blob: c6acc632637417c193394da4881fa19112ace761 [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 Molnar940010c2009-06-11 17:55:42 +020013#include <linux/perf_counter.h> /* perf_swcounter_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_*(), ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
Harvey Harrison33cb5242008-01-30 13:32:19 +010018/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010019 * Page fault error code bits:
20 *
21 * bit 0 == 0: no page found 1: protection fault
22 * bit 1 == 0: read access 1: write access
23 * bit 2 == 0: kernel-mode access 1: user-mode access
24 * bit 3 == 1: use of reserved bit detected
25 * bit 4 == 1: fault was an instruction fetch
Harvey Harrison33cb5242008-01-30 13:32:19 +010026 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +010027enum x86_pf_error_code {
28
29 PF_PROT = 1 << 0,
30 PF_WRITE = 1 << 1,
31 PF_USER = 1 << 2,
32 PF_RSVD = 1 << 3,
33 PF_INSTR = 1 << 4,
34};
Andi Kleen66c58152006-01-11 22:44:09 +010035
Ingo Molnarb814d412009-02-20 22:32:10 +010036/*
Ingo Molnarb319eed2009-02-22 10:24:18 +010037 * Returns 0 if mmiotrace is disabled, or if the fault is not
38 * handled by mmiotrace:
Ingo Molnarb814d412009-02-20 22:32:10 +010039 */
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020040static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr)
Pekka Paalanen86069782008-05-12 21:20:56 +020041{
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020042 if (unlikely(is_kmmio_active()))
43 if (kmmio_handler(regs, addr) == 1)
44 return -1;
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020045 return 0;
Pekka Paalanen86069782008-05-12 21:20:56 +020046}
47
Christoph Hellwig74a0b572007-10-16 01:24:07 -070048static inline int notify_page_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070049{
Christoph Hellwig74a0b572007-10-16 01:24:07 -070050 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070051
Christoph Hellwig74a0b572007-10-16 01:24:07 -070052 /* kprobe_running() needs smp_processor_id() */
Ingo Molnarb1801812009-02-20 22:42:57 +010053 if (kprobes_built_in() && !user_mode_vm(regs)) {
Christoph Hellwig74a0b572007-10-16 01:24:07 -070054 preempt_disable();
55 if (kprobe_running() && kprobe_fault_handler(regs, 14))
56 ret = 1;
57 preempt_enable();
58 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070059
Christoph Hellwig74a0b572007-10-16 01:24:07 -070060 return ret;
Harvey Harrison33cb5242008-01-30 13:32:19 +010061}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070062
Harvey Harrison1dc85be2008-01-30 13:32:35 +010063/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010064 * Prefetch quirks:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010065 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010066 * 32-bit mode:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010067 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010068 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
69 * Check that here and ignore it.
70 *
71 * 64-bit mode:
72 *
73 * Sometimes the CPU reports invalid exceptions on prefetch.
74 * Check that here and ignore it.
75 *
76 * Opcode checker based on code by Richard Brunner.
Harvey Harrison1dc85be2008-01-30 13:32:35 +010077 */
Ingo Molnar107a0362009-02-20 20:37:05 +010078static inline int
79check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
80 unsigned char opcode, int *prefetch)
81{
82 unsigned char instr_hi = opcode & 0xf0;
83 unsigned char instr_lo = opcode & 0x0f;
84
85 switch (instr_hi) {
86 case 0x20:
87 case 0x30:
88 /*
89 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
90 * In X86_64 long mode, the CPU will signal invalid
91 * opcode if some of these prefixes are present so
92 * X86_64 will never get here anyway
93 */
94 return ((instr_lo & 7) == 0x6);
95#ifdef CONFIG_X86_64
96 case 0x40:
97 /*
98 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
99 * Need to figure out under what instruction mode the
100 * instruction was issued. Could check the LDT for lm,
101 * but for now it's good enough to assume that long
102 * mode only uses well known segments or kernel.
