blob: 20e3f8702d1e5701b130d2891e23bf39fbe81663 [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 */
Andi Kleenf672b492010-09-27 22:05:55 +020014#include <linux/hugetlb.h> /* hstate_index_to_shift */
Linus Torvalds1da177e2005-04-16 15:20:36 -070015
Ingo Molnara2bcd472009-03-29 23:47:48 +020016#include <asm/traps.h> /* dotraplinkage, ... */
17#include <asm/pgalloc.h> /* pgd_*(), ... */
Vegard Nossumf8561292008-04-04 00:53:23 +020018#include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
Harvey Harrison33cb5242008-01-30 13:32:19 +010020/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010021 * Page fault error code bits:
22 *
23 * bit 0 == 0: no page found 1: protection fault
24 * bit 1 == 0: read access 1: write access
25 * bit 2 == 0: kernel-mode access 1: user-mode access
26 * bit 3 == 1: use of reserved bit detected
27 * bit 4 == 1: fault was an instruction fetch
Harvey Harrison33cb5242008-01-30 13:32:19 +010028 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +010029enum x86_pf_error_code {
30
31 PF_PROT = 1 << 0,
32 PF_WRITE = 1 << 1,
33 PF_USER = 1 << 2,
34 PF_RSVD = 1 << 3,
35 PF_INSTR = 1 << 4,
36};
Andi Kleen66c58152006-01-11 22:44:09 +010037
Ingo Molnarb814d412009-02-20 22:32:10 +010038/*
Ingo Molnarb319eed2009-02-22 10:24:18 +010039 * Returns 0 if mmiotrace is disabled, or if the fault is not
40 * handled by mmiotrace:
Ingo Molnarb814d412009-02-20 22:32:10 +010041 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -040042static inline int __kprobes
43kmmio_fault(struct pt_regs *regs, unsigned long addr)
Pekka Paalanen86069782008-05-12 21:20:56 +020044{
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020045 if (unlikely(is_kmmio_active()))
46 if (kmmio_handler(regs, addr) == 1)
47 return -1;
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020048 return 0;
Pekka Paalanen86069782008-05-12 21:20:56 +020049}
50
Masami Hiramatsu62c92952009-08-27 13:23:11 -040051static inline int __kprobes notify_page_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070052{
Christoph Hellwig74a0b572007-10-16 01:24:07 -070053 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070054
Christoph Hellwig74a0b572007-10-16 01:24:07 -070055 /* kprobe_running() needs smp_processor_id() */
Ingo Molnarb1801812009-02-20 22:42:57 +010056 if (kprobes_built_in() && !user_mode_vm(regs)) {
Christoph Hellwig74a0b572007-10-16 01:24:07 -070057 preempt_disable();
58 if (kprobe_running() && kprobe_fault_handler(regs, 14))
59 ret = 1;
60 preempt_enable();
61 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070062
Christoph Hellwig74a0b572007-10-16 01:24:07 -070063 return ret;
Harvey Harrison33cb5242008-01-30 13:32:19 +010064}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070065
Harvey Harrison1dc85be2008-01-30 13:32:35 +010066/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010067 * Prefetch quirks:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010068 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010069 * 32-bit mode:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010070 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010071 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
72 * Check that here and ignore it.
73 *
74 * 64-bit mode:
75 *
76 * Sometimes the CPU reports invalid exceptions on prefetch.
77 * Check that here and ignore it.
78 *
79 * Opcode checker based on code by Richard Brunner.
Harvey Harrison1dc85be2008-01-30 13:32:35 +010080 */
Ingo Molnar107a0362009-02-20 20:37:05 +010081static inline int
82check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
83 unsigned char opcode, int *prefetch)
84{
85 unsigned char instr_hi = opcode & 0xf0;
86 unsigned char instr_lo = opcode & 0x0f;
87
88 switch (instr_hi) {
89 case 0x20:
90 case 0x30:
91 /*
92 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
93 * In X86_64 long mode, the CPU will signal invalid
94 * opcode if some of these prefixes are present so
95 * X86_64 will never get here anyway
96 */
97 return ((instr_lo & 7) == 0x6);
98#ifdef CONFIG_X86_64
99 case 0x40:
100 /*
101 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
102 * Need to figure out under what instruction mode the
103 * instruction was issued. Could check the LDT for lm,
104 * but for now it's good enough to assume that long
105 * mode only uses well known segments or kernel.
