x86: cpumask: x86 mmio-mod.c use cpumask_var_t for downed_cpus
[linux-2.6.git] / 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 #define DEBUG 1
23
24 #include <linux/module.h>
25 #include <linux/debugfs.h>
26 #include <linux/uaccess.h>
27 #include <linux/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 #include <linux/cpu.h>
36
37 #include "pf_in.h"
38
39 #define NAME "mmiotrace: "
40
41 struct trap_reason {
42         unsigned long addr;
43         unsigned long ip;
44         enum reason_type type;
45         int active_traces;
46 };
47
48 struct remap_trace {
49         struct list_head list;
50         struct kmmio_probe probe;
51         resource_size_t phys;
52         unsigned long id;
53 };
54
55 /* Accessed per-cpu. */
56 static DEFINE_PER_CPU(struct trap_reason, pf_reason);
57 static DEFINE_PER_CPU(struct mmiotrace_rw, cpu_trace);
58
59 static DEFINE_MUTEX(mmiotrace_mutex);
60 static DEFINE_SPINLOCK(trace_lock);
61 static atomic_t mmiotrace_enabled;
62 static LIST_HEAD(trace_list);           /* struct remap_trace */
63
64 /*
65  * Locking in this file:
66  * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections.
67  * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex
68  *   and trace_lock.
69  * - Routines depending on is_enabled() must take trace_lock.
70  * - trace_list users must hold trace_lock.
71  * - is_enabled() guarantees that mmio_trace_{rw,mapping} are allowed.
72  * - pre/post callbacks assume the effect of is_enabled() being true.
73  */
74
75 /* module parameters */
76 static unsigned long    filter_offset;
77 static int              nommiotrace;
78 static int              trace_pc;
79
80 module_param(filter_offset, ulong, 0);
81 module_param(nommiotrace, bool, 0);
82 module_param(trace_pc, bool, 0);
83
84 MODULE_PARM_DESC(filter_offset, "Start address of traced mappings.");
85 MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing.");
86 MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions.");
87
88 static bool is_enabled(void)
89 {
90         return atomic_read(&mmiotrace_enabled);
91 }
92
93 static void print_pte(unsigned long address)
94 {
95         unsigned int level;
96         pte_t *pte = lookup_address(address, &level);
97
98         if (!pte) {
99                 pr_err(NAME "Error in %s: no pte for page 0x%08lx\n",
100                                                         __func__, address);
101                 return;
102         }
103
104         if (level == PG_LEVEL_2M) {
105                 pr_emerg(NAME "4MB pages are not currently supported: "
106                                                         "0x%08lx\n", address);
107                 BUG();
108         }
109         pr_info(NAME "pte for 0x%lx: 0x%llx 0x%llx\n", address,
110                 (unsigned long long)pte_val(*pte),
111                 (unsigned long long)pte_val(*pte) & _PAGE_PRESENT);
112 }
113
114 /*
115  * For some reason the pre/post pairs have been called in an
116  * unmatched order. Report and die.
117  */
118 static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr)
119 {
120         const struct trap_reason *my_reason = &get_cpu_var(pf_reason);
121         pr_emerg(NAME "unexpected fault for address: 0x%08lx, "
122                                         "last fault for address: 0x%08lx\n",
123                                         addr, my_reason->addr);
124         print_pte(addr);
125         print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip);
126         print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip);
127 #ifdef __i386__
128         pr_emerg("eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
129                         regs->ax, regs->bx, regs->cx, regs->dx);
130         pr_emerg("esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
131                         regs->si, regs->di, regs->bp, regs->sp);
132 #else
133         pr_emerg("rax: %016lx   rcx: %016lx   rdx: %016lx\n",
134                                         regs->ax, regs->cx, regs->dx);
135         pr_emerg("rsi: %016lx   rdi: %016lx   rbp: %016lx   rsp: %016lx\n",
136                                 regs->si, regs->di, regs->bp, regs->sp);
137 #endif
138         put_cpu_var(pf_reason);
139         BUG();
140 }
141
142 static void pre(struct kmmio_probe *p, struct pt_regs *regs,
143                                                 unsigned long addr)
144 {
145         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
146         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
147         const unsigned long instptr = instruction_pointer(regs);
148         const enum reason_type type = get_ins_type(instptr);
149         struct remap_trace *trace = p->private;
150
151         /* it doesn't make sense to have more than one active trace per cpu */
152         if (my_reason->active_traces)
153                 die_kmmio_nesting_error(regs, addr);
154         else
155                 my_reason->active_traces++;
156
157         my_reason->type = type;
158         my_reason->addr = addr;
159         my_reason->ip = instptr;
160
161         my_trace->phys = addr - trace->probe.addr + trace->phys;
162         my_trace->map_id = trace->id;
163
164         /*
165          * Only record the program counter when requested.