103 */
104 return (!user_mode(regs)) || (regs->cs == __USER_CS);
105#endif
106 case 0x60:
107 /* 0x64 thru 0x67 are valid prefixes in all modes. */
108 return (instr_lo & 0xC) == 0x4;
109 case 0xF0:
110 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
111 return !instr_lo || (instr_lo>>1) == 1;
112 case 0x00:
113 /* Prefetch instruction is 0x0F0D or 0x0F18 */
114 if (probe_kernel_address(instr, opcode))
115 return 0;
116
117 *prefetch = (instr_lo == 0xF) &&
118 (opcode == 0x0D || opcode == 0x18);
119 return 0;
120 default:
121 return 0;
122 }
123}
124
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100125static int
126is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
Harvey Harrison33cb5242008-01-30 13:32:19 +0100127{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100128 unsigned char *max_instr;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100129 unsigned char *instr;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100130 int prefetch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Ingo Molnar30853542008-03-27 21:29:09 +0100132 /*
133 * If it was a exec (instruction fetch) fault on NX page, then
134 * do not ignore the fault:
135 */
Andi Kleen66c58152006-01-11 22:44:09 +0100136 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100138
Ingo Molnar107a0362009-02-20 20:37:05 +0100139 instr = (void *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700140 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700142 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 return 0;
144
Ingo Molnar107a0362009-02-20 20:37:05 +0100145 while (instr < max_instr) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100146 unsigned char opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100148 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100149 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 instr++;
152
Ingo Molnar107a0362009-02-20 20:37:05 +0100153 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156 return prefetch;
157}
158
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100159static void
160force_sig_info_fault(int si_signo, int si_code, unsigned long address,
161 struct task_struct *tsk)
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100162{
163 siginfo_t info;
164
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100165 info.si_signo = si_signo;
166 info.si_errno = 0;
167 info.si_code = si_code;
168 info.si_addr = (void __user *)address;
169
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100170 force_sig_info(si_signo, &info, tsk);
171}
172
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100173DEFINE_SPINLOCK(pgd_lock);
174LIST_HEAD(pgd_list);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100175
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100176#ifdef CONFIG_X86_32
177static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
178{
179 unsigned index = pgd_index(address);
180 pgd_t *pgd_k;
181 pud_t *pud, *pud_k;
182 pmd_t *pmd, *pmd_k;
183
184 pgd += index;
185 pgd_k = init_mm.pgd + index;
186
187 if (!pgd_present(*pgd_k))
188 return NULL;
189
190 /*
191 * set_pgd(pgd, *pgd_k); here would be useless on PAE
192 * and redundant with the set_pmd() on non-PAE. As would
193 * set_pud.
194 */
195 pud = pud_offset(pgd, address);
196 pud_k = pud_offset(pgd_k, address);
197 if (!pud_present(*pud_k))
198 return NULL;
199
200 pmd = pmd_offset(pud, address);
201 pmd_k = pmd_offset(pud_k, address);
202 if (!pmd_present(*pmd_k))
203 return NULL;
204
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800205 if (!pmd_present(*pmd))
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100206 set_pmd(pmd, *pmd_k);
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800207 else
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100208 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100209
210 return pmd_k;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100211}
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100212
213void vmalloc_sync_all(void)
214{
215 unsigned long address;
216
217 if (SHARED_KERNEL_PMD)
218 return;
219
220 for (address = VMALLOC_START & PMD_MASK;
221 address >= TASK_SIZE && address < FIXADDR_TOP;
222 address += PMD_SIZE) {
223
224 unsigned long flags;
225 struct page *page;
226
227 spin_lock_irqsave(&pgd_lock, flags);
228 list_for_each_entry(page, &pgd_list, lru) {
229 if (!vmalloc_sync_one(page_address(page), address))
230 break;
231 }
232 spin_unlock_irqrestore(&pgd_lock, flags);
233 }
234}
235
236/*
237 * 32-bit:
238 *
239 * Handle a fault on the vmalloc or module mapping area
240 */
241static noinline int vmalloc_fault(unsigned long address)
242{
243 unsigned long pgd_paddr;
244 pmd_t *pmd_k;
245 pte_t *pte_k;
246
247 /* Make sure we are in vmalloc area: */
248 if (!(address >= VMALLOC_START && address < VMALLOC_END))
249 return -1;
250
251 /*
252 * Synchronize this task's top level page-table
253 * with the 'reference' page table.
254 *
255 * Do _not_ use "current" here. We might be inside
256 * an interrupt in the middle of a task switch..
257 */
258 pgd_paddr = read_cr3();
259 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
260 if (!pmd_k)
261 return -1;
262
263 pte_k = pte_offset_kernel(pmd_k, address);
264 if (!pte_present(*pte_k))
265 return -1;
266
267 return 0;
268}
269
270/*
271 * Did it hit the DOS screen memory VA from vm86 mode?
272 */
273static inline void
274check_v8086_mode(struct pt_regs *regs, unsigned long address,
275 struct task_struct *tsk)
276{
277 unsigned long bit;
278
279 if (!v8086_mode(regs))
280 return;
281
282 bit = (address - 0xA0000) >> PAGE_SHIFT;
283 if (bit < 32)
284 tsk->thread.screen_bitmap |= 1 << bit;
285}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Adrian Bunkcae30f82008-02-13 23:31:31 +0200287static void dump_pagetable(unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
Harvey Harrison1156e092008-01-30 13:34:10 +0100289 __typeof__(pte_val(__pte(0))) page;
290
291 page = read_cr3();
292 page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT];
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100293
Harvey Harrison1156e092008-01-30 13:34:10 +0100294#ifdef CONFIG_X86_PAE
295 printk("*pdpt = %016Lx ", page);
296 if ((page >> PAGE_SHIFT) < max_low_pfn
297 && page & _PAGE_PRESENT) {
298 page &= PAGE_MASK;
299 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100300 & (PTRS_PER_PMD - 1)];
Harvey Harrison1156e092008-01-30 13:34:10 +0100301 printk(KERN_CONT "*pde = %016Lx ", page);
302 page &= ~_PAGE_NX;
303 }
304#else
305 printk("*pde = %08lx ", page);
306#endif
307
308 /*
309 * We must not directly access the pte in the highpte
310 * case if the page table is located in highmem.
