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 */
36 /* This app's relay channel files will appear in /debug/mmio-trace */
37 #define APP_DIR "mmio-trace"
38 /* the marker injection file in /proc */
39 #define MARKER_FILE "mmio-marker"
41 #define MODULE_NAME "mmiotrace"
46 enum reason_type type;
50 static struct trap_reason pf_reason[NR_CPUS];
51 static struct mm_io_header_rw cpu_trace[NR_CPUS];
53 static struct file_operations mmio_fops = {
57 static const size_t subbuf_size = 256*1024;
58 static struct rchan *chan;
59 static struct dentry *dir;
60 static int suspended; /* XXX should this be per cpu? */
61 static struct proc_dir_entry *proc_marker_file;
63 /* module parameters */
64 static unsigned int n_subbufs = 32*4;
65 static unsigned long filter_offset;
66 static int nommiotrace;
70 module_param(n_subbufs, uint, 0);
71 module_param(filter_offset, ulong, 0);
72 module_param(nommiotrace, bool, 0);
73 module_param(ISA_trace, bool, 0);
74 module_param(trace_pc, bool, 0);
76 MODULE_PARM_DESC(n_subbufs, "Number of 256kB buffers, default 128.");
77 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
78 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
79 MODULE_PARM_DESC(ISA_trace, "Do not exclude the low ISA range.");
80 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
82 static void record_timestamp(struct mm_io_header *header)
87 header->sec = now.tv_sec;
88 header->nsec = now.tv_nsec;
92 * Write callback for the /proc entry:
93 * Read a marker and write it to the mmio trace log
95 static int write_marker(struct file *file, const char __user *buffer,
96 unsigned long count, void *data)
99 struct mm_io_header *headp;
100 int len = (count > 65535) ? 65535 : count;
102 event = kzalloc(sizeof(*headp) + len, GFP_KERNEL);
106 headp = (struct mm_io_header *)event;
107 headp->type = MMIO_MAGIC | (MMIO_MARKER << MMIO_OPCODE_SHIFT);
108 headp->data_len = len;
109 record_timestamp(headp);
111 if (copy_from_user(event + sizeof(*headp), buffer, len)) {
116 relay_write(chan, event, sizeof(*headp) + len);
121 static void print_pte(unsigned long address)
123 pgd_t *pgd = pgd_offset_k(address);
124 pud_t *pud = pud_offset(pgd, address);
125 pmd_t *pmd = pmd_offset(pud, address);
126 if (pmd_large(*pmd)) {
127 printk(KERN_EMERG MODULE_NAME ": 4MB pages are not "
128 "currently supported: %lx\n",
132 printk(KERN_DEBUG MODULE_NAME ": pte for 0x%lx: 0x%lx 0x%lx\n",
134 pte_val(*pte_offset_kernel(pmd, address)),
135 pte_val(*pte_offset_kernel(pmd, address)) & _PAGE_PRESENT);
139 * For some reason the pre/post pairs have been called in an
140 * unmatched order. Report and die.
142 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
144 const unsigned long cpu = smp_processor_id();
145 printk(KERN_EMERG MODULE_NAME ": unexpected fault for address: %lx, "
146 "last fault for address: %lx\n",
147 addr, pf_reason[cpu].addr);
150 print_symbol(KERN_EMERG "faulting EIP is at %s\n", regs->ip);
151 print_symbol(KERN_EMERG "last faulting EIP was at %s\n",
154 "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
155 regs->ax, regs->bx, regs->cx, regs->dx);
157 "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
158 regs->si, regs->di, regs->bp, regs->sp);
160 print_symbol(KERN_EMERG "faulting RIP is at %s\n", regs->ip);
161 print_symbol(KERN_EMERG "last faulting RIP was at %s\n",
163 printk(KERN_EMERG "rax: %016lx rcx: %016lx rdx: %016lx\n",
164 regs->ax, regs->cx, regs->dx);
165 printk(KERN_EMERG "rsi: %016lx rdi: %016lx "
166 "rbp: %016lx rsp: %016lx\n",
167 regs->si, regs->di, regs->bp, regs->sp);
172 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
175 const unsigned long cpu = smp_processor_id();
176 const unsigned long instptr = instruction_pointer(regs);
177 const enum reason_type type = get_ins_type(instptr);
179 /* it doesn't make sense to have more than one active trace per cpu */
180 if (pf_reason[cpu].active_traces)
181 die_kmmio_nesting_error(regs, addr);
183 pf_reason[cpu].active_traces++;
185 pf_reason[cpu].type = type;
186 pf_reason[cpu].addr = addr;
187 pf_reason[cpu].ip = instptr;
189 cpu_trace[cpu].header.type = MMIO_MAGIC;
190 cpu_trace[cpu].header.pid = 0;
191 cpu_trace[cpu].header.data_len = sizeof(struct mm_io_rw);
192 cpu_trace[cpu].rw.address = addr;
195 * Only record the program counter when requested.
