blob: 4c4508e8a2043015c1cce3eb49a46c0c435e9297 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1995 Linus Torvalds
Ingo Molnar2d4a7162009-02-20 19:56:40 +01003 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
Ingo Molnarf8eeb2e2009-02-20 23:13:36 +01004 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 */
Ingo Molnara2bcd472009-03-29 23:47:48 +02006#include <linux/magic.h> /* STACK_END_MAGIC */
7#include <linux/sched.h> /* test_thread_flag(), ... */
8#include <linux/kdebug.h> /* oops_begin/end, ... */
9#include <linux/module.h> /* search_exception_table */
10#include <linux/bootmem.h> /* max_low_pfn */
11#include <linux/kprobes.h> /* __kprobes, ... */
12#include <linux/mmiotrace.h> /* kmmio_handler, ... */
Ingo Molnarcdd6c482009-09-21 12:02:48 +020013#include <linux/perf_event.h> /* perf_sw_event */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014
Ingo Molnara2bcd472009-03-29 23:47:48 +020015#include <asm/traps.h> /* dotraplinkage, ... */
16#include <asm/pgalloc.h> /* pgd_*(), ... */
Vegard Nossumf8561292008-04-04 00:53:23 +020017#include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Harvey Harrison33cb5242008-01-30 13:32:19 +010019/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010020 * Page fault error code bits:
21 *
22 * bit 0 == 0: no page found 1: protection fault
23 * bit 1 == 0: read access 1: write access
24 * bit 2 == 0: kernel-mode access 1: user-mode access
25 * bit 3 == 1: use of reserved bit detected
26 * bit 4 == 1: fault was an instruction fetch
Harvey Harrison33cb5242008-01-30 13:32:19 +010027 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +010028enum x86_pf_error_code {
29
30 PF_PROT = 1 << 0,
31 PF_WRITE = 1 << 1,
32 PF_USER = 1 << 2,
33 PF_RSVD = 1 << 3,
34 PF_INSTR = 1 << 4,
35};
Andi Kleen66c58152006-01-11 22:44:09 +010036
Ingo Molnarb814d412009-02-20 22:32:10 +010037/*
Ingo Molnarb319eed2009-02-22 10:24:18 +010038 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
Ingo Molnarb814d412009-02-20 22:32:10 +010040 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -040041static inline int __kprobes
42kmmio_fault(struct pt_regs *regs, unsigned long addr)
Pekka Paalanen86069782008-05-12 21:20:56 +020043{
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020044 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs, addr) == 1)
46 return -1;
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020047 return 0;
Pekka Paalanen86069782008-05-12 21:20:56 +020048}
49
Masami Hiramatsu62c92952009-08-27 13:23:11 -040050static inline int __kprobes notify_page_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070051{
Christoph Hellwig74a0b572007-10-16 01:24:07 -070052 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070053
Christoph Hellwig74a0b572007-10-16 01:24:07 -070054 /* kprobe_running() needs smp_processor_id() */
Ingo Molnarb1801812009-02-20 22:42:57 +010055 if (kprobes_built_in() && !user_mode_vm(regs)) {
Christoph Hellwig74a0b572007-10-16 01:24:07 -070056 preempt_disable();
57 if (kprobe_running() && kprobe_fault_handler(regs, 14))
58 ret = 1;
59 preempt_enable();
60 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070061
Christoph Hellwig74a0b572007-10-16 01:24:07 -070062 return ret;
Harvey Harrison33cb5242008-01-30 13:32:19 +010063}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070064
Harvey Harrison1dc85be2008-01-30 13:32:35 +010065/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010066 * Prefetch quirks:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010067 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010068 * 32-bit mode:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010069 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010070 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
71 * Check that here and ignore it.
72 *
73 * 64-bit mode:
74 *
75 * Sometimes the CPU reports invalid exceptions on prefetch.
76 * Check that here and ignore it.
77 *
78 * Opcode checker based on code by Richard Brunner.
Harvey Harrison1dc85be2008-01-30 13:32:35 +010079 */
Ingo Molnar107a0362009-02-20 20:37:05 +010080static inline int
81check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
82 unsigned char opcode, int *prefetch)
83{
84 unsigned char instr_hi = opcode & 0xf0;
85 unsigned char instr_lo = opcode & 0x0f;
86
87 switch (instr_hi) {
88 case 0x20:
89 case 0x30:
90 /*
91 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
92 * In X86_64 long mode, the CPU will signal invalid
93 * opcode if some of these prefixes are present so
94 * X86_64 will never get here anyway
95 */
96 return ((instr_lo & 7) == 0x6);
97#ifdef CONFIG_X86_64
98 case 0x40:
99 /*
100 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
101 * Need to figure out under what instruction mode the
102 * instruction was issued. Could check the LDT for lm,
103 * but for now it's good enough to assume that long
104 * mode only uses well known segments or kernel.