311 * And let's rather not kmap-atomic the pte, just in case
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100312 * it's allocated already:
Harvey Harrison1156e092008-01-30 13:34:10 +0100313 */
314 if ((page >> PAGE_SHIFT) < max_low_pfn
315 && (page & _PAGE_PRESENT)
316 && !(page & _PAGE_PSE)) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100317
Harvey Harrison1156e092008-01-30 13:34:10 +0100318 page &= PAGE_MASK;
319 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100320 & (PTRS_PER_PTE - 1)];
Harvey Harrison1156e092008-01-30 13:34:10 +0100321 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
322 }
323
324 printk("\n");
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100325}
326
327#else /* CONFIG_X86_64: */
328
329void vmalloc_sync_all(void)
330{
331 unsigned long address;
332
333 for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END;
334 address += PGDIR_SIZE) {
335
336 const pgd_t *pgd_ref = pgd_offset_k(address);
337 unsigned long flags;
338 struct page *page;
339
340 if (pgd_none(*pgd_ref))
341 continue;
342
343 spin_lock_irqsave(&pgd_lock, flags);
344 list_for_each_entry(page, &pgd_list, lru) {
345 pgd_t *pgd;
346 pgd = (pgd_t *)page_address(page) + pgd_index(address);
347 if (pgd_none(*pgd))
348 set_pgd(pgd, *pgd_ref);
349 else
350 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
351 }
352 spin_unlock_irqrestore(&pgd_lock, flags);
353 }
354}
355
356/*
357 * 64-bit:
358 *
359 * Handle a fault on the vmalloc area
360 *
361 * This assumes no large pages in there.
362 */
363static noinline int vmalloc_fault(unsigned long address)
364{
365 pgd_t *pgd, *pgd_ref;
366 pud_t *pud, *pud_ref;
367 pmd_t *pmd, *pmd_ref;
368 pte_t *pte, *pte_ref;
369
370 /* Make sure we are in vmalloc area: */
371 if (!(address >= VMALLOC_START && address < VMALLOC_END))
372 return -1;
373
374 /*
375 * Copy kernel mappings over when needed. This can also
376 * happen within a race in page table update. In the later
377 * case just flush:
378 */
379 pgd = pgd_offset(current->active_mm, address);
380 pgd_ref = pgd_offset_k(address);
381 if (pgd_none(*pgd_ref))
382 return -1;
383
384 if (pgd_none(*pgd))
385 set_pgd(pgd, *pgd_ref);
386 else
387 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
388
389 /*
390 * Below here mismatches are bugs because these lower tables
391 * are shared:
392 */
393
394 pud = pud_offset(pgd, address);
395 pud_ref = pud_offset(pgd_ref, address);
396 if (pud_none(*pud_ref))
397 return -1;
398
399 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
400 BUG();
401
402 pmd = pmd_offset(pud, address);
403 pmd_ref = pmd_offset(pud_ref, address);
404 if (pmd_none(*pmd_ref))
405 return -1;
406
407 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
408 BUG();
409
410 pte_ref = pte_offset_kernel(pmd_ref, address);
411 if (!pte_present(*pte_ref))
412 return -1;
413
414 pte = pte_offset_kernel(pmd, address);
415
416 /*
417 * Don't use pte_page here, because the mappings can point
418 * outside mem_map, and the NUMA hash lookup cannot handle
419 * that:
420 */
421 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
422 BUG();
423
424 return 0;
425}
426
427static const char errata93_warning[] =
428KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
429KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
430KERN_ERR "******* Please consider a BIOS update.\n"
431KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
432
433/*
434 * No vm86 mode in 64-bit mode:
435 */
436static inline void
437check_v8086_mode(struct pt_regs *regs, unsigned long address,
438 struct task_struct *tsk)
439{
440}
441
442static int bad_address(void *p)
443{
444 unsigned long dummy;
445
446 return probe_kernel_address((unsigned long *)p, dummy);
447}
448
449static void dump_pagetable(unsigned long address)
450{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 pgd_t *pgd;
452 pud_t *pud;
453 pmd_t *pmd;
454 pte_t *pte;
455
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200456 pgd = (pgd_t *)read_cr3();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Harvey Harrison33cb5242008-01-30 13:32:19 +0100458 pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 pgd += pgd_index(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100461 if (bad_address(pgd))
462 goto bad;
463
Jan Beulichd646bce2006-02-03 21:51:47 +0100464 printk("PGD %lx ", pgd_val(*pgd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100465
466 if (!pgd_present(*pgd))
467 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Andi Kleend2ae5b52006-06-26 13:57:56 +0200469 pud = pud_offset(pgd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100470 if (bad_address(pud))
471 goto bad;
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 printk("PUD %lx ", pud_val(*pud));
Andi Kleenb5360222008-02-04 16:48:09 +0100474 if (!pud_present(*pud) || pud_large(*pud))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100475 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477 pmd = pmd_offset(pud, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100478 if (bad_address(pmd))
479 goto bad;
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 printk("PMD %lx ", pmd_val(*pmd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100482 if (!pmd_present(*pmd) || pmd_large(*pmd))
483 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485 pte = pte_offset_kernel(pmd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100486 if (bad_address(pte))
487 goto bad;
488
Harvey Harrison33cb5242008-01-30 13:32:19 +0100489 printk("PTE %lx", pte_val(*pte));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100490out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 printk("\n");
492 return;
493bad:
494 printk("BAD\n");
495}
496
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100497#endif /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100499/*
500 * Workaround for K8 erratum #93 & buggy BIOS.
