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.
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.
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.
16 * Copyright (C) IBM Corporation, 2005
17 * Jeff Muizelaar, 2006, 2007
18 * Pekka Paalanen, 2008 <pq@iki.fi>
20 * Derived from the read-mod example from relay-examples by Tom Zanussi.
22 #include <linux/module.h>
23 #include <linux/relay.h>
24 #include <linux/debugfs.h>
25 #include <linux/proc_fs.h>
27 #include <linux/version.h>
28 #include <linux/kallsyms.h>
29 #include <asm/pgtable.h>
30 #include <linux/mmiotrace.h>
31 #include <asm/e820.h> /* for ISA_START_ADDRESS */
32 #include <asm/atomic.h>
37 /* This app's relay channel files will appear in /debug/mmio-trace */
38 #define APP_DIR "mmio-trace"
39 /* the marker injection file in /proc */
40 #define MARKER_FILE "mmio-marker"
42 #define MODULE_NAME "mmiotrace"
47 enum reason_type type;
51 /* Accessed per-cpu. */
52 static struct trap_reason pf_reason[NR_CPUS];
53 static struct mm_io_header_rw cpu_trace[NR_CPUS];
55 /* Access to this is not per-cpu. */
56 static atomic_t dropped[NR_CPUS];
58 static struct file_operations mmio_fops = {
62 static const size_t subbuf_size = 256*1024;
63 static struct rchan *chan;
64 static struct dentry *dir;
65 static struct proc_dir_entry *proc_marker_file;
67 /* module parameters */
68 static unsigned int n_subbufs = 32*4;
69 static unsigned long filter_offset;
70 static int nommiotrace;
74 module_param(n_subbufs, uint, 0);
75 module_param(filter_offset, ulong, 0);
76 module_param(nommiotrace, bool, 0);
77 module_param(ISA_trace, bool, 0);
78 module_param(trace_pc, bool, 0);
80 MODULE_PARM_DESC(n_subbufs, "Number of 256kB buffers, default 128.");
81 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
82 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
83 MODULE_PARM_DESC(ISA_trace, "Do not exclude the low ISA range.");
84 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
86 static void record_timestamp(struct mm_io_header *header)
91 header->sec = now.tv_sec;
92 header->nsec = now.tv_nsec;
96 * Write callback for the /proc entry:
97 * Read a marker and write it to the mmio trace log
99 static int write_marker(struct file *file, const char __user *buffer,
100 unsigned long count, void *data)
103 struct mm_io_header *headp;
104 int len = (count > 65535) ? 65535 : count;
106 event = kzalloc(sizeof(*headp) + len, GFP_KERNEL);
110 headp = (struct mm_io_header *)event;
111 headp->type = MMIO_MAGIC | (MMIO_MARKER << MMIO_OPCODE_SHIFT);
112 headp->data_len = len;
113 record_timestamp(headp);
115 if (copy_from_user(event + sizeof(*headp), buffer, len)) {
120 relay_write(chan, event, sizeof(*headp) + len);
125 static void print_pte(unsigned long address)
128 pte_t *pte = lookup_address(address, &level);
131 printk(KERN_ERR "Error in %s: no pte for page 0x%08lx\n",
132 __FUNCTION__, address);
136 if (level == PG_LEVEL_2M) {
137 printk(KERN_EMERG MODULE_NAME ": 4MB pages are not "
138 "currently supported: %lx\n",
142 printk(KERN_DEBUG MODULE_NAME ": pte for 0x%lx: 0x%lx 0x%lx\n",
143 address, pte_val(*pte),
144 pte_val(*pte) & _PAGE_PRESENT);
148 * For some reason the pre/post pairs have been called in an
149 * unmatched order. Report and die.
151 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
153 const unsigned long cpu = smp_processor_id();
154 printk(KERN_EMERG MODULE_NAME ": unexpected fault for address: %lx, "
155 "last fault for address: %lx\n",
156 addr, pf_reason[cpu].addr);
159 print_symbol(KERN_EMERG "faulting EIP is at %s\n", regs->ip);
160 print_symbol(KERN_EMERG "last faulting EIP was at %s\n",
163 "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
164 regs->ax, regs->bx, regs->cx, regs->dx);
166 "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
167 regs->si, regs->di, regs->bp, regs->sp);
169 print_symbol(KERN_EMERG "faulting RIP is at %s\n", regs->ip);
170 print_symbol(KERN_EMERG "last faulting RIP was at %s\n",
172 printk(KERN_EMERG "rax: %016lx rcx: %016lx rdx: %016lx\n",
173 regs->ax, regs->cx, regs->dx);
174 printk(KERN_EMERG "rsi: %016lx rdi: %016lx "
175 "rbp: %016lx rsp: %016lx\n",
176 regs->si, regs->di, regs->bp, regs->sp);
181 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
184 const unsigned long cpu = smp_processor_id();
185 const unsigned long instptr = instruction_pointer(regs);
186 const enum reason_type type = get_ins_type(instptr);
188 /* it doesn't make sense to have more than one active trace per cpu */
189 if (pf_reason[cpu].active_traces)
190 die_kmmio_nesting_error(regs, addr);
192 pf_reason[cpu].active_traces++;
194 pf_reason[cpu].type = type;
195 pf_reason[cpu].addr = addr;
196 pf_reason[cpu].ip = instptr;
198 cpu_trace[cpu].header.type = MMIO_MAGIC;
199 cpu_trace[cpu].header.pid = 0;
200 cpu_trace[cpu].header.data_len = sizeof(struct mm_io_rw);
201 cpu_trace[cpu].rw.address = addr;
204 * Only record the program counter when requested.
