err.no Git - linux-2.6/blob - arch/x86/kernel/mmiotrace/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 #define DEBUG 1
  23
  24 #include <linux/module.h>
  25 #include <linux/relay.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/proc_fs.h>
  28 #include <asm/io.h>
  29 #include <linux/version.h>
  30 #include <linux/kallsyms.h>
  31 #include <asm/pgtable.h>
  32 #include <linux/mmiotrace.h>
  33 #include <asm/e820.h> /* for ISA_START_ADDRESS */
  34 #include <asm/atomic.h>
  35 #include <linux/percpu.h>
  36
  37 #include "pf_in.h"
  38
  39 #define NAME "mmiotrace: "
  40
  41 /* This app's relay channel files will appear in /debug/mmio-trace */
  42 static const char APP_DIR[] = "mmio-trace";
  43 /* the marker injection file in /debug/APP_DIR */
  44 static const char MARKER_FILE[] = "mmio-marker";
  45
  46 struct trap_reason {
  47         unsigned long addr;
  48         unsigned long ip;
  49         enum reason_type type;
  50         int active_traces;
  51 };
  52
  53 struct remap_trace {
  54         struct list_head list;
  55         struct kmmio_probe probe;
  56         unsigned long phys;
  57         unsigned long id;
  58 };
  59
  60 static const size_t subbuf_size = 256*1024;
  61
  62 /* Accessed per-cpu. */
  63 static DEFINE_PER_CPU(struct trap_reason, pf_reason);
  64 static DEFINE_PER_CPU(struct mm_io_header_rw, cpu_trace);
  65
  66 /* Access to this is not per-cpu. */
  67 static DEFINE_PER_CPU(atomic_t, dropped);
  68
  69 static struct dentry *dir;
  70 static struct dentry *enabled_file;
  71 static struct dentry *marker_file;
  72
  73 static DEFINE_MUTEX(mmiotrace_mutex);
  74 static DEFINE_SPINLOCK(trace_lock);
  75 static atomic_t mmiotrace_enabled;
  76 static LIST_HEAD(trace_list);           /* struct remap_trace */
  77 static struct rchan *chan;
  78
  79 /*
  80  * Locking in this file:
  81  * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections.
  82  * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex
  83  *   and trace_lock.
  84  * - Routines depending on is_enabled() must take trace_lock.
  85  * - trace_list users must hold trace_lock.
  86  * - is_enabled() guarantees that chan is valid.
  87  * - pre/post callbacks assume the effect of is_enabled() being true.
  88  */
  89
  90 /* module parameters */
  91 static unsigned int     n_subbufs = 32*4;
  92 static unsigned long    filter_offset;
  93 static int              nommiotrace;
  94 static int              ISA_trace;
  95 static int              trace_pc;
  96 static int              enable_now;
  97
  98 module_param(n_subbufs, uint, 0);
  99 module_param(filter_offset, ulong, 0);
 100 module_param(nommiotrace, bool, 0);
 101 module_param(ISA_trace, bool, 0);
 102 module_param(trace_pc, bool, 0);
 103 module_param(enable_now, bool, 0);
 104
 105 MODULE_PARM_DESC(n_subbufs, "Number of 256kB buffers, default 128.");
 106 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
 107 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
 108 MODULE_PARM_DESC(ISA_trace, "Do not exclude the low ISA range.");
 109 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
 110 MODULE_PARM_DESC(enable_now, "Start mmiotracing immediately on module load.");
 111
 112 static bool is_enabled(void)
 113 {
 114         return atomic_read(&mmiotrace_enabled);
 115 }
 116
 117 static void record_timestamp(struct mm_io_header *header)
 118 {
 119         struct timespec now;
 120
 121         getnstimeofday(&now);
 122         header->sec = now.tv_sec;
 123         header->nsec = now.tv_nsec;
 124 }
 125
 126 /*
 127  * Write callback for the debugfs entry:
 128  * Read a marker and write it to the mmio trace log
 129  */
 130 static ssize_t write_marker(struct file *file, const char __user *buffer,
 131                                                 size_t count, loff_t *ppos)
 132 {
 133         char *event = NULL;
 134         struct mm_io_header *headp;
 135         ssize_t len = (count > 65535) ? 65535 : count;
 136
 137         event = kzalloc(sizeof(*headp) + len, GFP_KERNEL);
 138         if (!event)
 139                 return -ENOMEM;
 140
