /* Accessed per-cpu. */
  58 static DEFINE_PER_CPU(struct trap_reason, pf_reason);
  59 static DEFINE_PER_CPU(struct mmiotrace_rw, cpu_trace);
  60
  61 static DEFINE_MUTEX(mmiotrace_mutex);
  62 static DEFINE_SPINLOCK(trace_lock);
  63 static atomic_t mmiotrace_enabled;
  64 static LIST_HEAD(trace_list);           /* struct remap_trace */
  65
  66 /*
  67  * Locking in this file:
  68  * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections.
  69  * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex
  70  *   and trace_lock.
  71  * - Routines depending on is_enabled() must take trace_lock.
  72  * - trace_list users must hold trace_lock.
  73  * - is_enabled() guarantees that mmio_trace_{rw,mapping} are allowed.
  74  * - pre/post callbacks assume the effect of is_enabled() being true.
  75  */
  76
  77 /* module parameters */
  78 static unsigned long    filter_offset;
  79 static int              nommiotrace;
  80 static int              trace_pc;
  81
  82 module_param(filter_offset, ulong, 0);
  83 module_param(nommiotrace, bool, 0);
  84 module_param(trace_pc, bool, 0);
  85
  86 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
  87 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
  88 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
  89
  90 static bool is_enabled(void)
  91 {
  92         return atomic_read(&mmiotrace_enabled);
  93 }
  94
  95 static void print_pte(unsigned long address)
  96 {
  97         unsigned int level;
  98         pte_t *pte = lookup_address(address, &level);
  99
 100         if (!pte) {
 101                 pr_err("Error in %s: no pte for page 0x%08lx\n",
 102                        __func__, address);
 103                 return;
 104         }
 105
 106         if (level == PG_LEVEL_2M) {
 107                 pr_emerg("4MB pages are not currently supported: 0x%08lx\n",
 108                          address);
 109                 BUG();
 110         }
 111         pr_info("pte for 0x%lx: 0x%llx 0x%llx\n",
 112                 address,
 113                 (unsigned long long)pte_val(*pte),
 114                 (unsigned long long)pte_val(*pte) & _PAGE_PRESENT);
 115 }
 116
 117 /*
 118  * For some reason the pre/post pairs have been called in an
 119  * unmatched order. Report and die.
 120  */
 121 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
 122 {
 123         const struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 124         pr_emerg("unexpected fault for address: 0x%08lx, last fault for address: 0x%08lx\n",
 125                  addr, my_reason->addr);
 126         print_pte(addr);
 127         print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip);
 128         print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip);
 129 #ifdef __i386__
 130         pr_emerg("eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
 131                  regs->ax, regs->bx, regs->cx, regs->dx);
 132         pr_emerg("esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
 133                  regs->si, regs->di, regs->bp, regs->sp);
 134 #else
 135         pr_emerg("rax: %016lx   rcx: %016lx   rdx: %016lx\n",
 136                  regs->ax, regs->cx, regs->dx);
 137         pr_emerg("rsi: %016lx   rdi: %016lx   rbp: %016lx   rsp: %016lx\n",
 138                  regs->si, regs->di, regs->bp, regs->sp);
 139 #endif
 140         put_cpu_var(pf_reason);
 141         BUG();
 142 }
 143
 144 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
 145                                                 unsigned long addr)
 146 {
 147         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 148         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
 149         const unsigned long instptr = instruction_pointer(regs);
 150         const enum reason_type type = get_ins_type(instptr);
 151         struct remap_trace *trace = p->private;
 152
 153         /* it doesn't make sense to have more than one active trace per cpu */
 154         if (my_reason->active_traces)
 155                 die_kmmio_nesting_error(regs, addr);
 156         else
 157                 my_reason->active_traces++;
 158
 159         my_reason->type = type;
 160         my_reason->addr = addr;
 161         my_reason->ip = instptr;
 162
 163         my_trace->phys = addr - trace->probe.addr + trace->phys;
 164         my_trace->map_id = trace->id;
 165
 166         /*
 167          * Only record the program counter when requested.
 168          * It may taint clean-room reverse engineering.
 169          */
 170         if (trace_pc)
 171                 my_trace->pc = instptr;
 172         else
 173                 my_trace->pc = 0;
 174
 175         /*
 176          * XXX: the timestamp recorded will be *after* the tracing has been
 177          * done, not at the time we hit the instruction. SMP implications
 178          * on event ordering?