166          * It may taint clean-room reverse engineering.
167          */
168         if (trace_pc)
169                 my_trace->pc = instptr;
170         else
171                 my_trace->pc = 0;
172
173         /*
174          * XXX: the timestamp recorded will be *after* the tracing has been
175          * done, not at the time we hit the instruction. SMP implications
176          * on event ordering?
177          */
178
179         switch (type) {
180         case REG_READ:
181                 my_trace->opcode = MMIO_READ;
182                 my_trace->width = get_ins_mem_width(instptr);
183                 break;
184         case REG_WRITE:
185                 my_trace->opcode = MMIO_WRITE;
186                 my_trace->width = get_ins_mem_width(instptr);
187                 my_trace->value = get_ins_reg_val(instptr, regs);
188                 break;
189         case IMM_WRITE:
190                 my_trace->opcode = MMIO_WRITE;
191                 my_trace->width = get_ins_mem_width(instptr);
192                 my_trace->value = get_ins_imm_val(instptr);
193                 break;
194         default:
195                 {
196                         unsigned char *ip = (unsigned char *)instptr;
197                         my_trace->opcode = MMIO_UNKNOWN_OP;
198                         my_trace->width = 0;
199                         my_trace->value = (*ip) << 16 | *(ip + 1) << 8 |
200                                                                 *(ip + 2);
201                 }
202         }
203         put_cpu_var(cpu_trace);
204         put_cpu_var(pf_reason);
205 }
206
207 static void post(struct kmmio_probe *p, unsigned long condition,
208                                                         struct pt_regs *regs)
209 {
210         struct trap_reason *my_reason = &get_cpu_var(pf_reason);
211         struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace);
212
213         /* this should always return the active_trace count to 0 */
214         my_reason->active_traces--;
215         if (my_reason->active_traces) {
216                 pr_emerg(NAME "unexpected post handler");
217                 BUG();
218         }
219
220         switch (my_reason->type) {
221         case REG_READ:
222                 my_trace->value = get_ins_reg_val(my_reason->ip, regs);
223                 break;
224         default:
225                 break;
226         }
227
228         mmio_trace_rw(my_trace);
229         put_cpu_var(cpu_trace);
230         put_cpu_var(pf_reason);
231 }
232
233 static void ioremap_trace_core(resource_size_t offset, unsigned long size,
234                                                         void __iomem *addr)
235 {
236         static atomic_t next_id;
237         struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL);
238         /* These are page-unaligned. */
239         struct mmiotrace_map map = {
240                 .phys = offset,
241                 .virt = (unsigned long)addr,
242                 .len = size,
243                 .opcode = MMIO_PROBE
244         };
245
246         if (!trace) {
247                 pr_err(NAME "kmalloc failed in ioremap\n");
248                 return;
249         }
250
251         *trace = (struct remap_trace) {
252                 .probe = {
253                         .addr = (unsigned long)addr,
254                         .len = size,
255                         .pre_handler = pre,
256                         .post_handler = post,
257                         .private = trace
258                 },
259                 .phys = offset,
260                 .id = atomic_inc_return(&next_id)
261         };
262         map.map_id = trace->id;
263
264         spin_lock_irq(&trace_lock);
265         if (!is_enabled()) {
266                 kfree(trace);
267                 goto not_enabled;
268         }
269
270         mmio_trace_mapping(&map);
271         list_add_tail(&trace->list, &trace_list);
272         if (!nommiotrace)
273                 register_kmmio_probe(&trace->probe);
274
275 not_enabled:
276         spin_unlock_irq(&trace_lock);
277 }
278
279 void mmiotrace_ioremap(resource_size_t offset, unsigned long size,
280                                                 void __iomem *addr)
281 {
282         if (!is_enabled()) /* recheck and proper locking in *_core() */
283                 return;
284
285         pr_debug(NAME "ioremap_*(0x%llx, 0x%lx) = %p\n",
286                                 (unsigned long long)offset, size, addr);
287         if ((filter_offset) && (offset != filter_offset))
288                 return;
289         ioremap_trace_core(offset, size, addr);
290 }
291
292 static void iounmap_trace_core(volatile void __iomem *addr)
293 {
294         struct mmiotrace_map map = {
295                 .phys = 0,
296                 .virt = (unsigned long)addr,
297                 .len = 0,
298                 .opcode = MMIO_UNPROBE
299         };
300         struct remap_trace *trace;
301         struct remap_trace *tmp;
302         struct remap_trace *found_trace = NULL;
303
304         pr_debug(NAME "Unmapping %p.\n", addr);
305
306         spin_lock_irq(&trace_lock);
307         if (!is_enabled())
308                 goto not_enabled;
309
310         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
311                 if ((unsigned long)addr == trace->probe.addr) {
312                         if (!nommiotrace)
313                                 unregister_kmmio_probe(&trace->probe);
314                         list_del(&trace->list);
315                         found_trace = trace;
316                         break;
317                 }
318         }
319         map.map_id = (found_trace) ? found_trace->id : -1;
320         mmio_trace_mapping(&map);
321
322 not_enabled:
323         spin_unlock_irq(&trace_lock);
324         if (found_trace) {
325                 synchronize_rcu(); /* unregister_kmmio_probe() requirement */
326                 kfree(found_trace);
327         }
328 }
329
330 void mmiotrace_iounmap(volatile void __iomem *addr)
331 {
332         might_sleep();
333         if (is_enabled()) /* recheck and proper locking in *_core() */
334                 iounmap_trace_core(addr);
335 }
336
337 int mmiotrace_printk(const char *fmt, ...)