501 *
502 * BIOS SMM functions are required to use a specific workaround
503 * to avoid corruption of the 64bit RIP register on C stepping K8.
504 *
505 * A lot of BIOS that didn't get tested properly miss this.
506 *
507 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
508 * Try to work around it here.
509 *
510 * Note we only handle faults in kernel here.
511 * Does nothing on 32-bit.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100512 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100513static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100515#ifdef CONFIG_X86_64
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100516 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100518
Harvey Harrison33cb5242008-01-30 13:32:19 +0100519 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100523 if ((address >= (u64)_stext && address <= (u64)_etext) ||
524 (address >= MODULES_VADDR && address <= MODULES_END)) {
Ingo Molnara454ab32009-05-03 10:09:03 +0200525 printk_once(errata93_warning);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100526 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return 1;
528 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100529#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100531}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Harvey Harrison35f32662008-01-30 13:34:09 +0100533/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100534 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
535 * to illegal addresses >4GB.
536 *
537 * We catch this in the page fault handler because these addresses
538 * are not reachable. Just detect this case and return. Any code
Harvey Harrison35f32662008-01-30 13:34:09 +0100539 * segment in LDT is compatibility mode.
540 */
541static int is_errata100(struct pt_regs *regs, unsigned long address)
542{
543#ifdef CONFIG_X86_64
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100544 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
Harvey Harrison35f32662008-01-30 13:34:09 +0100545 return 1;
546#endif
547 return 0;
548}
549
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100550static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
551{
552#ifdef CONFIG_X86_F00F_BUG
553 unsigned long nr;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100554
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100555 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100556 * Pentium F0 0F C7 C8 bug workaround:
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100557 */
558 if (boot_cpu_data.f00f_bug) {
559 nr = (address - idt_descr.address) >> 3;
560
561 if (nr == 6) {
562 do_invalid_op(regs, 0);
563 return 1;
564 }
565 }
566#endif
567 return 0;
568}
569
Ingo Molnar8f766142009-02-20 23:00:29 +0100570static const char nx_warning[] = KERN_CRIT
571"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
572
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100573static void
574show_fault_oops(struct pt_regs *regs, unsigned long error_code,
575 unsigned long address)
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100576{
Harvey Harrison1156e092008-01-30 13:34:10 +0100577 if (!oops_may_print())
578 return;
579
Harvey Harrison1156e092008-01-30 13:34:10 +0100580 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100581 unsigned int level;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100582
Harvey Harrison1156e092008-01-30 13:34:10 +0100583 pte_t *pte = lookup_address(address, &level);
584
Ingo Molnar8f766142009-02-20 23:00:29 +0100585 if (pte && pte_present(*pte) && !pte_exec(*pte))
586 printk(nx_warning, current_uid());
Harvey Harrison1156e092008-01-30 13:34:10 +0100587 }
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100588
Harvey Harrison1156e092008-01-30 13:34:10 +0100589 printk(KERN_ALERT "BUG: unable to handle kernel ");
590 if (address < PAGE_SIZE)
591 printk(KERN_CONT "NULL pointer dereference");
592 else
593 printk(KERN_CONT "paging request");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100594
Vegard Nossumf294a8c2008-07-01 15:38:13 +0200595 printk(KERN_CONT " at %p\n", (void *) address);
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100596 printk(KERN_ALERT "IP:");
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100597 printk_address(regs->ip, 1);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100598
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100599 dump_pagetable(address);
600}
601
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100602static noinline void
603pgtable_bad(struct pt_regs *regs, unsigned long error_code,
604 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100606 struct task_struct *tsk;
607 unsigned long flags;
608 int sig;
609
610 flags = oops_begin();
611 tsk = current;
612 sig = SIGKILL;
Jan Beulich12091402005-09-12 18:49:24 +0200613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
Nick Piggin92181f12009-01-20 04:24:26 +0100615 tsk->comm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 dump_pagetable(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100617
618 tsk->thread.cr2 = address;
619 tsk->thread.trap_no = 14;
620 tsk->thread.error_code = error_code;
621
Jan Beulich22f59912008-01-30 13:31:23 +0100622 if (__die("Bad pagetable", regs, error_code))
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200623 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100624