105 */
106 return (!user_mode(regs)) || (regs->cs == __USER_CS);
107#endif
108 case 0x60:
109 /* 0x64 thru 0x67 are valid prefixes in all modes. */
110 return (instr_lo & 0xC) == 0x4;
111 case 0xF0:
112 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
113 return !instr_lo || (instr_lo>>1) == 1;
114 case 0x00:
115 /* Prefetch instruction is 0x0F0D or 0x0F18 */
116 if (probe_kernel_address(instr, opcode))
117 return 0;
118
119 *prefetch = (instr_lo == 0xF) &&
120 (opcode == 0x0D || opcode == 0x18);
121 return 0;
122 default:
123 return 0;
124 }
125}
126
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100127static int
128is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
Harvey Harrison33cb5242008-01-30 13:32:19 +0100129{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100130 unsigned char *max_instr;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100131 unsigned char *instr;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100132 int prefetch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Ingo Molnar30853542008-03-27 21:29:09 +0100134 /*
135 * If it was a exec (instruction fetch) fault on NX page, then
136 * do not ignore the fault:
137 */
Andi Kleen66c58152006-01-11 22:44:09 +0100138 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100140
Ingo Molnar107a0362009-02-20 20:37:05 +0100141 instr = (void *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700142 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700144 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return 0;
146
Ingo Molnar107a0362009-02-20 20:37:05 +0100147 while (instr < max_instr) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100148 unsigned char opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100150 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100151 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 instr++;
154
Ingo Molnar107a0362009-02-20 20:37:05 +0100155 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
158 return prefetch;
159}
160
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100161static void
162force_sig_info_fault(int si_signo, int si_code, unsigned long address,
163 struct task_struct *tsk)
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100164{
165 siginfo_t info;
166
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100167 info.si_signo = si_signo;
168 info.si_errno = 0;
169 info.si_code = si_code;
170 info.si_addr = (void __user *)address;
Andi Kleena6e04aa2009-09-16 11:50:09 +0200171 info.si_addr_lsb = si_code == BUS_MCEERR_AR ? PAGE_SHIFT : 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100172
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100173 force_sig_info(si_signo, &info, tsk);
174}
175
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100176DEFINE_SPINLOCK(pgd_lock);
177LIST_HEAD(pgd_list);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100178
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100179#ifdef CONFIG_X86_32
180static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
181{
182 unsigned index = pgd_index(address);
183 pgd_t *pgd_k;
184 pud_t *pud, *pud_k;
185 pmd_t *pmd, *pmd_k;
186
187 pgd += index;
188 pgd_k = init_mm.pgd + index;
189
190 if (!pgd_present(*pgd_k))
191 return NULL;
192
193 /*
194 * set_pgd(pgd, *pgd_k); here would be useless on PAE
195 * and redundant with the set_pmd() on non-PAE. As would
196 * set_pud.
197 */
198 pud = pud_offset(pgd, address);
199 pud_k = pud_offset(pgd_k, address);
200 if (!pud_present(*pud_k))
201 return NULL;
202
203 pmd = pmd_offset(pud, address);
204 pmd_k = pmd_offset(pud_k, address);
205 if (!pmd_present(*pmd_k))
206 return NULL;
207
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800208 if (!pmd_present(*pmd))
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100209 set_pmd(pmd, *pmd_k);
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800210 else
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100211 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100212
213 return pmd_k;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100214}
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100215
216void vmalloc_sync_all(void)
217{
218 unsigned long address;
219
220 if (SHARED_KERNEL_PMD)
221 return;
222
223 for (address = VMALLOC_START & PMD_MASK;
224 address >= TASK_SIZE && address < FIXADDR_TOP;
225 address += PMD_SIZE) {
226
227 unsigned long flags;
228 struct page *page;
229
230 spin_lock_irqsave(&pgd_lock, flags);
231 list_for_each_entry(page, &pgd_list, lru) {
232 if (!vmalloc_sync_one(page_address(page), address))
233 break;
234 }
235 spin_unlock_irqrestore(&pgd_lock, flags);
236 }
237}
238
239/*
240 * 32-bit:
241 *
242 * Handle a fault on the vmalloc or module mapping area
243 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400244static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100245{
246 unsigned long pgd_paddr;
247 pmd_t *pmd_k;
248 pte_t *pte_k;
249
250 /* Make sure we are in vmalloc area: */
251 if (!(address >= VMALLOC_START && address < VMALLOC_END))
252 return -1;
253
254 /*
255 * Synchronize this task's top level page-table
256 * with the 'reference' page table.
257 *
258 * Do _not_ use "current" here. We might be inside
259 * an interrupt in the middle of a task switch..
260 */
261 pgd_paddr = read_cr3();
262 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
263 if (!pmd_k)
264 return -1;
265
266 pte_k = pte_offset_kernel(pmd_k, address);
267 if (!pte_present(*pte_k))
268 return -1;
269
270 return 0;
271}
272
273/*
274 * Did it hit the DOS screen memory VA from vm86 mode?