 141         headp = (struct mm_io_header *)event;
 142         headp->type = MMIO_MAGIC | (MMIO_MARKER << MMIO_OPCODE_SHIFT);
 143         headp->data_len = len;
 144         record_timestamp(headp);
 145
 146         if (copy_from_user(event + sizeof(*headp), buffer, len)) {
 147                 kfree(event);
 148                 return -EFAULT;
 149         }
 150
 151         spin_lock_irq(&trace_lock);
 152         if (is_enabled())
 153                 relay_write(chan, event, sizeof(*headp) + len);
 154         else
 155                 len = -EINVAL;
 156         spin_unlock_irq(&trace_lock);
 157         kfree(event);
 158         return len;
 159 }
 160
 161 static void print_pte(unsigned long address)
 162 {
 163         int level;
 164         pte_t *pte = lookup_address(address, &level);
 165
 166         if (!pte) {
 167                 pr_err(NAME "Error in %s: no pte for page 0x%08lx\n",
 168                                                         __func__, address);
 169                 return;
 170         }
 171
 172         if (level == PG_LEVEL_2M) {
 173                 pr_emerg(NAME "4MB pages are not currently supported: "
 174                                                         "0x%08lx\n", address);
 175                 BUG();
 176         }
 177         pr_info(NAME "pte for 0x%lx: 0x%lx 0x%lx\n", address, pte_val(*pte),
 178                                                 pte_val(*pte) & _PAGE_PRESENT);
 179 }
 180
 181 /*
 182  * For some reason the pre/post pairs have been called in an
 183  * unmatched order. Report and die.
 184  */
 185 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
 186 {
 187         const struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 188         pr_emerg(NAME "unexpected fault for address: 0x%08lx, "
 189                                         "last fault for address: 0x%08lx\n",
 190                                         addr, my_reason->addr);
 191         print_pte(addr);
 192         print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip);
 193         print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip);
 194 #ifdef __i386__
 195         pr_emerg("eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
 196                         regs->ax, regs->bx, regs->cx, regs->dx);
 197         pr_emerg("esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
 198                         regs->si, regs->di, regs->bp, regs->sp);
 199 #else
 200         pr_emerg("rax: %016lx   rcx: %016lx   rdx: %016lx\n",
 201                                         regs->ax, regs->cx, regs->dx);
 202         pr_emerg("rsi: %016lx   rdi: %016lx   rbp: %016lx   rsp: %016lx\n",
 203                                 regs->si, regs->di, regs->bp, regs->sp);
 204 #endif
 205         put_cpu_var(pf_reason);
 206         BUG();
 207 }
 208
 209 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
 210                                                 unsigned long addr)
 211 {
 212         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 213         struct mm_io_header_rw *my_trace = &get_cpu_var(cpu_trace);
 214         const unsigned long instptr = instruction_pointer(regs);
 215         const enum reason_type type = get_ins_type(instptr);
 216
 217         /* it doesn't make sense to have more than one active trace per cpu */
 218         if (my_reason->active_traces)
 219                 die_kmmio_nesting_error(regs, addr);
 220         else
 221                 my_reason->active_traces++;
 222
 223         my_reason->type = type;
 224         my_reason->addr = addr;
 225         my_reason->ip = instptr;
 226
 227         my_trace->header.type = MMIO_MAGIC;
 228         my_trace->header.pid = 0;
 229         my_trace->header.data_len = sizeof(struct mm_io_rw);
 230         my_trace->rw.address = addr;
 231         /*
 232          * struct remap_trace *trace = p->user_data;
 233          * phys = addr - trace->probe.addr + trace->phys;
 234          */
 235
 236         /*
 237          * Only record the program counter when requested.
 238          * It may taint clean-room reverse engineering.