106 */
107 return (!user_mode(regs)) || (regs->cs == __USER_CS);
108#endif
109 case 0x60:
110 /* 0x64 thru 0x67 are valid prefixes in all modes. */
111 return (instr_lo & 0xC) == 0x4;
112 case 0xF0:
113 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
114 return !instr_lo || (instr_lo>>1) == 1;
115 case 0x00:
116 /* Prefetch instruction is 0x0F0D or 0x0F18 */
117 if (probe_kernel_address(instr, opcode))
118 return 0;
119
120 *prefetch = (instr_lo == 0xF) &&
121 (opcode == 0x0D || opcode == 0x18);
122 return 0;
123 default:
124 return 0;
125 }
126}
127
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100128static int
129is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
Harvey Harrison33cb5242008-01-30 13:32:19 +0100130{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100131 unsigned char *max_instr;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100132 unsigned char *instr;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100133 int prefetch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Ingo Molnar30853542008-03-27 21:29:09 +0100135 /*
136 * If it was a exec (instruction fetch) fault on NX page, then
137 * do not ignore the fault:
138 */
Andi Kleen66c58152006-01-11 22:44:09 +0100139 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100141
Ingo Molnar107a0362009-02-20 20:37:05 +0100142 instr = (void *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700143 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700145 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 return 0;
147
Ingo Molnar107a0362009-02-20 20:37:05 +0100148 while (instr < max_instr) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100149 unsigned char opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100151 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100152 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 instr++;
155
Ingo Molnar107a0362009-02-20 20:37:05 +0100156 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 }
159 return prefetch;
160}
161
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100162static void
163force_sig_info_fault(int si_signo, int si_code, unsigned long address,
Andi Kleenf672b492010-09-27 22:05:55 +0200164 struct task_struct *tsk, int fault)
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100165{
Andi Kleenf672b492010-09-27 22:05:55 +0200166 unsigned lsb = 0;
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100167 siginfo_t info;
168
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100169 info.si_signo = si_signo;
170 info.si_errno = 0;
171 info.si_code = si_code;
172 info.si_addr = (void __user *)address;
Andi Kleenf672b492010-09-27 22:05:55 +0200173 if (fault & VM_FAULT_HWPOISON_LARGE)
174 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
175 if (fault & VM_FAULT_HWPOISON)
176 lsb = PAGE_SHIFT;
177 info.si_addr_lsb = lsb;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100178
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100179 force_sig_info(si_signo, &info, tsk);
180}
181
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100182DEFINE_SPINLOCK(pgd_lock);
183LIST_HEAD(pgd_list);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100184
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100185#ifdef CONFIG_X86_32
186static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
187{
188 unsigned index = pgd_index(address);
189 pgd_t *pgd_k;
190 pud_t *pud, *pud_k;
191 pmd_t *pmd, *pmd_k;
192
193 pgd += index;
194 pgd_k = init_mm.pgd + index;
195
196 if (!pgd_present(*pgd_k))
197 return NULL;
198
199 /*
200 * set_pgd(pgd, *pgd_k); here would be useless on PAE
201 * and redundant with the set_pmd() on non-PAE. As would
202 * set_pud.
203 */
204 pud = pud_offset(pgd, address);
205 pud_k = pud_offset(pgd_k, address);
206 if (!pud_present(*pud_k))
207 return NULL;
208
209 pmd = pmd_offset(pud, address);
210 pmd_k = pmd_offset(pud_k, address);
211 if (!pmd_present(*pmd_k))
212 return NULL;
213
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800214 if (!pmd_present(*pmd))
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100215 set_pmd(pmd, *pmd_k);
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800216 else
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100217 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100218
219 return pmd_k;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100220}
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100221
222void vmalloc_sync_all(void)
223{
224 unsigned long address;
225
226 if (SHARED_KERNEL_PMD)
227 return;
228
229 for (address = VMALLOC_START & PMD_MASK;
230 address >= TASK_SIZE && address < FIXADDR_TOP;
231 address += PMD_SIZE) {
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100232 struct page *page;
233
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800234 spin_lock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100235 list_for_each_entry(page, &pgd_list, lru) {
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700236 spinlock_t *pgt_lock;
Borislav Petkovf01f7c52010-10-19 22:17:37 +0000237 pmd_t *ret;
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700238
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800239 /* the pgt_lock only for Xen */
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700240 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
241
242 spin_lock(pgt_lock);
243 ret = vmalloc_sync_one(page_address(page), address);
244 spin_unlock(pgt_lock);
245
246 if (!ret)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100247 break;
248 }
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800249 spin_unlock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100250 }
251}
252
253/*
254 * 32-bit:
255 *
256 * Handle a fault on the vmalloc or module mapping area
257 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400258static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100259{
260 unsigned long pgd_paddr;
261 pmd_t *pmd_k;
262 pte_t *pte_k;
263
264 /* Make sure we are in vmalloc area: */
265 if (!(address >= VMALLOC_START && address < VMALLOC_END))
266 return -1;
267
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200268 WARN_ON_ONCE(in_nmi());
269
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100270 /*
271 * Synchronize this task's top level page-table
272 * with the 'reference' page table.