275 */
276static inline void
277check_v8086_mode(struct pt_regs *regs, unsigned long address,
278 struct task_struct *tsk)
279{
280 unsigned long bit;
281
282 if (!v8086_mode(regs))
283 return;
284
285 bit = (address - 0xA0000) >> PAGE_SHIFT;
286 if (bit < 32)
287 tsk->thread.screen_bitmap |= 1 << bit;
288}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Akinobu Mita087975b2009-06-27 15:35:15 +0900290static bool low_pfn(unsigned long pfn)
291{
292 return pfn < max_low_pfn;
293}
294
Adrian Bunkcae30f822008-02-13 23:31:31 +0200295static void dump_pagetable(unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Akinobu Mita087975b2009-06-27 15:35:15 +0900297 pgd_t *base = __va(read_cr3());
298 pgd_t *pgd = &base[pgd_index(address)];
299 pmd_t *pmd;
300 pte_t *pte;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100301
Harvey Harrison1156e092008-01-30 13:34:10 +0100302#ifdef CONFIG_X86_PAE
Akinobu Mita087975b2009-06-27 15:35:15 +0900303 printk("*pdpt = %016Lx ", pgd_val(*pgd));
304 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
305 goto out;
Harvey Harrison1156e092008-01-30 13:34:10 +0100306#endif
Akinobu Mita087975b2009-06-27 15:35:15 +0900307 pmd = pmd_offset(pud_offset(pgd, address), address);
308 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
Harvey Harrison1156e092008-01-30 13:34:10 +0100309
310 /*
311 * We must not directly access the pte in the highpte
312 * case if the page table is located in highmem.
313 * And let's rather not kmap-atomic the pte, just in case
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100314 * it's allocated already:
Harvey Harrison1156e092008-01-30 13:34:10 +0100315 */
Akinobu Mita087975b2009-06-27 15:35:15 +0900316 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
317 goto out;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100318
Akinobu Mita087975b2009-06-27 15:35:15 +0900319 pte = pte_offset_kernel(pmd, address);
320 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
321out:
Harvey Harrison1156e092008-01-30 13:34:10 +0100322 printk("\n");
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100323}
324
325#else /* CONFIG_X86_64: */
326
327void vmalloc_sync_all(void)
328{
329 unsigned long address;
330
331 for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END;
332 address += PGDIR_SIZE) {
333
334 const pgd_t *pgd_ref = pgd_offset_k(address);
335 unsigned long flags;
336 struct page *page;
337
338 if (pgd_none(*pgd_ref))
339 continue;
340
341 spin_lock_irqsave(&pgd_lock, flags);
342 list_for_each_entry(page, &pgd_list, lru) {
343 pgd_t *pgd;
344 pgd = (pgd_t *)page_address(page) + pgd_index(address);
345 if (pgd_none(*pgd))
346 set_pgd(pgd, *pgd_ref);
347 else
348 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
349 }
350 spin_unlock_irqrestore(&pgd_lock, flags);
351 }
352}
353
354/*
355 * 64-bit:
356 *
357 * Handle a fault on the vmalloc area
358 *
359 * This assumes no large pages in there.
360 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400361static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100362{
363 pgd_t *pgd, *pgd_ref;
364 pud_t *pud, *pud_ref;
365 pmd_t *pmd, *pmd_ref;
366 pte_t *pte, *pte_ref;
367
368 /* Make sure we are in vmalloc area: */
369 if (!(address >= VMALLOC_START && address < VMALLOC_END))
370 return -1;
371
372 /*
373 * Copy kernel mappings over when needed. This can also
374 * happen within a race in page table update. In the later
375 * case just flush:
376 */
377 pgd = pgd_offset(current->active_mm, address);
378 pgd_ref = pgd_offset_k(address);
379 if (pgd_none(*pgd_ref))
380 return -1;
381
382 if (pgd_none(*pgd))
383 set_pgd(pgd, *pgd_ref);
384 else
385 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
386
387 /*
388 * Below here mismatches are bugs because these lower tables
389 * are shared:
390 */
391
392 pud = pud_offset(pgd, address);
393 pud_ref = pud_offset(pgd_ref, address);
394 if (pud_none(*pud_ref))
395 return -1;
396
397 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
398 BUG();
399
400 pmd = pmd_offset(pud, address);
401 pmd_ref = pmd_offset(pud_ref, address);
402 if (pmd_none(*pmd_ref))
403 return -1;
404
405 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
406 BUG();
407
408 pte_ref = pte_offset_kernel(pmd_ref, address);
409 if (!pte_present(*pte_ref))
410 return -1;
411
412 pte = pte_offset_kernel(pmd, address);
413
414 /*
415 * Don't use pte_page here, because the mappings can point
416 * outside mem_map, and the NUMA hash lookup cannot handle
417 * that:
418 */
419 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
420 BUG();
421
422 return 0;
423}
424
425static const char errata93_warning[] =
Joe Perchesad361c92009-07-06 13:05:40 -0700426KERN_ERR
427"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
428"******* Working around it, but it may cause SEGVs or burn power.\n"
429"******* Please consider a BIOS update.\n"
430"******* Disabling USB legacy in the BIOS may also help.\n";
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100431
432/*
433 * No vm86 mode in 64-bit mode:
434 */
435static inline void
436check_v8086_mode(struct pt_regs *regs, unsigned long address,
437 struct task_struct *tsk)
438{
439}
440
441static int bad_address(void *p)
442{
443 unsigned long dummy;
444
445 return probe_kernel_address((unsigned long *)p, dummy);
446}
447
448static void dump_pagetable(unsigned long address)
449{
Akinobu Mita087975b2009-06-27 15:35:15 +0900450 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
451 pgd_t *pgd = base + pgd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 pud_t *pud;
453 pmd_t *pmd;
454 pte_t *pte;
455
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100456 if (bad_address(pgd))
457 goto bad;
458
Jan Beulichd646bce2006-02-03 21:51:47 +0100459 printk("PGD %lx ", pgd_val(*pgd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100460
461 if (!pgd_present(*pgd))
462 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
Andi Kleend2ae5b52006-06-26 13:57:56 +0200464 pud = pud_offset(pgd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100465 if (bad_address(pud))