205 * It may taint clean-room reverse engineering.
208 cpu_trace[cpu].rw.pc = instptr;
210 cpu_trace[cpu].rw.pc = 0;
212 record_timestamp(&cpu_trace[cpu].header);
216 cpu_trace[cpu].header.type |=
217 (MMIO_READ << MMIO_OPCODE_SHIFT) |
218 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
221 cpu_trace[cpu].header.type |=
222 (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
223 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
224 cpu_trace[cpu].rw.value = get_ins_reg_val(instptr, regs);
227 cpu_trace[cpu].header.type |=
228 (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
229 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
230 cpu_trace[cpu].rw.value = get_ins_imm_val(instptr);
234 unsigned char *ip = (unsigned char *)instptr;
235 cpu_trace[cpu].header.type |=
236 (MMIO_UNKNOWN_OP << MMIO_OPCODE_SHIFT);
237 cpu_trace[cpu].rw.value = (*ip) << 16 |
244 static void post(struct kmmio_probe *p, unsigned long condition,
245 struct pt_regs *regs)
247 const unsigned long cpu = smp_processor_id();
249 /* this should always return the active_trace count to 0 */
250 pf_reason[cpu].active_traces--;
251 if (pf_reason[cpu].active_traces) {
252 printk(KERN_EMERG MODULE_NAME ": unexpected post handler");
256 switch (pf_reason[cpu].type) {
258 cpu_trace[cpu].rw.value = get_ins_reg_val(pf_reason[cpu].ip,
264 relay_write(chan, &cpu_trace[cpu], sizeof(struct mm_io_header_rw));
268 * subbuf_start() relay callback.
270 * Defined so that we know when events are dropped due to the buffer-full
273 static int subbuf_start_handler(struct rchan_buf *buf, void *subbuf,
274 void *prev_subbuf, size_t prev_padding)
276 unsigned int cpu = buf->cpu;
277 atomic_t *drop = &dropped[cpu];
279 if (relay_buf_full(buf)) {
280 if (atomic_inc_return(drop) == 1) {
281 printk(KERN_ERR MODULE_NAME ": cpu %d buffer full!\n",
285 } else if ((count = atomic_read(drop))) {
286 printk(KERN_ERR MODULE_NAME
287 ": cpu %d buffer no longer full, "
288 "missed %d events.\n",
290 atomic_sub(count, drop);
296 /* file_create() callback. Creates relay file in debugfs. */
297 static struct dentry *create_buf_file_handler(const char *filename,
298 struct dentry *parent,
300 struct rchan_buf *buf,
303 struct dentry *buf_file;
305 mmio_fops.read = relay_file_operations.read;
306 mmio_fops.open = relay_file_operations.open;
307 mmio_fops.poll = relay_file_operations.poll;
308 mmio_fops.mmap = relay_file_operations.mmap;
309 mmio_fops.release = relay_file_operations.release;
310 mmio_fops.splice_read = relay_file_operations.splice_read;
312 buf_file = debugfs_create_file(filename, mode, parent, buf,
318 /* file_remove() default callback. Removes relay file in debugfs. */
319 static int remove_buf_file_handler(struct dentry *dentry)
321 debugfs_remove(dentry);
325 static struct rchan_callbacks relay_callbacks = {
326 .subbuf_start = subbuf_start_handler,
327 .create_buf_file = create_buf_file_handler,
328 .remove_buf_file = remove_buf_file_handler,
332 * create_channel - creates channel /debug/APP_DIR/cpuXXX
333 * Returns channel on success, NULL otherwise
335 static struct rchan *create_channel(unsigned size, unsigned n)
337 return relay_open("cpu", dir, size, n, &relay_callbacks, NULL);
340 /* destroy_channel - destroys channel /debug/APP_DIR/cpuXXX */
341 static void destroy_channel(void)
350 struct list_head list;
351 struct kmmio_probe probe;
353 static LIST_HEAD(trace_list);
354 static DEFINE_SPINLOCK(trace_list_lock);
356 static void do_ioremap_trace_core(unsigned long offset, unsigned long size,
