ftrace: mmiotrace, updates
[linux-2.6.git] / 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/debugfs.h>
26 #include <linux/uaccess.h>
27 #include <asm/io.h>
28 #include <linux/version.h>
29 #include <linux/kallsyms.h>
30 #include <asm/pgtable.h>
31 #include <linux/mmiotrace.h>
32 #include <asm/e820.h> /* for ISA_START_ADDRESS */
33 #include <asm/atomic.h>
34 #include <linux/percpu.h>
35
36 #include "pf_in.h"
37
38 #define NAME "mmiotrace: "
39
40 struct trap_reason {
41         unsigned long addr;
42         unsigned long ip;
43         enum reason_type type;
44         int active_traces;
45 };
46
47 struct remap_trace {
48         struct list_head list;
49         struct kmmio_probe probe;
50         unsigned long phys;
51         unsigned long id;
52 };
53
54 /* Accessed per-cpu. */
55 static DEFINE_PER_CPU(struct trap_reason, pf_reason);
56 static DEFINE_PER_CPU(struct mmiotrace_rw, cpu_trace);
57
58 #if 0 /* XXX: no way gather this info anymore */
59 /* Access to this is not per-cpu. */
60 static DEFINE_PER_CPU(atomic_t, dropped);
61 #endif
62
63 static struct dentry *marker_file;
64
65 static DEFINE_MUTEX(mmiotrace_mutex);
66 static DEFINE_SPINLOCK(trace_lock);
67 static atomic_t mmiotrace_enabled;
68 static LIST_HEAD(trace_list);           /* struct remap_trace */
69
70 /*
71  * Locking in this file:
72  * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections.
73  * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex
74  *   and trace_lock.
75  * - Routines depending on is_enabled() must take trace_lock.
76  * - trace_list users must hold trace_lock.
77  * - is_enabled() guarantees that mmio_trace_record is allowed.
78  * - pre/post callbacks assume the effect of is_enabled() being true.
79  */
80
81 /* module parameters */
82 static unsigned long    filter_offset;
83 static int              nommiotrace;
84 static int              ISA_trace;
85 static int              trace_pc;
86
87 module_param(filter_offset, ulong, 0);
88 module_param(nommiotrace, bool, 0);
89 module_param(ISA_trace, bool, 0);
90 module_param(trace_pc, bool, 0);
91
92 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
93 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
94 MODULE_PARM_DESC(ISA_trace, "Do not exclude the low ISA range.");
95 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
96
97 static bool is_enabled(void)
98 {
99         return atomic_read(&mmiotrace_enabled);
100 }
101
102 #if 0 /* XXX: needs rewrite */
103 /*
104  * Write callback for the debugfs entry:
105  * Read a marker and write it to the mmio trace log
106  */
107 static ssize_t write_marker(struct file *file, const char __user *buffer,
108                                                 size_t count, loff_t *ppos)
109 {
110         char *event = NULL;
111         struct mm_io_header *headp;
112         ssize_t len = (count > 65535) ? 65535 : count;
113
114         event = kzalloc(sizeof(*headp) + len, GFP_KERNEL);
115         if (!event)
116                 return -ENOMEM;
117
118         headp = (struct mm_io_header *)event;
119         headp->type = MMIO_MAGIC | (MMIO_MARKER << MMIO_OPCODE_SHIFT);
120         headp->data_len = len;
121
122         if (copy_from_user(event + sizeof(*headp), buffer, len)) {
123                 kfree(event);
124                 return -EFAULT;
125         }
126
127         spin_lock_irq(&trace_lock);
128 #if 0 /* XXX: convert this to use tracing */
129         if (is_enabled())
130                 relay_write(chan, event, sizeof(*headp) + len);
131         else
132 #endif
133                 len = -EINVAL;
134         spin_unlock_irq(&trace_lock);
135         kfree(event);
136         return len;
137 }
138 #endif
139
140 static void print_pte(unsigned long address)
141 {
142         int level;
143         pte_t *pte = lookup_address(address, &level);
144
145         if (!pte) {
146                 pr_err(NAME "Error in %s: no pte for page 0x%08lx\n",
147                                                         __func__, address);
148                 return;
149         }
150
151         if (level == PG_LEVEL_2M) {
152                 pr_emerg(NAME "4MB pages are not currently supported: "
153                                                         "0x%08lx\n", address);
154                 BUG();
155         }
156         pr_info(NAME "pte for 0x%lx: 0x%lx 0x%lx\n", address, pte_val(*pte),
157                                                 pte_val(*pte) & _PAGE_PRESENT);
158 }
159
160 /*
161  * For some reason the pre/post pairs have been called in an
162  * unmatched order. Report and die.