466 goto bad;
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 printk("PUD %lx ", pud_val(*pud));
Andi Kleenb5360222008-02-04 16:48:09 +0100469 if (!pud_present(*pud) || pud_large(*pud))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100470 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 pmd = pmd_offset(pud, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100473 if (bad_address(pmd))
474 goto bad;
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 printk("PMD %lx ", pmd_val(*pmd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100477 if (!pmd_present(*pmd) || pmd_large(*pmd))
478 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 pte = pte_offset_kernel(pmd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100481 if (bad_address(pte))
482 goto bad;
483
Harvey Harrison33cb5242008-01-30 13:32:19 +0100484 printk("PTE %lx", pte_val(*pte));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100485out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 printk("\n");
487 return;
488bad:
489 printk("BAD\n");
490}
491
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100492#endif /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100494/*
495 * Workaround for K8 erratum #93 & buggy BIOS.
496 *
497 * BIOS SMM functions are required to use a specific workaround
498 * to avoid corruption of the 64bit RIP register on C stepping K8.
499 *
500 * A lot of BIOS that didn't get tested properly miss this.
501 *
502 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
503 * Try to work around it here.
504 *
505 * Note we only handle faults in kernel here.
506 * Does nothing on 32-bit.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100507 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100508static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100510#ifdef CONFIG_X86_64
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100511 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100513
Harvey Harrison33cb5242008-01-30 13:32:19 +0100514 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100518 if ((address >= (u64)_stext && address <= (u64)_etext) ||
519 (address >= MODULES_VADDR && address <= MODULES_END)) {
Ingo Molnara454ab32009-05-03 10:09:03 +0200520 printk_once(errata93_warning);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100521 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return 1;
523 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100524#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100526}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Harvey Harrison35f32662008-01-30 13:34:09 +0100528/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100529 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
530 * to illegal addresses >4GB.
531 *
532 * We catch this in the page fault handler because these addresses
533 * are not reachable. Just detect this case and return. Any code
Harvey Harrison35f32662008-01-30 13:34:09 +0100534 * segment in LDT is compatibility mode.
535 */
536static int is_errata100(struct pt_regs *regs, unsigned long address)
537{
538#ifdef CONFIG_X86_64
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100539 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
Harvey Harrison35f32662008-01-30 13:34:09 +0100540 return 1;
541#endif
542 return 0;
543}
544
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100545static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
546{
547#ifdef CONFIG_X86_F00F_BUG
548 unsigned long nr;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100549
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100550 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100551 * Pentium F0 0F C7 C8 bug workaround:
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100552 */
553 if (boot_cpu_data.f00f_bug) {
554 nr = (address - idt_descr.address) >> 3;
555
556 if (nr == 6) {
557 do_invalid_op(regs, 0);
558 return 1;
559 }
560 }
561#endif
562 return 0;
563}
564
Ingo Molnar8f766142009-02-20 23:00:29 +0100565static const char nx_warning[] = KERN_CRIT
566"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
567
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100568static void
569show_fault_oops(struct pt_regs *regs, unsigned long error_code,
570 unsigned long address)
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100571{
Harvey Harrison1156e092008-01-30 13:34:10 +0100572 if (!oops_may_print())
573 return;
574
Harvey Harrison1156e092008-01-30 13:34:10 +0100575 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100576 unsigned int level;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100577
Harvey Harrison1156e092008-01-30 13:34:10 +0100578 pte_t *pte = lookup_address(address, &level);
579
Ingo Molnar8f766142009-02-20 23:00:29 +0100580 if (pte && pte_present(*pte) && !pte_exec(*pte))
581 printk(nx_warning, current_uid());
Harvey Harrison1156e092008-01-30 13:34:10 +0100582 }
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100583
Harvey Harrison1156e092008-01-30 13:34:10 +0100584 printk(KERN_ALERT "BUG: unable to handle kernel ");
585 if (address < PAGE_SIZE)
586 printk(KERN_CONT "NULL pointer dereference");
587 else
588 printk(KERN_CONT "paging request");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100589
Vegard Nossumf294a8c2008-07-01 15:38:13 +0200590 printk(KERN_CONT " at %p\n", (void *) address);
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100591 printk(KERN_ALERT "IP:");
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100592 printk_address(regs->ip, 1);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100593
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100594 dump_pagetable(address);
595}
596
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100597static noinline void
598pgtable_bad(struct pt_regs *regs, unsigned long error_code,
599 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100601 struct task_struct *tsk;
602 unsigned long flags;
603 int sig;
604
605 flags = oops_begin();
606 tsk = current;
607 sig = SIGKILL;