359 struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
360 struct mm_io_header_map event = {
363 (MMIO_PROBE << MMIO_OPCODE_SHIFT),
367 .data_len = sizeof(struct mm_io_map)
371 .addr = (unsigned long)addr,
376 record_timestamp(&event.header);
378 *trace = (struct remap_trace) {
380 .addr = (unsigned long)addr,
383 .post_handler = post,
387 relay_write(chan, &event, sizeof(event));
388 spin_lock(&trace_list_lock);
389 list_add_tail(&trace->list, &trace_list);
390 spin_unlock(&trace_list_lock);
392 register_kmmio_probe(&trace->probe);
395 static void ioremap_trace_core(unsigned long offset, unsigned long size,
398 if ((filter_offset) && (offset != filter_offset))
401 /* Don't trace the low PCI/ISA area, it's always mapped.. */
402 if (!ISA_trace && (offset < ISA_END_ADDRESS) &&
403 (offset + size > ISA_START_ADDRESS)) {
404 printk(KERN_NOTICE MODULE_NAME ": Ignoring map of low "
405 "PCI/ISA area (0x%lx-0x%lx)\n",
406 offset, offset + size);
409 do_ioremap_trace_core(offset, size, addr);
412 void __iomem *ioremap_cache_trace(unsigned long offset, unsigned long size)
414 void __iomem *p = ioremap_cache(offset, size);
415 printk(KERN_DEBUG MODULE_NAME ": ioremap_cache(0x%lx, 0x%lx) = %p\n",
417 ioremap_trace_core(offset, size, p);
420 EXPORT_SYMBOL(ioremap_cache_trace);
422 void __iomem *ioremap_nocache_trace(unsigned long offset, unsigned long size)
424 void __iomem *p = ioremap_nocache(offset, size);
425 printk(KERN_DEBUG MODULE_NAME ": ioremap_nocache(0x%lx, 0x%lx) = %p\n",
427 ioremap_trace_core(offset, size, p);
430 EXPORT_SYMBOL(ioremap_nocache_trace);
432 void iounmap_trace(volatile void __iomem *addr)
434 struct mm_io_header_map event = {
437 (MMIO_UNPROBE << MMIO_OPCODE_SHIFT),
441 .data_len = sizeof(struct mm_io_map)
445 .addr = (unsigned long)addr,
450 struct remap_trace *trace;
451 struct remap_trace *tmp;
452 printk(KERN_DEBUG MODULE_NAME ": Unmapping %p.\n", addr);
453 record_timestamp(&event.header);
455 spin_lock(&trace_list_lock);
456 list_for_each_entry_safe(trace, tmp, &trace_list, list) {
457 if ((unsigned long)addr == trace->probe.addr) {
459 unregister_kmmio_probe(&trace->probe);
460 list_del(&trace->list);
465 spin_unlock(&trace_list_lock);
466 relay_write(chan, &event, sizeof(event));
469 EXPORT_SYMBOL(iounmap_trace);
471 static void clear_trace_list(void)
473 struct remap_trace *trace;
474 struct remap_trace *tmp;
476 spin_lock(&trace_list_lock);
477 list_for_each_entry_safe(trace, tmp, &trace_list, list) {
478 printk(KERN_WARNING MODULE_NAME ": purging non-iounmapped "
479 "trace @0x%08lx, size 0x%lx.\n",
480 trace->probe.addr, trace->probe.len);
482 unregister_kmmio_probe(&trace->probe);
483 list_del(&trace->list);
487 spin_unlock(&trace_list_lock);
490 static int __init init(void)
495 dir = debugfs_create_dir(APP_DIR, NULL);
497 printk(KERN_ERR MODULE_NAME
498 ": Couldn't create relay app directory.\n");
502 chan = create_channel(subbuf_size, n_subbufs);
505 printk(KERN_ERR MODULE_NAME
506 ": relay app channel creation failed\n");
512 proc_marker_file = create_proc_entry(MARKER_FILE, 0, NULL);
513 if (proc_marker_file)
514 proc_marker_file->write_proc = write_marker;
516 printk(KERN_DEBUG MODULE_NAME ": loaded.\n");
518 printk(KERN_DEBUG MODULE_NAME ": MMIO tracing disabled.\n");
520 printk(KERN_WARNING MODULE_NAME
521 ": Warning! low ISA range will be traced.\n");
525 static void __exit cleanup(void)
527 printk(KERN_DEBUG MODULE_NAME ": unload...\n");
530 remove_proc_entry(MARKER_FILE, NULL);
537 module_exit(cleanup);
538 MODULE_LICENSE("GPL");