196 * It may taint clean-room reverse engineering.
199 cpu_trace[cpu].rw.pc = instptr;
201 cpu_trace[cpu].rw.pc = 0;
203 record_timestamp(&cpu_trace[cpu].header);
207 cpu_trace[cpu].header.type |=
208 (MMIO_READ << MMIO_OPCODE_SHIFT) |
209 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
212 cpu_trace[cpu].header.type |=
213 (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
214 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
215 cpu_trace[cpu].rw.value = get_ins_reg_val(instptr, regs);
218 cpu_trace[cpu].header.type |=
219 (MMIO_WRITE << MMIO_OPCODE_SHIFT) |
220 (get_ins_mem_width(instptr) << MMIO_WIDTH_SHIFT);
221 cpu_trace[cpu].rw.value = get_ins_imm_val(instptr);
225 unsigned char *ip = (unsigned char *)instptr;
226 cpu_trace[cpu].header.type |=
227 (MMIO_UNKNOWN_OP << MMIO_OPCODE_SHIFT);
228 cpu_trace[cpu].rw.value = (*ip) << 16 |
235 static void post(struct kmmio_probe *p, unsigned long condition,
236 struct pt_regs *regs)
238 const unsigned long cpu = smp_processor_id();
240 /* this should always return the active_trace count to 0 */
241 pf_reason[cpu].active_traces--;
242 if (pf_reason[cpu].active_traces) {
243 printk(KERN_EMERG MODULE_NAME ": unexpected post handler");
247 switch (pf_reason[cpu].type) {
249 cpu_trace[cpu].rw.value = get_ins_reg_val(pf_reason[cpu].ip,
255 relay_write(chan, &cpu_trace[cpu], sizeof(struct mm_io_header_rw));
259 * subbuf_start() relay callback.
261 * Defined so that we know when events are dropped due to the buffer-full
264 static int subbuf_start_handler(struct rchan_buf *buf, void *subbuf,
265 void *prev_subbuf, size_t prev_padding)
267 if (relay_buf_full(buf)) {
270 printk(KERN_ERR MODULE_NAME
271 ": cpu %d buffer full!!!\n",
275 } else if (suspended) {
277 printk(KERN_ERR MODULE_NAME
278 ": cpu %d buffer no longer full.\n",
285 /* file_create() callback. Creates relay file in debugfs. */
286 static struct dentry *create_buf_file_handler(const char *filename,
287 struct dentry *parent,
289 struct rchan_buf *buf,
292 struct dentry *buf_file;
294 mmio_fops.read = relay_file_operations.read;
295 mmio_fops.open = relay_file_operations.open;
296 mmio_fops.poll = relay_file_operations.poll;
297 mmio_fops.mmap = relay_file_operations.mmap;
298 mmio_fops.release = relay_file_operations.release;
299 mmio_fops.splice_read = relay_file_operations.splice_read;
301 buf_file = debugfs_create_file(filename, mode, parent, buf,
307 /* file_remove() default callback. Removes relay file in debugfs. */
308 static int remove_buf_file_handler(struct dentry *dentry)
310 debugfs_remove(dentry);
314 static struct rchan_callbacks relay_callbacks = {
315 .subbuf_start = subbuf_start_handler,
316 .create_buf_file = create_buf_file_handler,
317 .remove_buf_file = remove_buf_file_handler,
321 * create_channel - creates channel /debug/APP_DIR/cpuXXX
322 * Returns channel on success, NULL otherwise
324 static struct rchan *create_channel(unsigned size, unsigned n)
326 return relay_open("cpu", dir, size, n, &relay_callbacks, NULL);
329 /* destroy_channel - destroys channel /debug/APP_DIR/cpuXXX */
330 static void destroy_channel(void)
339 struct list_head list;
340 struct kmmio_probe probe;
342 static LIST_HEAD(trace_list);
343 static DEFINE_SPINLOCK(trace_list_lock);
345 static void do_ioremap_trace_core(unsigned long offset, unsigned long size,
348 struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
349 struct mm_io_header_map event = {