Jan Beulich12091402005-09-12 18:49:24 +0200608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
Nick Piggin92181f12009-01-20 04:24:26 +0100610 tsk->comm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 dump_pagetable(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100612
613 tsk->thread.cr2 = address;
614 tsk->thread.trap_no = 14;
615 tsk->thread.error_code = error_code;
616
Jan Beulich22f59912008-01-30 13:31:23 +0100617 if (__die("Bad pagetable", regs, error_code))
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200618 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100619
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200620 oops_end(flags, regs, sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100623static noinline void
624no_context(struct pt_regs *regs, unsigned long error_code,
625 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100626{
627 struct task_struct *tsk = current;
Ingo Molnar19803072009-01-21 10:39:51 +0100628 unsigned long *stackend;
Nick Piggin92181f12009-01-20 04:24:26 +0100629 unsigned long flags;
630 int sig;
Nick Piggin92181f12009-01-20 04:24:26 +0100631
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100632 /* Are we prepared to handle this kernel fault? */
Nick Piggin92181f12009-01-20 04:24:26 +0100633 if (fixup_exception(regs))
634 return;
635
636 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100637 * 32-bit:
Nick Piggin92181f12009-01-20 04:24:26 +0100638 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100639 * Valid to do another page fault here, because if this fault
640 * had been triggered by is_prefetch fixup_exception would have
641 * handled it.
642 *
643 * 64-bit:
644 *
645 * Hall of shame of CPU/BIOS bugs.
Nick Piggin92181f12009-01-20 04:24:26 +0100646 */
647 if (is_prefetch(regs, error_code, address))
648 return;
649
650 if (is_errata93(regs, address))
651 return;
652
653 /*
654 * Oops. The kernel tried to access some bad page. We'll have to
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100655 * terminate things with extreme prejudice:
Nick Piggin92181f12009-01-20 04:24:26 +0100656 */
Nick Piggin92181f12009-01-20 04:24:26 +0100657 flags = oops_begin();
Nick Piggin92181f12009-01-20 04:24:26 +0100658
659 show_fault_oops(regs, error_code, address);
660
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100661 stackend = end_of_stack(tsk);
Jan Beulich0e7810b2009-11-20 14:00:14 +0000662 if (tsk != &init_task && *stackend != STACK_END_MAGIC)
Ingo Molnar19803072009-01-21 10:39:51 +0100663 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
664
Ingo Molnar1cc99542009-02-20 23:07:48 +0100665 tsk->thread.cr2 = address;
666 tsk->thread.trap_no = 14;
667 tsk->thread.error_code = error_code;
Nick Piggin92181f12009-01-20 04:24:26 +0100668
Nick Piggin92181f12009-01-20 04:24:26 +0100669 sig = SIGKILL;
670 if (__die("Oops", regs, error_code))
671 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100672
Nick Piggin92181f12009-01-20 04:24:26 +0100673 /* Executive summary in case the body of the oops scrolled away */
674 printk(KERN_EMERG "CR2: %016lx\n", address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100675
Nick Piggin92181f12009-01-20 04:24:26 +0100676 oops_end(flags, regs, sig);
Nick Piggin92181f12009-01-20 04:24:26 +0100677}
678
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100679/*
680 * Print out info about fatal segfaults, if the show_unhandled_signals
681 * sysctl is set:
682 */
683static inline void
684show_signal_msg(struct pt_regs *regs, unsigned long error_code,
685 unsigned long address, struct task_struct *tsk)
686{
687 if (!unhandled_signal(tsk, SIGSEGV))
688 return;
689
690 if (!printk_ratelimit())
691 return;
692
Roland Dreiera1a08d12009-07-11 00:10:04 -0700693 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100694 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
695 tsk->comm, task_pid_nr(tsk), address,
696 (void *)regs->ip, (void *)regs->sp, error_code);
697
698 print_vma_addr(KERN_CONT " in ", regs->ip);
699
700 printk(KERN_CONT "\n");
701}
702
703static void
704__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
705 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100706{
707 struct task_struct *tsk = current;
708
709 /* User mode accesses just cause a SIGSEGV */
710 if (error_code & PF_USER) {
711 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100712 * It's possible to have interrupts off here:
Nick Piggin92181f12009-01-20 04:24:26 +0100713 */
714 local_irq_enable();
715
716 /*
717 * Valid to do another page fault here because this one came
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100718 * from user space:
Nick Piggin92181f12009-01-20 04:24:26 +0100719 */
720 if (is_prefetch(regs, error_code, address))
721 return;
722
723 if (is_errata100(regs, address))
724 return;
725
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100726 if (unlikely(show_unhandled_signals))
727 show_signal_msg(regs, error_code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100728
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100729 /* Kernel addresses are always protection faults: */
730 tsk->thread.cr2 = address;
731 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
732 tsk->thread.trap_no = 14;
733
Nick Piggin92181f12009-01-20 04:24:26 +0100734 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100735
Nick Piggin92181f12009-01-20 04:24:26 +0100736 return;
737 }
738
739 if (is_f00f_bug(regs, address))
740 return;
741
742 no_context(regs, error_code, address);
743}
744
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100745static noinline void
746bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
747 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100748{
749 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
750}
751
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100752static void
753__bad_area(struct pt_regs *regs, unsigned long error_code,
754 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100755{
756 struct mm_struct *mm = current->mm;
757
758 /*
759 * Something tried to access memory that isn't in our memory map..