163  */
164 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
165 {
166         const struct trap_reason *my_reason = &get_cpu_var(pf_reason);
167         pr_emerg(NAME "unexpected fault for address: 0x%08lx, "
168                                         "last fault for address: 0x%08lx\n",
169                                         addr, my_reason->addr);
170         print_pte(addr);
171         print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip);
172         print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip);
173 #ifdef __i386__
174         pr_emerg("eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
175                         regs->ax, regs->bx, regs->cx, regs->dx);
176         pr_emerg("esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
177                         regs->si, regs->di, regs->bp, regs->sp);
178 #else
179         pr_emerg("rax: %016lx   rcx: %016lx   rdx: %016lx\n",
180                                         regs->ax, regs->cx, regs->dx);
181         pr_emerg("rsi: %016lx   rdi: %016lx   rbp: %016lx   rsp: %016lx\n",
182                                 regs->si, regs->di, regs->bp, regs->sp);
183 #endif
184         put_cpu_var(pf_reason);
185         BUG();
186 }
187
188 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
189                                                 unsigned long addr)
190 {
191         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
192         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
193         const unsigned long instptr = instruction_pointer(regs);
194         const enum reason_type type = get_ins_type(instptr);
195         struct remap_trace *trace = p->user_data;
196
197         /* it doesn't make sense to have more than one active trace per cpu */
198         if (my_reason->active_traces)
199                 die_kmmio_nesting_error(regs, addr);
200         else
201                 my_reason->active_traces++;
202
203         my_reason->type = type;
204         my_reason->addr = addr;
205         my_reason->ip = instptr;
206
207         my_trace->phys = addr - trace->probe.addr + trace->phys;
208         my_trace->map_id = trace->id;
209
210         /*
211          * Only record the program counter when requested.
212          * It may taint clean-room reverse engineering.
213          */
214         if (trace_pc)
215                 my_trace->pc = instptr;
216         else
217                 my_trace->pc = 0;
218
219         /*
220          * XXX: the timestamp recorded will be *after* the tracing has been
221          * done, not at the time we hit the instruction. SMP implications
222          * on event ordering?
223          */
224
225         switch (type) {
226         case REG_READ:
227                 my_trace->opcode = MMIO_READ;
228                 my_trace->width = get_ins_mem_width(instptr);
229                 break;
230         case REG_WRITE:
231                 my_trace->opcode = MMIO_WRITE;
232                 my_trace->width = get_ins_mem_width(instptr);
233                 my_trace->value = get_ins_reg_val(instptr, regs);
234                 break;
235         case IMM_WRITE:
236                 my_trace->opcode = MMIO_WRITE;
237                 my_trace->width = get_ins_mem_width(instptr);
238                 my_trace->value = get_ins_imm_val(instptr);
239                 break;
240         default:
241                 {
242                         unsigned char *ip = (unsigned char *)instptr;
243                         my_trace->opcode = MMIO_UNKNOWN_OP;
244                         my_trace->width = 0;
245                         my_trace->value = (*ip) << 16 | *(ip + 1) << 8 |
246                                                                 *(ip + 2);
247                 }
248         }
249         put_cpu_var(cpu_trace);
250         put_cpu_var(pf_reason);
251 }
252
253 static void post(struct kmmio_probe *p, unsigned long condition,
254                                                         struct pt_regs *regs)
255 {
256         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
257         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
258
259         /* this should always return the active_trace count to 0 */
260         my_reason->active_traces--;
261         if (my_reason->active_traces) {
262                 pr_emerg(NAME "unexpected post handler");
263                 BUG();
264         }
265
266         switch (my_reason->type) {
267         case REG_READ:
268                 my_trace->value = get_ins_reg_val(my_reason->ip, regs);
269                 break;
270         default:
271                 break;
272         }
273
274         mmio_trace_rw(my_trace);
275         put_cpu_var(cpu_trace);
276         put_cpu_var(pf_reason);
277 }
278
279 static void ioremap_trace_core(unsigned long offset, unsigned long size,
280                                                         void __iomem *addr)
281 {
282         static atomic_t next_id;
283         struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
284         struct mmiotrace_map map = {
285                 .phys = offset,
286                 .virt = (unsigned long)addr,
287                 .len = size,
288                 .opcode = MMIO_PROBE
289         };
290
291         if (!trace) {
292                 pr_err(NAME "kmalloc failed in ioremap\n");
293                 return;
294         }
295
296         *trace = (struct remap_trace) {
297                 .probe = {
298                         .addr = (unsigned long)addr,