352 (MMIO_PROBE << MMIO_OPCODE_SHIFT),
356 .data_len = sizeof(struct mm_io_map)
360 .addr = (unsigned long)addr,
365 record_timestamp(&event.header);
367 *trace = (struct remap_trace) {
369 .addr = (unsigned long)addr,
372 .post_handler = post,
376 relay_write(chan, &event, sizeof(event));
377 spin_lock(&trace_list_lock);
378 list_add_tail(&trace->list, &trace_list);
379 spin_unlock(&trace_list_lock);
381 register_kmmio_probe(&trace->probe);
384 static void ioremap_trace_core(unsigned long offset, unsigned long size,
387 if ((filter_offset) && (offset != filter_offset))
390 /* Don't trace the low PCI/ISA area, it's always mapped.. */
391 if (!ISA_trace && (offset < ISA_END_ADDRESS) &&
392 (offset + size > ISA_START_ADDRESS)) {
393 printk(KERN_NOTICE MODULE_NAME ": Ignoring map of low "
394 "PCI/ISA area (0x%lx-0x%lx)\n",
395 offset, offset + size);
398 do_ioremap_trace_core(offset, size, addr);
401 void __iomem *ioremap_cache_trace(unsigned long offset, unsigned long size)
403 void __iomem *p = ioremap_cache(offset, size);
404 printk(KERN_DEBUG MODULE_NAME ": ioremap_cache(0x%lx, 0x%lx) = %p\n",
406 ioremap_trace_core(offset, size, p);
409 EXPORT_SYMBOL(ioremap_cache_trace);
411 void __iomem *ioremap_nocache_trace(unsigned long offset, unsigned long size)
413 void __iomem *p = ioremap_nocache(offset, size);
414 printk(KERN_DEBUG MODULE_NAME ": ioremap_nocache(0x%lx, 0x%lx) = %p\n",
416 ioremap_trace_core(offset, size, p);
419 EXPORT_SYMBOL(ioremap_nocache_trace);
421 void iounmap_trace(volatile void __iomem *addr)
423 struct mm_io_header_map event = {
426 (MMIO_UNPROBE << MMIO_OPCODE_SHIFT),
430 .data_len = sizeof(struct mm_io_map)
434 .addr = (unsigned long)addr,
439 struct remap_trace *trace;
440 struct remap_trace *tmp;
441 printk(KERN_DEBUG MODULE_NAME ": Unmapping %p.\n", addr);
442 record_timestamp(&event.header);
444 spin_lock(&trace_list_lock);
445 list_for_each_entry_safe(trace, tmp, &trace_list, list) {
446 if ((unsigned long)addr == trace->probe.addr) {
448 unregister_kmmio_probe(&trace->probe);
449 list_del(&trace->list);
454 spin_unlock(&trace_list_lock);
455 relay_write(chan, &event, sizeof(event));
458 EXPORT_SYMBOL(iounmap_trace);
460 static void clear_trace_list(void)
462 struct remap_trace *trace;
463 struct remap_trace *tmp;
465 spin_lock(&trace_list_lock);
466 list_for_each_entry_safe(trace, tmp, &trace_list, list) {
467 printk(KERN_WARNING MODULE_NAME ": purging non-iounmapped "
468 "trace @0x%08lx, size 0x%lx.\n",
469 trace->probe.addr, trace->probe.len);
471 unregister_kmmio_probe(&trace->probe);
472 list_del(&trace->list);
476 spin_unlock(&trace_list_lock);
479 static int __init init(void)
484 dir = debugfs_create_dir(APP_DIR, NULL);
486 printk(KERN_ERR MODULE_NAME
487 ": Couldn't create relay app directory.\n");
491 chan = create_channel(subbuf_size, n_subbufs);
494 printk(KERN_ERR MODULE_NAME
495 ": relay app channel creation failed\n");
501 proc_marker_file = create_proc_entry(MARKER_FILE, 0, NULL);
502 if (proc_marker_file)
503 proc_marker_file->write_proc = write_marker;
505 printk(KERN_DEBUG MODULE_NAME ": loaded.\n");
507 printk(KERN_DEBUG MODULE_NAME ": MMIO tracing disabled.\n");
509 printk(KERN_WARNING MODULE_NAME
510 ": Warning! low ISA range will be traced.\n");
514 static void __exit cleanup(void)
516 printk(KERN_DEBUG MODULE_NAME ": unload...\n");
519 remove_proc_entry(MARKER_FILE, NULL);
526 module_exit(cleanup);
527 MODULE_LICENSE("GPL");