760 * Fix it, but check if it's kernel or user first..
761 */
762 up_read(&mm->mmap_sem);
763
764 __bad_area_nosemaphore(regs, error_code, address, si_code);
765}
766
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100767static noinline void
768bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100769{
770 __bad_area(regs, error_code, address, SEGV_MAPERR);
771}
772
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100773static noinline void
774bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
775 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100776{
777 __bad_area(regs, error_code, address, SEGV_ACCERR);
778}
779
780/* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100781static void
782out_of_memory(struct pt_regs *regs, unsigned long error_code,
783 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100784{
785 /*
786 * We ran out of memory, call the OOM killer, and return the userspace
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100787 * (which will retry the fault, or kill us if we got oom-killed):
Nick Piggin92181f12009-01-20 04:24:26 +0100788 */
789 up_read(&current->mm->mmap_sem);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100790
Nick Piggin92181f12009-01-20 04:24:26 +0100791 pagefault_out_of_memory();
792}
793
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100794static void
Andi Kleena6e04aa2009-09-16 11:50:09 +0200795do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
796 unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100797{
798 struct task_struct *tsk = current;
799 struct mm_struct *mm = tsk->mm;
Andi Kleena6e04aa2009-09-16 11:50:09 +0200800 int code = BUS_ADRERR;
Nick Piggin92181f12009-01-20 04:24:26 +0100801
802 up_read(&mm->mmap_sem);
803
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100804 /* Kernel mode? Handle exceptions or die: */
Linus Torvalds96054562010-08-13 09:49:20 -0700805 if (!(error_code & PF_USER)) {
Nick Piggin92181f12009-01-20 04:24:26 +0100806 no_context(regs, error_code, address);
Linus Torvalds96054562010-08-13 09:49:20 -0700807 return;
808 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100809
Ingo Molnarcd1b68f2009-02-20 23:39:02 +0100810 /* User-space => ok to do another page fault: */
Nick Piggin92181f12009-01-20 04:24:26 +0100811 if (is_prefetch(regs, error_code, address))
812 return;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100813
814 tsk->thread.cr2 = address;
815 tsk->thread.error_code = error_code;
816 tsk->thread.trap_no = 14;
817
Andi Kleena6e04aa2009-09-16 11:50:09 +0200818#ifdef CONFIG_MEMORY_FAILURE
819 if (fault & VM_FAULT_HWPOISON) {
820 printk(KERN_ERR
821 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
822 tsk->comm, tsk->pid, address);
823 code = BUS_MCEERR_AR;
824 }
825#endif
826 force_sig_info_fault(SIGBUS, code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100827}
828
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100829static noinline void
830mm_fault_error(struct pt_regs *regs, unsigned long error_code,
831 unsigned long address, unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100832{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100833 if (fault & VM_FAULT_OOM) {
Nick Piggin92181f12009-01-20 04:24:26 +0100834 out_of_memory(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100835 } else {
Andi Kleena6e04aa2009-09-16 11:50:09 +0200836 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON))
837 do_sigbus(regs, error_code, address, fault);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100838 else
839 BUG();
840 }
Nick Piggin92181f12009-01-20 04:24:26 +0100841}
842
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100843static int spurious_fault_check(unsigned long error_code, pte_t *pte)
844{
845 if ((error_code & PF_WRITE) && !pte_write(*pte))
846 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100847
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100848 if ((error_code & PF_INSTR) && !pte_exec(*pte))
849 return 0;
850
851 return 1;
852}
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100855 * Handle a spurious fault caused by a stale TLB entry.
856 *
857 * This allows us to lazily refresh the TLB when increasing the
858 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
859 * eagerly is very expensive since that implies doing a full
860 * cross-processor TLB flush, even if no stale TLB entries exist
861 * on other processors.
862 *
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100863 * There are no security implications to leaving a stale TLB when
864 * increasing the permissions on a page.
865 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400866static noinline __kprobes int
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100867spurious_fault(unsigned long error_code, unsigned long address)
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100868{
869 pgd_t *pgd;
870 pud_t *pud;
871 pmd_t *pmd;
872 pte_t *pte;
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500873 int ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100874
875 /* Reserved-bit violation or user access to kernel space? */
876 if (error_code & (PF_USER | PF_RSVD))
877 return 0;
878
879 pgd = init_mm.pgd + pgd_index(address);
880 if (!pgd_present(*pgd))
881 return 0;
882
883 pud = pud_offset(pgd, address);
884 if (!pud_present(*pud))
885 return 0;
886
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100887 if (pud_large(*pud))
888 return spurious_fault_check(error_code, (pte_t *) pud);
889
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100890 pmd = pmd_offset(pud, address);
891 if (!pmd_present(*pmd))
892 return 0;
893
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100894 if (pmd_large(*pmd))
895 return spurious_fault_check(error_code, (pte_t *) pmd);
896
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100897 pte = pte_offset_kernel(pmd, address);
898 if (!pte_present(*pte))
899 return 0;
900
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500901 ret = spurious_fault_check(error_code, pte);
902 if (!ret)
903 return 0;
904
905 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100906 * Make sure we have permissions in PMD.