299                         .len = size,
300                         .pre_handler = pre,
301                         .post_handler = post,
302                         .user_data = trace
303                 },
304                 .phys = offset,
305                 .id = atomic_inc_return(&next_id)
306         };
307         map.map_id = trace->id;
308
309         spin_lock_irq(&trace_lock);
310         if (!is_enabled())
311                 goto not_enabled;
312
313         mmio_trace_mapping(&map);
314         list_add_tail(&trace->list, &trace_list);
315         if (!nommiotrace)
316                 register_kmmio_probe(&trace->probe);
317
318 not_enabled:
319         spin_unlock_irq(&trace_lock);
320 }
321
322 void
323 mmiotrace_ioremap(unsigned long offset, unsigned long size, void __iomem *addr)
324 {
325         if (!is_enabled()) /* recheck and proper locking in *_core() */
326                 return;
327
328         pr_debug(NAME "ioremap_*(0x%lx, 0x%lx) = %p\n", offset, size, addr);
329         if ((filter_offset) && (offset != filter_offset))
330                 return;
331         ioremap_trace_core(offset, size, addr);
332 }
333
334 static void iounmap_trace_core(volatile void __iomem *addr)
335 {
336         struct mmiotrace_map map = {
337                 .phys = 0,
338                 .virt = (unsigned long)addr,
339                 .len = 0,
340                 .opcode = MMIO_UNPROBE
341         };
342         struct remap_trace *trace;
343         struct remap_trace *tmp;
344         struct remap_trace *found_trace = NULL;
345
346         pr_debug(NAME "Unmapping %p.\n", addr);
347
348         spin_lock_irq(&trace_lock);
349         if (!is_enabled())
350                 goto not_enabled;
351
352         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
353                 if ((unsigned long)addr == trace->probe.addr) {
354                         if (!nommiotrace)
355                                 unregister_kmmio_probe(&trace->probe);
356                         list_del(&trace->list);
357                         found_trace = trace;
358                         break;
359                 }
360         }
361         map.map_id = (found_trace) ? found_trace->id : -1;
362         mmio_trace_mapping(&map);
363
364 not_enabled:
365         spin_unlock_irq(&trace_lock);
366         if (found_trace) {
367                 synchronize_rcu(); /* unregister_kmmio_probe() requirement */
368                 kfree(found_trace);
369         }
370 }
371
372 void mmiotrace_iounmap(volatile void __iomem *addr)
373 {
374         might_sleep();
375         if (is_enabled()) /* recheck and proper locking in *_core() */
376                 iounmap_trace_core(addr);
377 }
378
379 static void clear_trace_list(void)
380 {
381         struct remap_trace *trace;
382         struct remap_trace *tmp;
383
384         /*
385          * No locking required, because the caller ensures we are in a
386          * critical section via mutex, and is_enabled() is false,
387          * i.e. nothing can traverse or modify this list.
388          * Caller also ensures is_enabled() cannot change.
389          */
390         list_for_each_entry(trace, &trace_list, list) {
391                 pr_notice(NAME "purging non-iounmapped "
392                                         "trace @0x%08lx, size 0x%lx.\n",
393                                         trace->probe.addr, trace->probe.len);
394                 if (!nommiotrace)
395                         unregister_kmmio_probe(&trace->probe);
396         }
397         synchronize_rcu(); /* unregister_kmmio_probe() requirement */
398
399         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
400                 list_del(&trace->list);
401                 kfree(trace);
402         }
403 }
404
405 #if 0 /* XXX: out of order */
406 static struct file_operations fops_marker = {
407         .owner =        THIS_MODULE,
408         .write =        write_marker
409 };
410 #endif
411
412 void enable_mmiotrace(void)
413 {
414         mutex_lock(&mmiotrace_mutex);
415         if (is_enabled())
416                 goto out;
417
418         reference_kmmio();
419
420 #if 0 /* XXX: tracing does not support text entries */
421         marker_file = debugfs_create_file("marker", 0660, dir, NULL,
422                                                                 &fops_marker);
423         if (!marker_file)
424                 pr_err(NAME "marker file creation failed.\n");
425 #endif
426
427         if (nommiotrace)
428                 pr_info(NAME "MMIO tracing disabled.\n");
429         if (ISA_trace)
430                 pr_warning(NAME "Warning! low ISA range will be traced.\n");
431         spin_lock_irq(&trace_lock);
432         atomic_inc(&mmiotrace_enabled);
433         spin_unlock_irq(&trace_lock);
434         pr_info(NAME "enabled.\n");
435 out:
436         mutex_unlock(&mmiotrace_mutex);
437 }
438
439 void disable_mmiotrace(void)
440 {
441         mutex_lock(&mmiotrace_mutex);
442         if (!is_enabled())
443                 goto out;
444
445         spin_lock_irq(&trace_lock);
446         atomic_dec(&mmiotrace_enabled);
447         BUG_ON(is_enabled());
448         spin_unlock_irq(&trace_lock);
449
450         clear_trace_list(); /* guarantees: no more kmmio callbacks */
451         unreference_kmmio();
452         if (marker_file) {
453                 debugfs_remove(marker_file);
454                 marker_file = NULL;
455         }
456
457         pr_info(NAME "disabled.\n");
458 out:
459         mutex_unlock(&mmiotrace_mutex);
460 }