273 *
274 * Do _not_ use "current" here. We might be inside
275 * an interrupt in the middle of a task switch..
276 */
277 pgd_paddr = read_cr3();
278 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
279 if (!pmd_k)
280 return -1;
281
282 pte_k = pte_offset_kernel(pmd_k, address);
283 if (!pte_present(*pte_k))
284 return -1;
285
286 return 0;
287}
288
289/*
290 * Did it hit the DOS screen memory VA from vm86 mode?
291 */
292static inline void
293check_v8086_mode(struct pt_regs *regs, unsigned long address,
294 struct task_struct *tsk)
295{
296 unsigned long bit;
297
298 if (!v8086_mode(regs))
299 return;
300
301 bit = (address - 0xA0000) >> PAGE_SHIFT;
302 if (bit < 32)
303 tsk->thread.screen_bitmap |= 1 << bit;
304}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Akinobu Mita087975b2009-06-27 15:35:15 +0900306static bool low_pfn(unsigned long pfn)
307{
308 return pfn < max_low_pfn;
309}
310
Adrian Bunkcae30f82008-02-13 23:31:31 +0200311static void dump_pagetable(unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Akinobu Mita087975b2009-06-27 15:35:15 +0900313 pgd_t *base = __va(read_cr3());
314 pgd_t *pgd = &base[pgd_index(address)];
315 pmd_t *pmd;
316 pte_t *pte;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100317
Harvey Harrison1156e092008-01-30 13:34:10 +0100318#ifdef CONFIG_X86_PAE
Akinobu Mita087975b2009-06-27 15:35:15 +0900319 printk("*pdpt = %016Lx ", pgd_val(*pgd));
320 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
321 goto out;
Harvey Harrison1156e092008-01-30 13:34:10 +0100322#endif
Akinobu Mita087975b2009-06-27 15:35:15 +0900323 pmd = pmd_offset(pud_offset(pgd, address), address);
324 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
Harvey Harrison1156e092008-01-30 13:34:10 +0100325
326 /*
327 * We must not directly access the pte in the highpte
328 * case if the page table is located in highmem.
329 * And let's rather not kmap-atomic the pte, just in case
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100330 * it's allocated already:
Harvey Harrison1156e092008-01-30 13:34:10 +0100331 */
Akinobu Mita087975b2009-06-27 15:35:15 +0900332 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
333 goto out;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100334
Akinobu Mita087975b2009-06-27 15:35:15 +0900335 pte = pte_offset_kernel(pmd, address);
336 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
337out:
Harvey Harrison1156e092008-01-30 13:34:10 +0100338 printk("\n");
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100339}
340
341#else /* CONFIG_X86_64: */
342
343void vmalloc_sync_all(void)
344{
Haicheng Li6afb5152010-05-19 17:42:14 +0800345 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100346}
347
348/*
349 * 64-bit:
350 *
351 * Handle a fault on the vmalloc area
352 *
353 * This assumes no large pages in there.