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200625 oops_end(flags, regs, sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100628static noinline void
629no_context(struct pt_regs *regs, unsigned long error_code,
630 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100631{
632 struct task_struct *tsk = current;
Ingo Molnar19803072009-01-21 10:39:51 +0100633 unsigned long *stackend;
Nick Piggin92181f12009-01-20 04:24:26 +0100634 unsigned long flags;
635 int sig;
Nick Piggin92181f12009-01-20 04:24:26 +0100636
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100637 /* Are we prepared to handle this kernel fault? */
Nick Piggin92181f12009-01-20 04:24:26 +0100638 if (fixup_exception(regs))
639 return;
640
641 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100642 * 32-bit:
Nick Piggin92181f12009-01-20 04:24:26 +0100643 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100644 * Valid to do another page fault here, because if this fault
645 * had been triggered by is_prefetch fixup_exception would have
646 * handled it.
647 *
648 * 64-bit:
649 *
650 * Hall of shame of CPU/BIOS bugs.
Nick Piggin92181f12009-01-20 04:24:26 +0100651 */
652 if (is_prefetch(regs, error_code, address))
653 return;
654
655 if (is_errata93(regs, address))
656 return;
657
658 /*
659 * Oops. The kernel tried to access some bad page. We'll have to
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100660 * terminate things with extreme prejudice:
Nick Piggin92181f12009-01-20 04:24:26 +0100661 */
Nick Piggin92181f12009-01-20 04:24:26 +0100662 flags = oops_begin();
Nick Piggin92181f12009-01-20 04:24:26 +0100663
664 show_fault_oops(regs, error_code, address);
665
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100666 stackend = end_of_stack(tsk);
Ingo Molnar19803072009-01-21 10:39:51 +0100667 if (*stackend != STACK_END_MAGIC)
668 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
669
Ingo Molnar1cc99542009-02-20 23:07:48 +0100670 tsk->thread.cr2 = address;
671 tsk->thread.trap_no = 14;
672 tsk->thread.error_code = error_code;
Nick Piggin92181f12009-01-20 04:24:26 +0100673
Nick Piggin92181f12009-01-20 04:24:26 +0100674 sig = SIGKILL;
675 if (__die("Oops", regs, error_code))
676 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100677
Nick Piggin92181f12009-01-20 04:24:26 +0100678 /* Executive summary in case the body of the oops scrolled away */
679 printk(KERN_EMERG "CR2: %016lx\n", address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100680
Nick Piggin92181f12009-01-20 04:24:26 +0100681 oops_end(flags, regs, sig);
Nick Piggin92181f12009-01-20 04:24:26 +0100682}
683
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100684/*
685 * Print out info about fatal segfaults, if the show_unhandled_signals
686 * sysctl is set:
687 */
688static inline void
689show_signal_msg(struct pt_regs *regs, unsigned long error_code,
690 unsigned long address, struct task_struct *tsk)
691{
692 if (!unhandled_signal(tsk, SIGSEGV))
693 return;
694
695 if (!printk_ratelimit())
696 return;
697
698 printk(KERN_CONT "%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
699 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
700 tsk->comm, task_pid_nr(tsk), address,
701 (void *)regs->ip, (void *)regs->sp, error_code);
702
703 print_vma_addr(KERN_CONT " in ", regs->ip);
704
705 printk(KERN_CONT "\n");
706}
707
708static void
709__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
710 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100711{
712 struct task_struct *tsk = current;
713
714 /* User mode accesses just cause a SIGSEGV */
715 if (error_code & PF_USER) {
716 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100717 * It's possible to have interrupts off here:
Nick Piggin92181f12009-01-20 04:24:26 +0100718 */
719 local_irq_enable();
720
721 /*
722 * Valid to do another page fault here because this one came
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100723 * from user space:
Nick Piggin92181f12009-01-20 04:24:26 +0100724 */
725 if (is_prefetch(regs, error_code, address))
726 return;
727
728 if (is_errata100(regs, address))
729 return;
730
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100731 if (unlikely(show_unhandled_signals))
732 show_signal_msg(regs, error_code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100733
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100734 /* Kernel addresses are always protection faults: */
735 tsk->thread.cr2 = address;
736 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
737 tsk->thread.trap_no = 14;
738
Nick Piggin92181f12009-01-20 04:24:26 +0100739 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100740
Nick Piggin92181f12009-01-20 04:24:26 +0100741 return;
742 }
743
744 if (is_f00f_bug(regs, address))
745 return;
746
747 no_context(regs, error_code, address);
748}
749
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100750static noinline void
751bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
752 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100753{
754 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
755}
756
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100757static void
758__bad_area(struct pt_regs *regs, unsigned long error_code,
759 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100760{
761 struct mm_struct *mm = current->mm;
762
763 /*
764 * Something tried to access memory that isn't in our memory map..