338 {
339         int ret = 0;
340         va_list args;
341         unsigned long flags;
342         va_start(args, fmt);
343
344         spin_lock_irqsave(&trace_lock, flags);
345         if (is_enabled())
346                 ret = mmio_trace_printk(fmt, args);
347         spin_unlock_irqrestore(&trace_lock, flags);
348
349         va_end(args);
350         return ret;
351 }
352 EXPORT_SYMBOL(mmiotrace_printk);
353
354 static void clear_trace_list(void)
355 {
356         struct remap_trace *trace;
357         struct remap_trace *tmp;
358
359         /*
360          * No locking required, because the caller ensures we are in a
361          * critical section via mutex, and is_enabled() is false,
362          * i.e. nothing can traverse or modify this list.
363          * Caller also ensures is_enabled() cannot change.
364          */
365         list_for_each_entry(trace, &trace_list, list) {
366                 pr_notice(NAME "purging non-iounmapped "
367                                         "trace @0x%08lx, size 0x%lx.\n",
368                                         trace->probe.addr, trace->probe.len);
369                 if (!nommiotrace)
370                         unregister_kmmio_probe(&trace->probe);
371         }
372         synchronize_rcu(); /* unregister_kmmio_probe() requirement */
373
374         list_for_each_entry_safe(trace, tmp, &trace_list, list) {
375                 list_del(&trace->list);
376                 kfree(trace);
377         }
378 }
379
380 #ifdef CONFIG_HOTPLUG_CPU
381 static cpumask_var_t downed_cpus;
382
383 static void enter_uniprocessor(void)
384 {
385         int cpu;
386         int err;
387
388         if (downed_cpus == NULL &&
389             !alloc_cpumask_var(&downed_cpus, GFP_KERNEL)) {
390                 pr_notice(NAME "Failed to allocate mask\n");
391                 goto out;
392         }
393
394         get_online_cpus();
395         cpumask_copy(downed_cpus, cpu_online_mask);
396         cpumask_clear_cpu(cpumask_first(cpu_online_mask), downed_cpus);
397         if (num_online_cpus() > 1)
398                 pr_notice(NAME "Disabling non-boot CPUs...\n");
399         put_online_cpus();
400
401         for_each_cpu(cpu, downed_cpus) {
402                 err = cpu_down(cpu);
403                 if (!err)
404                         pr_info(NAME "CPU%d is down.\n", cpu);
405                 else
406                         pr_err(NAME "Error taking CPU%d down: %d\n", cpu, err);
407         }
408 out:
409         if (num_online_cpus() > 1)
410                 pr_warning(NAME "multiple CPUs still online, "
411                                                 "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(NAME "Re-enabling CPUs...\n");
424         for_each_cpu(cpu, downed_cpus) {
425                 err = cpu_up(cpu);
426                 if (!err)
427                         pr_info(NAME "enabled CPU%d.\n", cpu);
428                 else
429                         pr_err(NAME "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(NAME "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(NAME "MMIO tracing disabled.\n");
454         enter_uniprocessor();
455         spin_lock_irq(&trace_lock);
456         atomic_inc(&mmiotrace_enabled);
457         spin_unlock_irq(&trace_lock);
458         pr_info(NAME "enabled.\n");
459 out:
460         mutex_unlock(&mmiotrace_mutex);
461 }
462
463 void disable_mmiotrace(void)
464 {
465         mutex_lock(&mmiotrace_mutex);
466         if (!is_enabled())
467                 goto out;
468
469         spin_lock_irq(&trace_lock);
470         atomic_dec(&mmiotrace_enabled);
471         BUG_ON(is_enabled());
472         spin_unlock_irq(&trace_lock);
473
474         clear_trace_list(); /* guarantees: no more kmmio callbacks */
475         leave_uniprocessor();
476         pr_info(NAME "disabled.\n");
477 out:
478         mutex_unlock(&mmiotrace_mutex);
479 }