354 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400355static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100356{
357 pgd_t *pgd, *pgd_ref;
358 pud_t *pud, *pud_ref;
359 pmd_t *pmd, *pmd_ref;
360 pte_t *pte, *pte_ref;
361
362 /* Make sure we are in vmalloc area: */
363 if (!(address >= VMALLOC_START && address < VMALLOC_END))
364 return -1;
365
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200366 WARN_ON_ONCE(in_nmi());
367
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100368 /*
369 * Copy kernel mappings over when needed. This can also
370 * happen within a race in page table update. In the later
371 * case just flush:
372 */
373 pgd = pgd_offset(current->active_mm, address);
374 pgd_ref = pgd_offset_k(address);
375 if (pgd_none(*pgd_ref))
376 return -1;
377
378 if (pgd_none(*pgd))
379 set_pgd(pgd, *pgd_ref);
380 else
381 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
382
383 /*
384 * Below here mismatches are bugs because these lower tables
385 * are shared:
386 */
387
388 pud = pud_offset(pgd, address);
389 pud_ref = pud_offset(pgd_ref, address);
390 if (pud_none(*pud_ref))
391 return -1;
392
393 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
394 BUG();
395
396 pmd = pmd_offset(pud, address);
397 pmd_ref = pmd_offset(pud_ref, address);
398 if (pmd_none(*pmd_ref))
399 return -1;
400
401 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
402 BUG();
403
404 pte_ref = pte_offset_kernel(pmd_ref, address);
405 if (!pte_present(*pte_ref))
406 return -1;
407
408 pte = pte_offset_kernel(pmd, address);
409
410 /*
411 * Don't use pte_page here, because the mappings can point
412 * outside mem_map, and the NUMA hash lookup cannot handle
413 * that:
414 */
415 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
416 BUG();
417
418 return 0;
419}
420
421static const char errata93_warning[] =
Joe Perchesad361c92009-07-06 13:05:40 -0700422KERN_ERR
423"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
424"******* Working around it, but it may cause SEGVs or burn power.\n"
425"******* Please consider a BIOS update.\n"
426"******* Disabling USB legacy in the BIOS may also help.\n";
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100427
428/*
429 * No vm86 mode in 64-bit mode:
430 */
431static inline void
432check_v8086_mode(struct pt_regs *regs, unsigned long address,
433 struct task_struct *tsk)
434{
435}
436
437static int bad_address(void *p)
438{
439 unsigned long dummy;
440
441 return probe_kernel_address((unsigned long *)p, dummy);
442}
443
444static void dump_pagetable(unsigned long address)
445{
Akinobu Mita087975b2009-06-27 15:35:15 +0900446 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
447 pgd_t *pgd = base + pgd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 pud_t *pud;
449 pmd_t *pmd;
450 pte_t *pte;
451
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100452 if (bad_address(pgd))
453 goto bad;
454
Jan Beulichd646bce2006-02-03 21:51:47 +0100455 printk("PGD %lx ", pgd_val(*pgd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100456
457 if (!pgd_present(*pgd))
458 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Andi Kleend2ae5b52006-06-26 13:57:56 +0200460 pud = pud_offset(pgd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100461 if (bad_address(pud))
462 goto bad;
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 printk("PUD %lx ", pud_val(*pud));
Andi Kleenb5360222008-02-04 16:48:09 +0100465 if (!pud_present(*pud) || pud_large(*pud))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100466 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 pmd = pmd_offset(pud, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100469 if (bad_address(pmd))
470 goto bad;
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 printk("PMD %lx ", pmd_val(*pmd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100473 if (!pmd_present(*pmd) || pmd_large(*pmd))
474 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476 pte = pte_offset_kernel(pmd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100477 if (bad_address(pte))
478 goto bad;
479
Harvey Harrison33cb5242008-01-30 13:32:19 +0100480 printk("PTE %lx", pte_val(*pte));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100481out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 printk("\n");
483 return;
484bad:
485 printk("BAD\n");
486}
487
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100488#endif /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100490/*
491 * Workaround for K8 erratum #93 & buggy BIOS.
492 *
493 * BIOS SMM functions are required to use a specific workaround
494 * to avoid corruption of the 64bit RIP register on C stepping K8.
495 *
496 * A lot of BIOS that didn't get tested properly miss this.
497 *
498 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
499 * Try to work around it here.
500 *
501 * Note we only handle faults in kernel here.
502 * Does nothing on 32-bit.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100503 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100504static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100506#ifdef CONFIG_X86_64
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100507 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100509
Harvey Harrison33cb5242008-01-30 13:32:19 +0100510 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100514 if ((address >= (u64)_stext && address <= (u64)_etext) ||
515 (address >= MODULES_VADDR && address <= MODULES_END)) {
Ingo Molnara454ab32009-05-03 10:09:03 +0200516 printk_once(errata93_warning);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100517 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return 1;
519 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100520#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100522}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Harvey Harrison35f32662008-01-30 13:34:09 +0100524/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100525 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
526 * to illegal addresses >4GB.