907 * If not, then there's a bug in the page tables:
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500908 */
909 ret = spurious_fault_check(error_code, (pte_t *) pmd);
910 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100911
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500912 return ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100913}
914
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200915int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100917static inline int
918access_error(unsigned long error_code, int write, struct vm_area_struct *vma)
Nick Piggin92181f12009-01-20 04:24:26 +0100919{
920 if (write) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100921 /* write, present and write, not present: */
Nick Piggin92181f12009-01-20 04:24:26 +0100922 if (unlikely(!(vma->vm_flags & VM_WRITE)))
923 return 1;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100924 return 0;
Nick Piggin92181f12009-01-20 04:24:26 +0100925 }
926
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100927 /* read, present: */
928 if (unlikely(error_code & PF_PROT))
929 return 1;
930
931 /* read, not present: */
932 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
933 return 1;
934
Nick Piggin92181f12009-01-20 04:24:26 +0100935 return 0;
936}
937
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800938static int fault_in_kernel_space(unsigned long address)
939{
Ingo Molnard9517342009-02-20 23:32:28 +0100940 return address >= TASK_SIZE_MAX;
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800941}
942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943/*
944 * This routine handles page faults. It determines the address,
945 * and the problem, and then passes it off to one of the appropriate
946 * routines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 */
Ingo Molnarc3731c62009-02-20 23:22:34 +0100948dotraplinkage void __kprobes
949do_page_fault(struct pt_regs *regs, unsigned long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
Harvey Harrison33cb5242008-01-30 13:32:19 +0100951 struct vm_area_struct *vma;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100952 struct task_struct *tsk;
953 unsigned long address;
954 struct mm_struct *mm;
Nick Piggin92181f12009-01-20 04:24:26 +0100955 int write;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100956 int fault;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100958 tsk = current;
959 mm = tsk->mm;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100960
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100961 /* Get the faulting address: */
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200962 address = read_cr2();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Vegard Nossumf8561292008-04-04 00:53:23 +0200964 /*
965 * Detect and handle instructions that would cause a page fault for
966 * both a tracked kernel page and a userspace page.
967 */
968 if (kmemcheck_active(regs))
969 kmemcheck_hide(regs);
Ingo Molnar5dfaf902009-06-16 10:23:32 +0200970 prefetchw(&mm->mmap_sem);
Vegard Nossumf8561292008-04-04 00:53:23 +0200971
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +0200972 if (unlikely(kmmio_fault(regs, address)))
Pekka Paalanen86069782008-05-12 21:20:56 +0200973 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 /*
976 * We fault-in kernel-space virtual memory on-demand. The
977 * 'reference' page table is init_mm.pgd.
978 *
979 * NOTE! We MUST NOT take any locks for this case. We may
980 * be in an interrupt or a critical region, and should
981 * only copy the information from the master page table,
982 * nothing more.
983 *
984 * This verifies that the fault happens in kernel space
985 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +0100986 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 */
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800988 if (unlikely(fault_in_kernel_space(address))) {
Vegard Nossumf8561292008-04-04 00:53:23 +0200989 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
990 if (vmalloc_fault(address) >= 0)
991 return;
992
993 if (kmemcheck_fault(regs, address, error_code))
994 return;
995 }
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100996
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100997 /* Can handle a stale RO->RW TLB: */
Nick Piggin92181f12009-01-20 04:24:26 +0100998 if (spurious_fault(error_code, address))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100999 return;
1000
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001001 /* kprobes don't want to hook the spurious faults: */
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001002 if (notify_page_fault(regs))
1003 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001004 /*
1005 * Don't take the mm semaphore here. If we fixup a prefetch
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001006 * fault we could otherwise deadlock:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001007 */
Nick Piggin92181f12009-01-20 04:24:26 +01001008 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001009
Nick Piggin92181f12009-01-20 04:24:26 +01001010 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001011 }
1012
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001013 /* kprobes don't want to hook the spurious faults: */
Ingo Molnarf8a6b2b2009-02-13 09:44:22 +01001014 if (unlikely(notify_page_fault(regs)))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001015 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001016 /*
Linus Torvalds891cffb2008-10-12 13:16:12 -07001017 * It's safe to allow irq's after cr2 has been saved and the
1018 * vmalloc fault has been handled.