 239          */
 240         if (trace_pc)
 241                 my_trace->rw.pc = instptr;
 242         else
 243                 my_trace->rw.pc = 0;
 244
 245         record_timestamp(&my_trace->header);
 246
 247         switch (type) {
 248         case REG_READ:
 249                 my_trace->header.type |=
 250                         (MMIO_READ << MMIO_OPCODE_SHIFT) |
 251                         (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
 252                 break;
 253         case REG_WRITE:
 254                 my_trace->header.type |=
 255                         (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
 256                         (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
 257                 my_trace->rw.value = get_ins_reg_val(instptr, regs);
 258                 break;
 259         case IMM_WRITE:
 260                 my_trace->header.type |=
 261                         (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
 262                         (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
 263                 my_trace->rw.value = get_ins_imm_val(instptr);
 264                 break;
 265         default:
 266                 {
 267                         unsigned char *ip = (unsigned char *)instptr;
 268                         my_trace->header.type |=
 269                                         (MMIO_UNKNOWN_OP << MMIO_OPCODE_SHIFT);
 270                         my_trace->rw.value = (*ip) << 16 | *(ip + 1) << 8 |
 271                                                                 *(ip + 2);
 272                 }
 273         }
 274         put_cpu_var(cpu_trace);
 275         put_cpu_var(pf_reason);
 276 }
 277
 278 static void post(struct kmmio_probe *p, unsigned long condition,
 279                                                         struct pt_regs *regs)
 280 {
 281         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
 282         struct mm_io_header_rw *my_trace = &get_cpu_var(cpu_trace);
 283
 284         /* this should always return the active_trace count to 0 */
 285         my_reason->active_traces--;
 286         if (my_reason->active_traces) {
 287                 pr_emerg(NAME "unexpected post handler");
 288                 BUG();
 289         }
 290
 291         switch (my_reason->type) {
 292         case REG_READ:
 293                 my_trace->rw.value = get_ins_reg_val(my_reason->ip, regs);
 294                 break;
 295         default:
 296                 break;
 297         }
 298         relay_write(chan, my_trace, sizeof(*my_trace));
 299         put_cpu_var(cpu_trace);
 300         put_cpu_var(pf_reason);
 301 }
 302
 303 /*
 304  * subbuf_start() relay callback.
 305  *
 306  * Defined so that we know when events are dropped due to the buffer-full
 307  * condition.
 308  */
 309 static int subbuf_start_handler(struct rchan_buf *buf, void *subbuf,
 310                                         void *prev_subbuf, size_t prev_padding)
 311 {
 312         unsigned int cpu = buf->cpu;
 313         atomic_t *drop = &per_cpu(dropped, cpu);
 314         int count;
 315         if (relay_buf_full(buf)) {
 316                 if (atomic_inc_return(drop) == 1)
 317                         pr_err(NAME "cpu %d buffer full!\n", cpu);
 318                 return 0;
 319         }
 320         count = atomic_read(drop);
 321         if (count) {
 322                 pr_err(NAME "cpu %d buffer no longer full, missed %d events.\n",
 323                                                                 cpu, count);
 324                 atomic_sub(count, drop);
 325         }
 326
 327         return 1;
 328 }
 329
 330 static struct file_operations mmio_fops = {
 331         .owner = THIS_MODULE,
 332 };
 333
 334 /* file_create() callback.  Creates relay file in debugfs. */
 335 static struct dentry *create_buf_file_handler(const char *filename,
 336                                                 struct dentry *parent,
 337                                                 int mode,
 338                                                 struct rchan_buf *buf,
 339                                                 int *is_global)
 340 {
 341         struct dentry *buf_file;
 342
 343         mmio_fops.read = relay_file_operations.read;
 344         mmio_fops.open = relay_file_operations.open;
 345         mmio_fops.poll = relay_file_operations.poll;
 346         mmio_fops.mmap = relay_file_operations.mmap;
 347         mmio_fops.release = relay_file_operations.release;
 348         mmio_fops.splice_read = relay_file_operations.splice_read;
 349
 350         buf_file = debugfs_create_file(filename, mode, parent, buf,
 351                                                                 &mmio_fops);
 352
 353         return buf_file;
 354 }
 355
 356 /* file_remove() default callback.  Removes relay file in debugfs. */
 357 static int remove_buf_file_handler(struct dentry *dentry)
 358 {
 359         debugfs_remove(dentry);
 360         return 0;
 361 }
 362
 363 static struct rchan_callbacks relay_callbacks = {