765 * Fix it, but check if it's kernel or user first..
766 */
767 up_read(&mm->mmap_sem);
768
769 __bad_area_nosemaphore(regs, error_code, address, si_code);
770}
771
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100772static noinline void
773bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100774{
775 __bad_area(regs, error_code, address, SEGV_MAPERR);
776}
777
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100778static noinline void
779bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
780 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100781{
782 __bad_area(regs, error_code, address, SEGV_ACCERR);
783}
784
785/* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100786static void
787out_of_memory(struct pt_regs *regs, unsigned long error_code,
788 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100789{
790 /*
791 * We ran out of memory, call the OOM killer, and return the userspace
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100792 * (which will retry the fault, or kill us if we got oom-killed):
Nick Piggin92181f12009-01-20 04:24:26 +0100793 */
794 up_read(&current->mm->mmap_sem);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100795
Nick Piggin92181f12009-01-20 04:24:26 +0100796 pagefault_out_of_memory();
797}
798
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100799static void
800do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100801{
802 struct task_struct *tsk = current;
803 struct mm_struct *mm = tsk->mm;
804
805 up_read(&mm->mmap_sem);
806
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100807 /* Kernel mode? Handle exceptions or die: */
Nick Piggin92181f12009-01-20 04:24:26 +0100808 if (!(error_code & PF_USER))
809 no_context(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100810
Ingo Molnarcd1b68f2009-02-20 23:39:02 +0100811 /* User-space => ok to do another page fault: */
Nick Piggin92181f12009-01-20 04:24:26 +0100812 if (is_prefetch(regs, error_code, address))
813 return;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100814
815 tsk->thread.cr2 = address;
816 tsk->thread.error_code = error_code;
817 tsk->thread.trap_no = 14;
818
Nick Piggin92181f12009-01-20 04:24:26 +0100819 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
820}
821
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100822static noinline void
823mm_fault_error(struct pt_regs *regs, unsigned long error_code,
824 unsigned long address, unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100825{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100826 if (fault & VM_FAULT_OOM) {
Nick Piggin92181f12009-01-20 04:24:26 +0100827 out_of_memory(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100828 } else {
829 if (fault & VM_FAULT_SIGBUS)
830 do_sigbus(regs, error_code, address);
831 else
832 BUG();
833 }
Nick Piggin92181f12009-01-20 04:24:26 +0100834}
835
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100836static int spurious_fault_check(unsigned long error_code, pte_t *pte)
837{
838 if ((error_code & PF_WRITE) && !pte_write(*pte))
839 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100840
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100841 if ((error_code & PF_INSTR) && !pte_exec(*pte))
842 return 0;
843
844 return 1;
845}
846
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100848 * Handle a spurious fault caused by a stale TLB entry.
849 *
850 * This allows us to lazily refresh the TLB when increasing the
851 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
852 * eagerly is very expensive since that implies doing a full
853 * cross-processor TLB flush, even if no stale TLB entries exist
854 * on other processors.
855 *
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100856 * There are no security implications to leaving a stale TLB when
857 * increasing the permissions on a page.
858 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100859static noinline int
860spurious_fault(unsigned long error_code, unsigned long address)
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100861{
862 pgd_t *pgd;
863 pud_t *pud;
864 pmd_t *pmd;
865 pte_t *pte;
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500866 int ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100867
868 /* Reserved-bit violation or user access to kernel space? */
869 if (error_code & (PF_USER | PF_RSVD))
870 return 0;
871
872 pgd = init_mm.pgd + pgd_index(address);
873 if (!pgd_present(*pgd))
874 return 0;
875
876 pud = pud_offset(pgd, address);
877 if (!pud_present(*pud))
878 return 0;
879
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100880 if (pud_large(*pud))
881 return spurious_fault_check(error_code, (pte_t *) pud);
882
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100883 pmd = pmd_offset(pud, address);
884 if (!pmd_present(*pmd))
885 return 0;
886
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100887 if (pmd_large(*pmd))
888 return spurious_fault_check(error_code, (pte_t *) pmd);
889
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100890 pte = pte_offset_kernel(pmd, address);
891 if (!pte_present(*pte))
892 return 0;
893
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500894 ret = spurious_fault_check(error_code, pte);
895 if (!ret)
896 return 0;
897
898 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100899 * Make sure we have permissions in PMD.