1019 *
1020 * User-mode registers count as a user access even for any
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001021 * potential system fault or CPU buglet:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001022 */
Linus Torvalds891cffb2008-10-12 13:16:12 -07001023 if (user_mode_vm(regs)) {
1024 local_irq_enable();
1025 error_code |= PF_USER;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001026 } else {
1027 if (regs->flags & X86_EFLAGS_IF)
1028 local_irq_enable();
1029 }
Jan Beulich8c914cb2006-03-25 16:29:40 +01001030
Andi Kleen66c58152006-01-11 22:44:09 +01001031 if (unlikely(error_code & PF_RSVD))
Nick Piggin92181f12009-01-20 04:24:26 +01001032 pgtable_bad(regs, error_code, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001034 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
Peter Zijlstra7dd1fcc2009-03-13 12:21:33 +01001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001037 * If we're in an interrupt, have no user context or are running
1038 * in an atomic region then we must not take the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 */
Nick Piggin92181f12009-01-20 04:24:26 +01001040 if (unlikely(in_atomic() || !mm)) {
1041 bad_area_nosemaphore(regs, error_code, address);
1042 return;
1043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Ingo Molnar3a1dfe62008-10-13 17:49:02 +02001045 /*
1046 * When running in the kernel we expect faults to occur only to
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001047 * addresses in user space. All other faults represent errors in
1048 * the kernel and should generate an OOPS. Unfortunately, in the
1049 * case of an erroneous fault occurring in a code path which already
1050 * holds mmap_sem we will deadlock attempting to validate the fault
1051 * against the address space. Luckily the kernel only validly
1052 * references user space from well defined areas of code, which are
1053 * listed in the exceptions table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 *
1055 * As the vast majority of faults will be valid we will only perform
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001056 * the source reference check when there is a possibility of a
1057 * deadlock. Attempt to lock the address space, if we cannot we then
1058 * validate the source. If this is invalid we can skip the address
1059 * space check, thus avoiding the deadlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 */
Nick Piggin92181f12009-01-20 04:24:26 +01001061 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
Andi Kleen66c58152006-01-11 22:44:09 +01001062 if ((error_code & PF_USER) == 0 &&
Nick Piggin92181f12009-01-20 04:24:26 +01001063 !search_exception_tables(regs->ip)) {
1064 bad_area_nosemaphore(regs, error_code, address);
1065 return;
1066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 down_read(&mm->mmap_sem);
Peter Zijlstra01006072009-01-29 16:02:12 +01001068 } else {
1069 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001070 * The above down_read_trylock() might have succeeded in
1071 * which case we'll have missed the might_sleep() from
1072 * down_read():
Peter Zijlstra01006072009-01-29 16:02:12 +01001073 */
1074 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
1076
1077 vma = find_vma(mm, address);
Nick Piggin92181f12009-01-20 04:24:26 +01001078 if (unlikely(!vma)) {
1079 bad_area(regs, error_code, address);
1080 return;
1081 }
1082 if (likely(vma->vm_start <= address))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 goto good_area;
Nick Piggin92181f12009-01-20 04:24:26 +01001084 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1085 bad_area(regs, error_code, address);
1086 return;
1087 }
Harvey Harrison33cb5242008-01-30 13:32:19 +01001088 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001089 /*
1090 * Accessing the stack below %sp is always a bug.
1091 * The large cushion allows instructions like enter
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001092 * and pusha to work. ("enter $65535, $31" pushes
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001093 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +02001094 */
Nick Piggin92181f12009-01-20 04:24:26 +01001095 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1096 bad_area(regs, error_code, address);
1097 return;
1098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
Nick Piggin92181f12009-01-20 04:24:26 +01001100 if (unlikely(expand_stack(vma, address))) {
1101 bad_area(regs, error_code, address);
1102 return;
1103 }
1104
1105 /*
1106 * Ok, we have a good vm_area for this memory access, so
1107 * we can handle it..
1108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109good_area:
Nick Piggin92181f12009-01-20 04:24:26 +01001110 write = error_code & PF_WRITE;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001111
Nick Piggin92181f12009-01-20 04:24:26 +01001112 if (unlikely(access_error(error_code, write, vma))) {
1113 bad_area_access_error(regs, error_code, address);
1114 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 }
1116
1117 /*
1118 * If for any reason at all we couldn't handle the fault,
1119 * make sure we exit gracefully rather than endlessly redo
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001120 * the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 */
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001122 fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001123
Nick Piggin83c54072007-07-19 01:47:05 -07001124 if (unlikely(fault & VM_FAULT_ERROR)) {
Nick Piggin92181f12009-01-20 04:24:26 +01001125 mm_fault_error(regs, error_code, address, fault);
1126 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001128
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001129 if (fault & VM_FAULT_MAJOR) {
Nick Piggin83c54072007-07-19 01:47:05 -07001130 tsk->maj_flt++;
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001131 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
Peter Zijlstra78f13e92009-04-08 15:01:33 +02001132 regs, address);
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001133 } else {
Nick Piggin83c54072007-07-19 01:47:05 -07001134 tsk->min_flt++;
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001135 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
Peter Zijlstra78f13e92009-04-08 15:01:33 +02001136 regs, address);
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001137 }
Harvey Harrisond729ab32008-01-30 13:33:23 +01001138
Ingo Molnar8c938f92009-02-20 22:12:18 +01001139 check_v8086_mode(regs, address, tsk);
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 up_read(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}