 364         .subbuf_start = subbuf_start_handler,
 365         .create_buf_file = create_buf_file_handler,
 366         .remove_buf_file = remove_buf_file_handler,
 367 };
 368
 369 static void ioremap_trace_core(unsigned long offset, unsigned long size,
 370                                                         void __iomem *addr)
 371 {
 372         static atomic_t next_id;
 373         struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
 374         struct mm_io_header_map event = {
 375                 .header = {
 376                         .type = MMIO_MAGIC |
 377                                         (MMIO_PROBE << MMIO_OPCODE_SHIFT),
 378                         .sec = 0,
 379                         .nsec = 0,
 380                         .pid = 0,
 381                         .data_len = sizeof(struct mm_io_map)
 382                 },
 383                 .map = {
 384                         .phys = offset,
 385                         .addr = (unsigned long)addr,
 386                         .len  = size,
 387                         .pc   = 0
 388                 }
 389         };
 390         record_timestamp(&event.header);
 391
 392         if (!trace) {
 393                 pr_err(NAME "kmalloc failed in ioremap\n");
 394                 return;
 395         }
 396
 397         *trace = (struct remap_trace) {
 398                 .probe = {
 399                         .addr = (unsigned long)addr,
 400                         .len = size,
 401                         .pre_handler = pre,
 402                         .post_handler = post,
 403                         .user_data = trace
 404                 },
 405                 .phys = offset,
 406                 .id = atomic_inc_return(&next_id)
 407         };
 408
 409         spin_lock_irq(&trace_lock);
 410         if (!is_enabled())
 411                 goto not_enabled;
 412
 413         relay_write(chan, &event, sizeof(event));
 414         list_add_tail(&trace->list, &trace_list);
 415         if (!nommiotrace)
 416                 register_kmmio_probe(&trace->probe);
 417
 418 not_enabled:
 419         spin_unlock_irq(&trace_lock);
 420 }
 421
 422 void
 423 mmiotrace_ioremap(unsigned long offset, unsigned long size, void __iomem *addr)
 424 {
 425         if (!is_enabled()) /* recheck and proper locking in *_core() */
 426                 return;
 427
 428         pr_debug(NAME "ioremap_*(0x%lx, 0x%lx) = %p\n", offset, size, addr);
 429         if ((filter_offset) && (offset != filter_offset))
 430                 return;
 431         ioremap_trace_core(offset, size, addr);
 432 }
 433
 434 static void iounmap_trace_core(volatile void __iomem *addr)
 435 {
 436         struct mm_io_header_map event = {
 437                 .header = {
 438                         .type = MMIO_MAGIC |
 439                                 (MMIO_UNPROBE << MMIO_OPCODE_SHIFT),
 440                         .sec = 0,
 441                         .nsec = 0,
 442                         .pid = 0,
 443                         .data_len = sizeof(struct mm_io_map)
 444                 },
 445                 .map = {
 446                         .phys = 0,
 447                         .addr = (unsigned long)addr,
 448                         .len  = 0,
 449                         .pc   = 0
 450                 }
 451         };
 452         struct remap_trace *trace;
 453         struct remap_trace *tmp;
 454         struct remap_trace *found_trace = NULL;
 455
 456         pr_debug(NAME "Unmapping %p.\n", addr);
 457         record_timestamp(&event.header);
 458
 459         spin_lock_irq(&trace_lock);
 460         if (!is_enabled())
 461                 goto not_enabled;
 462
 463         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
 464                 if ((unsigned long)addr == trace->probe.addr) {
 465                         if (!nommiotrace)
 466                                 unregister_kmmio_probe(&trace->probe);
 467                         list_del(&trace->list);
 468                         found_trace = trace;
 469                         break;
 470                 }
 471         }
 472         relay_write(chan, &event, sizeof(event));
 473
 474 not_enabled:
 475         spin_unlock_irq(&trace_lock);
 476         if (found_trace) {
 477                 synchronize_rcu(); /* unregister_kmmio_probe() requirement */
 478                 kfree(found_trace);
 479         }
 480 }
 481
 482 void mmiotrace_iounmap(volatile void __iomem *addr)
 483 {
 484         might_sleep();
 485         if (is_enabled()) /* recheck and proper locking in *_core() */
 486                 iounmap_trace_core(addr);
 487 }
 488
 489 static void clear_trace_list(void)
 490 {
 491         struct remap_trace *trace;
 492         struct remap_trace *tmp;
 493
 494         /*
 495          * No locking required, because the caller ensures we are in a
 496          * critical section via mutex, and is_enabled() is false,
 497          * i.e. nothing can traverse or modify this list.
 498          * Caller also ensures is_enabled() cannot change.