900 * If not, then there's a bug in the page tables:
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500901 */
902 ret = spurious_fault_check(error_code, (pte_t *) pmd);
903 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100904
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500905 return ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100906}
907
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200908int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100910static inline int
911access_error(unsigned long error_code, int write, struct vm_area_struct *vma)
Nick Piggin92181f12009-01-20 04:24:26 +0100912{
913 if (write) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100914 /* write, present and write, not present: */
Nick Piggin92181f12009-01-20 04:24:26 +0100915 if (unlikely(!(vma->vm_flags & VM_WRITE)))
916 return 1;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100917 return 0;
Nick Piggin92181f12009-01-20 04:24:26 +0100918 }
919
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100920 /* read, present: */
921 if (unlikely(error_code & PF_PROT))
922 return 1;
923
924 /* read, not present: */
925 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
926 return 1;
927
Nick Piggin92181f12009-01-20 04:24:26 +0100928 return 0;
929}
930
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800931static int fault_in_kernel_space(unsigned long address)
932{
Ingo Molnard9517342009-02-20 23:32:28 +0100933 return address >= TASK_SIZE_MAX;
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800934}
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936/*
937 * This routine handles page faults. It determines the address,
938 * and the problem, and then passes it off to one of the appropriate
939 * routines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 */
Ingo Molnarc3731c62009-02-20 23:22:34 +0100941dotraplinkage void __kprobes
942do_page_fault(struct pt_regs *regs, unsigned long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943{
Harvey Harrison33cb5242008-01-30 13:32:19 +0100944 struct vm_area_struct *vma;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100945 struct task_struct *tsk;
946 unsigned long address;
947 struct mm_struct *mm;
Nick Piggin92181f12009-01-20 04:24:26 +0100948 int write;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100949 int fault;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100951 tsk = current;
952 mm = tsk->mm;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100953
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100954 prefetchw(&mm->mmap_sem);
955
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100956 /* Get the faulting address: */
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200957 address = read_cr2();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +0200959 if (unlikely(kmmio_fault(regs, address)))
Pekka Paalanen86069782008-05-12 21:20:56 +0200960 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
962 /*
963 * We fault-in kernel-space virtual memory on-demand. The
964 * 'reference' page table is init_mm.pgd.
965 *
966 * NOTE! We MUST NOT take any locks for this case. We may
967 * be in an interrupt or a critical region, and should
968 * only copy the information from the master page table,
969 * nothing more.
970 *
971 * This verifies that the fault happens in kernel space
972 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +0100973 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 */
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800975 if (unlikely(fault_in_kernel_space(address))) {
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100976 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
977 vmalloc_fault(address) >= 0)
978 return;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100979
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100980 /* Can handle a stale RO->RW TLB: */
Nick Piggin92181f12009-01-20 04:24:26 +0100981 if (spurious_fault(error_code, address))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100982 return;
983
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100984 /* kprobes don't want to hook the spurious faults: */
Masami Hiramatsu9be260a2009-02-05 17:12:39 -0500985 if (notify_page_fault(regs))
986 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100987 /*
988 * Don't take the mm semaphore here. If we fixup a prefetch
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100989 * fault we could otherwise deadlock:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100990 */
Nick Piggin92181f12009-01-20 04:24:26 +0100991 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100992
Nick Piggin92181f12009-01-20 04:24:26 +0100993 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100994 }
995
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100996 /* kprobes don't want to hook the spurious faults: */
Ingo Molnarf8a6b2b2009-02-13 09:44:22 +0100997 if (unlikely(notify_page_fault(regs)))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -0500998 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100999 /*
Linus Torvalds891cffb2008-10-12 13:16:12 -07001000 * It's safe to allow irq's after cr2 has been saved and the
1001 * vmalloc fault has been handled.