 179          */
 180
 181         switch (type) {
 182         case REG_READ:
 183                 my_trace->opcode = MMIO_READ;
 184                 my_trace->width = get_ins_mem_width(instptr);
 185                 break;
 186         case REG_WRITE:
 187                 my_trace->opcode = MMIO_WRITE;
 188                 my_trace->width = get_ins_mem_width(instptr);
 189                 my_trace->value = get_ins_reg_val(instptr, regs);
 190                 break;
 191         case IMM_WRITE:
 192                 my_trace->opcode = MMIO_WRITE;
 193                 my_trace->width = get_ins_mem_width(instptr);
 194                 my_trace->value = get_ins_imm_val(instptr);
 195                 break;
 196         default:
 197                 {
 198                         unsigned char *ip = (unsigned char *)instptr;
 199                         my_trace->opcode = MMIO_UNKNOWN_OP;
 200                         my_trace->width = 0;
 201                         my_trace->value = (*ip) << 16 | *(ip + 1) << 8 |
 202                                                                 *(ip + 2);
 203                 }
 204         }
 205         put_cpu_var(cpu_trace);
 206         put_cpu_var(pf_reason);
 207 }
 208
 209 static void post(struct kmmio_probe *p, unsigned long condition,
 210                                                         struct pt_regs *regs)
 211 {
 212         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 213         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
 214
 215         /* this should always return the active_trace count to 0 */
 216         my_reason->active_traces--;
 217         if (my_reason->active_traces) {
 218                 pr_emerg("unexpected post handler");
 219                 BUG();
 220         }
 221
 222         switch (my_reason->type) {
 223         case REG_READ:
 224                 my_trace->value = get_ins_reg_val(my_reason->ip, regs);
 225                 break;
 226         default:
 227                 break;
 228         }
 229
 230         mmio_trace_rw(my_trace);
 231         put_cpu_var(cpu_trace);
 232         put_cpu_var(pf_reason);
 233 }
 234
 235 static void ioremap_trace_core(resource_size_t offset, unsigned long size,
 236                                                         void __iomem *addr)
 237 {
 238         static atomic_t next_id;
 239         struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
 240         /* These are page-unaligned. */
 241         struct mmiotrace_map map = {
 242                 .phys = offset,
 243                 .virt = (unsigned long)addr,
 244                 .len = size,
 245                 .opcode = MMIO_PROBE
 246         };
 247
 248         if (!trace) {
 249                 pr_err("kmalloc failed in ioremap\n");
 250                 return;
 251         }
 252
 253         *trace = (struct remap_trace) {
 254                 .probe = {
 255                         .addr = (unsigned long)addr,
 256                         .len = size,
 257                         .pre_handler = pre,
 258                         .post_handler = post,
 259                         .private = trace
 260                 },
 261                 .phys = offset,
 262                 .id = atomic_inc_return(&next_id)
 263         };
 264         map.map_id = trace->id;
 265
 266         spin_lock_irq(&trace_lock);
 267         if (!is_enabled()) {
 268                 kfree(trace);
 269                 goto not_enabled;
 270         }
 271
 272         mmio_trace_mapping(&map);
 273         list_add_tail(&trace->list, &trace_list);
 274         if (!nommiotrace)
 275                 register_kmmio_probe(&trace->probe);
 276
 277 not_enabled:
 278         spin_unlock_irq(&trace_lock);
 279 }
 280
 281 void mmiotrace_ioremap(resource_size_t offset, unsigned long size,
 282                                                 void __iomem *addr)
 283 {
 284         if (!is_enabled()) /* recheck and proper locking in *_core() */
 285                 return;
 286
 287         pr_debug("ioremap_*(0x%llx, 0x%lx) = %p\n",
 288                  (unsigned long long)offset, size, addr);
 289         if ((filter_offset) && (offset != filter_offset))
 290                 return;
 291         ioremap_trace_core(offset, size, addr);
 292 }
 293
 294 static void iounmap_trace_core(volatile void __iomem *addr)
 295 {
 296         struct mmiotrace_map map = {
 297                 .phys = 0,
 298                 .virt = (unsigned long)addr,
 299                 .len = 0,
 300                 .opcode = MMIO_UNPROBE
 301         };
 302         struct remap_trace *trace;
 303         struct remap_trace *tmp;
 304         struct remap_trace *found_trace = NULL;
 305
 306         pr_debug("Unmapping %p.\n", addr);
 307
 308         spin_lock_irq(&trace_lock);
 309         if (!is_enabled())
 310                 goto not_enabled;
 311
 312         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
 313                 if ((unsigned long)addr == trace->probe.addr) {
 314                         if (!nommiotrace)
 315                                 unregister_kmmio_probe(&trace->probe);
 316                         list_del(&trace->list);
 317                         found_trace = trace;
 318                         break;
 319                 }
 320         }
 321         map.map_id = (found_trace) ? found_trace->id : -1;
 322         mmio_trace_mapping(&map);
 323
 324 not_enabled:
 325         spin_unlock_irq(&trace_lock);
 326         if (found_trace) {
 327                 synchronize_rcu(); /* unregister_kmmio_probe() requirement */
 328                 kfree(found_trace);
 329         }
 330 }
 331
 332 void mmiotrace_iounmap(volatile void __iomem *addr)
 333 {
 334         might_sleep();
 335         if (is_enabled()) /* recheck and proper locking in *_core() */
 336                 iounmap_trace_core(addr);
 337 }
 338
 339 int mmiotrace_printk(const char *fmt, ...)