 499          */
 500         list_for_each_entry(trace, &trace_list, list) {
 501                 pr_notice(NAME "purging non-iounmapped "
 502                                         "trace @0x%08lx, size 0x%lx.\n",
 503                                         trace->probe.addr, trace->probe.len);
 504                 if (!nommiotrace)
 505                         unregister_kmmio_probe(&trace->probe);
 506         }
 507         synchronize_rcu(); /* unregister_kmmio_probe() requirement */
 508
 509         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
 510                 list_del(&trace->list);
 511                 kfree(trace);
 512         }
 513 }
 514
 515 static ssize_t read_enabled_file_bool(struct file *file,
 516                 char __user *user_buf, size_t count, loff_t *ppos)
 517 {
 518         char buf[3];
 519
 520         if (is_enabled())
 521                 buf[0] = '1';
 522         else
 523                 buf[0] = '0';
 524         buf[1] = '\n';
 525         buf[2] = '\0';
 526         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
 527 }
 528
 529 static void enable_mmiotrace(void);
 530 static void disable_mmiotrace(void);
 531
 532 static ssize_t write_enabled_file_bool(struct file *file,
 533                 const char __user *user_buf, size_t count, loff_t *ppos)
 534 {
 535         char buf[32];
 536         int buf_size = min(count, (sizeof(buf)-1));
 537
 538         if (copy_from_user(buf, user_buf, buf_size))
 539                 return -EFAULT;
 540
 541         switch (buf[0]) {
 542         case 'y':
 543         case 'Y':
 544         case '1':
 545                 enable_mmiotrace();
 546                 break;
 547         case 'n':
 548         case 'N':
 549         case '0':
 550                 disable_mmiotrace();
 551                 break;
 552         }
 553
 554         return count;
 555 }
 556
 557 /* this ripped from kernel/kprobes.c */
 558 static struct file_operations fops_enabled = {
 559         .owner =        THIS_MODULE,
 560         .read =         read_enabled_file_bool,
 561         .write =        write_enabled_file_bool
 562 };
 563
 564 static struct file_operations fops_marker = {
 565         .owner =        THIS_MODULE,
 566         .write =        write_marker
 567 };
 568
 569 static void enable_mmiotrace(void)
 570 {
 571         mutex_lock(&mmiotrace_mutex);
 572         if (is_enabled())
 573                 goto out;
 574
 575         chan = relay_open("cpu", dir, subbuf_size, n_subbufs,
 576                                                 &relay_callbacks, NULL);
 577         if (!chan) {
 578                 pr_err(NAME "relay app channel creation failed.\n");
 579                 goto out;
 580         }
 581
 582         reference_kmmio();
 583
 584         marker_file = debugfs_create_file("marker", 0660, dir, NULL,
 585                                                                 &fops_marker);
 586         if (!marker_file)
 587                 pr_err(NAME "marker file creation failed.\n");
 588
 589         if (nommiotrace)
 590                 pr_info(NAME "MMIO tracing disabled.\n");
 591         if (ISA_trace)
 592                 pr_warning(NAME "Warning! low ISA range will be traced.\n");
 593         spin_lock_irq(&trace_lock);
 594         atomic_inc(&mmiotrace_enabled);
 595         spin_unlock_irq(&trace_lock);
 596         pr_info(NAME "enabled.\n");
 597 out:
 598         mutex_unlock(&mmiotrace_mutex);
 599 }
 600
 601 static void disable_mmiotrace(void)
 602 {
 603         mutex_lock(&mmiotrace_mutex);
 604         if (!is_enabled())
 605                 goto out;
 606
 607         spin_lock_irq(&trace_lock);
 608         atomic_dec(&mmiotrace_enabled);
 609         BUG_ON(is_enabled());
 610         spin_unlock_irq(&trace_lock);
 611
 612         clear_trace_list(); /* guarantees: no more kmmio callbacks */
 613         unreference_kmmio();
 614         if (marker_file) {
 615                 debugfs_remove(marker_file);
 616                 marker_file = NULL;
 617         }
 618         if (chan) {
 619                 relay_close(chan);
 620                 chan = NULL;
 621         }
 622
 623         pr_info(NAME "disabled.\n");
 624 out:
 625         mutex_unlock(&mmiotrace_mutex);
 626 }
 627
 628 static int __init init(void)
 629 {
 630         pr_debug(NAME "load...\n");
 631         if (n_subbufs < 2)
 632                 return -EINVAL;
 633
 634         dir = debugfs_create_dir(APP_DIR, NULL);
 635         if (!dir) {
 636                 pr_err(NAME "Couldn't create relay app directory.\n");
 637                 return -ENOMEM;
 638         }
 639
 640         enabled_file = debugfs_create_file("enabled", 0600, dir, NULL,
 641                                                                 &fops_enabled);
 642         if (!enabled_file) {
 643                 pr_err(NAME "Couldn't create enabled file.\n");
 644                 debugfs_remove(dir);
 645                 return -ENOMEM;
 646         }
 647
 648         if (enable_now)
 649                 enable_mmiotrace();
 650
 651         return 0;
 652 }
 653
 654 static void __exit cleanup(void)
 655 {
 656         pr_debug(NAME "unload...\n");
 657         if (enabled_file)
 658                 debugfs_remove(enabled_file);
 659         disable_mmiotrace();
 660         if (dir)
 661                 debugfs_remove(dir);
 662 }
 663
 664 module_init(init);
 665 module_exit(cleanup);
 666 MODULE_LICENSE("GPL");