1002 *
1003 * User-mode registers count as a user access even for any
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001004 * potential system fault or CPU buglet:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001005 */
Linus Torvalds891cffb2008-10-12 13:16:12 -07001006 if (user_mode_vm(regs)) {
1007 local_irq_enable();
1008 error_code |= PF_USER;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001009 } else {
1010 if (regs->flags & X86_EFLAGS_IF)
1011 local_irq_enable();
1012 }
Jan Beulich8c914cb2006-03-25 16:29:40 +01001013
Andi Kleen66c58152006-01-11 22:44:09 +01001014 if (unlikely(error_code & PF_RSVD))
Nick Piggin92181f12009-01-20 04:24:26 +01001015 pgtable_bad(regs, error_code, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Peter Zijlstraf4dbfa82009-06-11 14:06:28 +02001017 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
Peter Zijlstra7dd1fcc2009-03-13 12:21:33 +01001018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001020 * If we're in an interrupt, have no user context or are running
1021 * in an atomic region then we must not take the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
Nick Piggin92181f12009-01-20 04:24:26 +01001023 if (unlikely(in_atomic() || !mm)) {
1024 bad_area_nosemaphore(regs, error_code, address);
1025 return;
1026 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Ingo Molnar3a1dfe62008-10-13 17:49:02 +02001028 /*
1029 * When running in the kernel we expect faults to occur only to
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001030 * addresses in user space. All other faults represent errors in
1031 * the kernel and should generate an OOPS. Unfortunately, in the
1032 * case of an erroneous fault occurring in a code path which already
1033 * holds mmap_sem we will deadlock attempting to validate the fault
1034 * against the address space. Luckily the kernel only validly
1035 * references user space from well defined areas of code, which are
1036 * listed in the exceptions table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 *
1038 * As the vast majority of faults will be valid we will only perform
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001039 * the source reference check when there is a possibility of a
1040 * deadlock. Attempt to lock the address space, if we cannot we then
1041 * validate the source. If this is invalid we can skip the address
1042 * space check, thus avoiding the deadlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Nick Piggin92181f12009-01-20 04:24:26 +01001044 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
Andi Kleen66c58152006-01-11 22:44:09 +01001045 if ((error_code & PF_USER) == 0 &&
Nick Piggin92181f12009-01-20 04:24:26 +01001046 !search_exception_tables(regs->ip)) {
1047 bad_area_nosemaphore(regs, error_code, address);
1048 return;
1049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 down_read(&mm->mmap_sem);
Peter Zijlstra01006072009-01-29 16:02:12 +01001051 } else {
1052 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001053 * The above down_read_trylock() might have succeeded in
1054 * which case we'll have missed the might_sleep() from
1055 * down_read():
Peter Zijlstra01006072009-01-29 16:02:12 +01001056 */
1057 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 }
1059
1060 vma = find_vma(mm, address);
Nick Piggin92181f12009-01-20 04:24:26 +01001061 if (unlikely(!vma)) {
1062 bad_area(regs, error_code, address);
1063 return;
1064 }
1065 if (likely(vma->vm_start <= address))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 goto good_area;
Nick Piggin92181f12009-01-20 04:24:26 +01001067 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1068 bad_area(regs, error_code, address);
1069 return;
1070 }
Harvey Harrison33cb5242008-01-30 13:32:19 +01001071 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001072 /*
1073 * Accessing the stack below %sp is always a bug.
1074 * The large cushion allows instructions like enter
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001075 * and pusha to work. ("enter $65535, $31" pushes
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001076 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +02001077 */
Nick Piggin92181f12009-01-20 04:24:26 +01001078 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1079 bad_area(regs, error_code, address);
1080 return;
1081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
Nick Piggin92181f12009-01-20 04:24:26 +01001083 if (unlikely(expand_stack(vma, address))) {
1084 bad_area(regs, error_code, address);
1085 return;
1086 }
1087
1088 /*
1089 * Ok, we have a good vm_area for this memory access, so
1090 * we can handle it..
1091 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092good_area:
Nick Piggin92181f12009-01-20 04:24:26 +01001093 write = error_code & PF_WRITE;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001094
Nick Piggin92181f12009-01-20 04:24:26 +01001095 if (unlikely(access_error(error_code, write, vma))) {
1096 bad_area_access_error(regs, error_code, address);
1097 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 }
1099
1100 /*
1101 * If for any reason at all we couldn't handle the fault,
1102 * make sure we exit gracefully rather than endlessly redo
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001103 * the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 */
Nick Piggin83c54072007-07-19 01:47:05 -07001105 fault = handle_mm_fault(mm, vma, address, write);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001106
Nick Piggin83c54072007-07-19 01:47:05 -07001107 if (unlikely(fault & VM_FAULT_ERROR)) {
Nick Piggin92181f12009-01-20 04:24:26 +01001108 mm_fault_error(regs, error_code, address, fault);
1109 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001111
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001112 if (fault & VM_FAULT_MAJOR) {
Nick Piggin83c54072007-07-19 01:47:05 -07001113 tsk->maj_flt++;
Peter Zijlstraf4dbfa82009-06-11 14:06:28 +02001114 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
Peter Zijlstra78f13e92009-04-08 15:01:33 +02001115 regs, address);
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001116 } else {
Nick Piggin83c54072007-07-19 01:47:05 -07001117 tsk->min_flt++;
Peter Zijlstraf4dbfa82009-06-11 14:06:28 +02001118 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
Peter Zijlstra78f13e92009-04-08 15:01:33 +02001119 regs, address);
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001120 }
Harvey Harrisond729ab32008-01-30 13:33:23 +01001121
Ingo Molnar8c938f92009-02-20 22:12:18 +01001122 check_v8086_mode(regs, address, tsk);
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 up_read(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}