 340 {
 341         int ret = 0;
 342         va_list args;
 343         unsigned long flags;
 344         va_start(args, fmt);
 345
 346         spin_lock_irqsave(&trace_lock, flags);
 347         if (is_enabled())
 348                 ret = mmio_trace_printk(fmt, args);
 349         spin_unlock_irqrestore(&trace_lock, flags);
 350
 351         va_end(args);
 352         return ret;
 353 }
 354 EXPORT_SYMBOL(mmiotrace_printk);
 355
 356 static void clear_trace_list(void)
 357 {
 358         struct remap_trace *trace;
 359         struct remap_trace *tmp;
 360
 361         /*
 362          * No locking required, because the caller ensures we are in a
 363          * critical section via mutex, and is_enabled() is false,
 364          * i.e. nothing can traverse or modify this list.
 365          * Caller also ensures is_enabled() cannot change.
 366          */
 367         list_for_each_entry(trace, &trace_list, list) {
 368                 pr_notice("purging non-iounmapped trace @0x%08lx, size 0x%lx.\n",
 369                           trace->probe.addr, trace->probe.len);
 370                 if (!nommiotrace)
 371                         unregister_kmmio_probe(&trace->probe);
 372         }
 373         synchronize_rcu(); /* unregister_kmmio_probe() requirement */
 374
 375         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
 376                 list_del(&trace->list);
 377                 kfree(trace);
 378         }
 379 }
 380
 381 #ifdef CONFIG_HOTPLUG_CPU
 382 static cpumask_var_t downed_cpus;
 383
 384 static void enter_uniprocessor(void)
 385 {
 386         int cpu;
 387         int err;
 388
 389         if (downed_cpus == NULL &&
 390             !alloc_cpumask_var(&downed_cpus, GFP_KERNEL)) {
 391                 pr_notice("Failed to allocate mask\n");
 392                 goto out;
 393         }
 394
 395         get_online_cpus();
 396         cpumask_copy(downed_cpus, cpu_online_mask);
 397         cpumask_clear_cpu(cpumask_first(cpu_online_mask), downed_cpus);
 398         if (num_online_cpus() > 1)
 399                 pr_notice("Disabling non-boot CPUs...\n");
 400         put_online_cpus();
 401
 402         for_each_cpu(cpu, downed_cpus) {
 403                 err = cpu_down(cpu);
 404                 if (!err)
 405                         pr_info("CPU%d is down.\n", cpu);
 406                 else
 407                         pr_err("Error taking CPU%d down: %d\n", cpu, err);
 408         }
 409 out:
 410         if (num_online_cpus() > 1)
 411                 pr_warning("multiple CPUs still online, may miss events.\n");
 412 }
 413
 414 /* __ref because leave_uniprocessor calls cpu_up which is __cpuinit,
 415    but this whole function is ifdefed CONFIG_HOTPLUG_CPU */
 416 static void __ref leave_uniprocessor(void)
 417 {
 418         int cpu;
 419         int err;
 420
 421         if (downed_cpus == NULL || cpumask_weight(downed_cpus) == 0)
 422                 return;
 423         pr_notice("Re-enabling CPUs...\n");
 424         for_each_cpu(cpu, downed_cpus) {
 425                 err = cpu_up(cpu);
 426                 if (!err)
 427                         pr_info("enabled CPU%d.\n", cpu);
 428                 else
 429                         pr_err("cannot re-enable CPU%d: %d\n", cpu, err);
 430         }
 431 }
 432
 433 #else /* !CONFIG_HOTPLUG_CPU */
 434 static void enter_uniprocessor(void)
 435 {
 436         if (num_online_cpus() > 1)
 437                 pr_warning("multiple CPUs are online, may miss events. "
 438                            "Suggest booting with maxcpus=1 kernel argument.\n");
 439 }
 440
 441 static void leave_uniprocessor(void)
 442 {
 443 }
 444 #endif
 445
 446 void enable_mmiotrace(void)
 447 {
 448         mutex_lock(&mmiotrace_mutex);
 449         if (is_enabled())
 450                 goto out;
 451
 452         if (nommiotrace)
 453                 pr_info("MMIO tracing disabled.\n");
 454         kmmio_init();
 455         enter_uniprocessor();
 456         spin_lock_irq(&trace_lock);
 457         atomic_inc(&mmiotrace_enabled);
 458         spin_unlock_irq(&trace_lock);
 459         pr_info("enabled.\n");
 460 out:
 461         mutex_unlock(&mmiotrace_mutex);
 462 }
 463
 464 void disable_mmiotrace(void)
 465 {
 466         mutex_lock(&mmiotrace_mutex);
 467         if (!is_enabled())
 468                 goto out;
 469
 470         spin_lock_irq(&trace_lock);
 471         atomic_dec(&mmiotrace_enabled);
 472         BUG_ON(is_enabled());
 473         spin_unlock_irq(&trace_lock);
 474
 475         clear_trace_list(); /* guarantees: no more kmmio callbacks */
 476         leave_uniprocessor();
 477         kmmio_cleanup();
 478         pr_info("disabled.\n");
 479 out:
 480         mutex_unlock(&mmiotrace_mutex);
 481 }