527 *
528 * We catch this in the page fault handler because these addresses
529 * are not reachable. Just detect this case and return. Any code
Harvey Harrison35f32662008-01-30 13:34:09 +0100530 * segment in LDT is compatibility mode.
531 */
532static int is_errata100(struct pt_regs *regs, unsigned long address)
533{
534#ifdef CONFIG_X86_64
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100535 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
Harvey Harrison35f32662008-01-30 13:34:09 +0100536 return 1;
537#endif
538 return 0;
539}
540
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100541static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
542{
543#ifdef CONFIG_X86_F00F_BUG
544 unsigned long nr;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100545
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100546 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100547 * Pentium F0 0F C7 C8 bug workaround:
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100548 */
549 if (boot_cpu_data.f00f_bug) {
550 nr = (address - idt_descr.address) >> 3;
551
552 if (nr == 6) {
553 do_invalid_op(regs, 0);
554 return 1;
555 }
556 }
557#endif
558 return 0;
559}
560
Ingo Molnar8f766142009-02-20 23:00:29 +0100561static const char nx_warning[] = KERN_CRIT
562"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
563
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100564static void
565show_fault_oops(struct pt_regs *regs, unsigned long error_code,
566 unsigned long address)
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100567{
Harvey Harrison1156e092008-01-30 13:34:10 +0100568 if (!oops_may_print())
569 return;
570
Harvey Harrison1156e092008-01-30 13:34:10 +0100571 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100572 unsigned int level;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100573
Harvey Harrison1156e092008-01-30 13:34:10 +0100574 pte_t *pte = lookup_address(address, &level);
575
Ingo Molnar8f766142009-02-20 23:00:29 +0100576 if (pte && pte_present(*pte) && !pte_exec(*pte))
577 printk(nx_warning, current_uid());
Harvey Harrison1156e092008-01-30 13:34:10 +0100578 }
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100579
Harvey Harrison1156e092008-01-30 13:34:10 +0100580 printk(KERN_ALERT "BUG: unable to handle kernel ");
581 if (address < PAGE_SIZE)
582 printk(KERN_CONT "NULL pointer dereference");
583 else
584 printk(KERN_CONT "paging request");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100585
Vegard Nossumf294a8c2008-07-01 15:38:13 +0200586 printk(KERN_CONT " at %p\n", (void *) address);
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100587 printk(KERN_ALERT "IP:");
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100588 printk_address(regs->ip, 1);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100589
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100590 dump_pagetable(address);
591}
592
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100593static noinline void
594pgtable_bad(struct pt_regs *regs, unsigned long error_code,
595 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100597 struct task_struct *tsk;
598 unsigned long flags;
599 int sig;
600
601 flags = oops_begin();
602 tsk = current;
603 sig = SIGKILL;
Jan Beulich12091402005-09-12 18:49:24 +0200604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
Nick Piggin92181f12009-01-20 04:24:26 +0100606 tsk->comm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 dump_pagetable(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100608
609 tsk->thread.cr2 = address;
610 tsk->thread.trap_no = 14;
611 tsk->thread.error_code = error_code;
612
Jan Beulich22f59912008-01-30 13:31:23 +0100613 if (__die("Bad pagetable", regs, error_code))
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200614 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100615
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200616 oops_end(flags, regs, sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100619static noinline void
620no_context(struct pt_regs *regs, unsigned long error_code,
621 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100622{
623 struct task_struct *tsk = current;
Ingo Molnar19803072009-01-21 10:39:51 +0100624 unsigned long *stackend;
Nick Piggin92181f12009-01-20 04:24:26 +0100625 unsigned long flags;
626 int sig;
Nick Piggin92181f12009-01-20 04:24:26 +0100627
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100628 /* Are we prepared to handle this kernel fault? */
Nick Piggin92181f12009-01-20 04:24:26 +0100629 if (fixup_exception(regs))
630 return;
631
632 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100633 * 32-bit:
Nick Piggin92181f12009-01-20 04:24:26 +0100634 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100635 * Valid to do another page fault here, because if this fault
636 * had been triggered by is_prefetch fixup_exception would have
637 * handled it.
638 *
639 * 64-bit:
640 *
641 * Hall of shame of CPU/BIOS bugs.
Nick Piggin92181f12009-01-20 04:24:26 +0100642 */
643 if (is_prefetch(regs, error_code, address))
644 return;
645
646 if (is_errata93(regs, address))
647 return;
648
649 /*
650 * Oops. The kernel tried to access some bad page. We'll have to
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100651 * terminate things with extreme prejudice:
Nick Piggin92181f12009-01-20 04:24:26 +0100652 */
Nick Piggin92181f12009-01-20 04:24:26 +0100653 flags = oops_begin();
Nick Piggin92181f12009-01-20 04:24:26 +0100654
655 show_fault_oops(regs, error_code, address);
656
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100657 stackend = end_of_stack(tsk);
Jan Beulich0e7810b2009-11-20 14:00:14 +0000658 if (tsk != &init_task && *stackend != STACK_END_MAGIC)
Ingo Molnar19803072009-01-21 10:39:51 +0100659 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
660
Ingo Molnar1cc99542009-02-20 23:07:48 +0100661 tsk->thread.cr2 = address;
662 tsk->thread.trap_no = 14;
663 tsk->thread.error_code = error_code;
Nick Piggin92181f12009-01-20 04:24:26 +0100664
Nick Piggin92181f12009-01-20 04:24:26 +0100665 sig = SIGKILL;
666 if (__die("Oops", regs, error_code))
667 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100668
Nick Piggin92181f12009-01-20 04:24:26 +0100669 /* Executive summary in case the body of the oops scrolled away */
670 printk(KERN_EMERG "CR2: %016lx\n", address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100671
Nick Piggin92181f12009-01-20 04:24:26 +0100672 oops_end(flags, regs, sig);
Nick Piggin92181f12009-01-20 04:24:26 +0100673}
674
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100675/*
676 * Print out info about fatal segfaults, if the show_unhandled_signals
677 * sysctl is set:
678 */
679static inline void
680show_signal_msg(struct pt_regs *regs, unsigned long error_code,
681 unsigned long address, struct task_struct *tsk)
682{
683 if (!unhandled_signal(tsk, SIGSEGV))
684 return;
685
686 if (!printk_ratelimit())
687 return;
688
Roland Dreiera1a08d12009-07-11 00:10:04 -0700689 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100690 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
691 tsk->comm, task_pid_nr(tsk), address,
692 (void *)regs->ip, (void *)regs->sp, error_code);
693
694 print_vma_addr(KERN_CONT " in ", regs->ip);
695
696 printk(KERN_CONT "\n");
697}
698
699static void
700__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
701 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100702{
703 struct task_struct *tsk = current;
704
705 /* User mode accesses just cause a SIGSEGV */
706 if (error_code & PF_USER) {
707 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100708 * It's possible to have interrupts off here:
Nick Piggin92181f12009-01-20 04:24:26 +0100709 */
710 local_irq_enable();
711
712 /*
713 * Valid to do another page fault here because this one came
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100714 * from user space:
Nick Piggin92181f12009-01-20 04:24:26 +0100715 */
716 if (is_prefetch(regs, error_code, address))
717 return;
718
719 if (is_errata100(regs, address))
720 return;
721
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100722 if (unlikely(show_unhandled_signals))
723 show_signal_msg(regs, error_code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100724
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100725 /* Kernel addresses are always protection faults: */
726 tsk->thread.cr2 = address;
727 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
728 tsk->thread.trap_no = 14;
729
Andi Kleenf672b492010-09-27 22:05:55 +0200730 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100731
Nick Piggin92181f12009-01-20 04:24:26 +0100732 return;
733 }
734
735 if (is_f00f_bug(regs, address))
736 return;
737
738 no_context(regs, error_code, address);
739}
740
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100741static noinline void
742bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
743 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100744{
745 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
746}
747
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100748static void
749__bad_area(struct pt_regs *regs, unsigned long error_code,
750 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100751{
752 struct mm_struct *mm = current->mm;
753
754 /*
755 * Something tried to access memory that isn't in our memory map..
756 * Fix it, but check if it's kernel or user first..
757 */
758 up_read(&mm->mmap_sem);
759
760 __bad_area_nosemaphore(regs, error_code, address, si_code);
761}
762
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100763static noinline void
764bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100765{
766 __bad_area(regs, error_code, address, SEGV_MAPERR);
767}
768
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100769static noinline void
770bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
771 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100772{
773 __bad_area(regs, error_code, address, SEGV_ACCERR);
774}
775
776/* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100777static void
778out_of_memory(struct pt_regs *regs, unsigned long error_code,
779 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100780{
781 /*
782 * We ran out of memory, call the OOM killer, and return the userspace
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100783 * (which will retry the fault, or kill us if we got oom-killed):
Nick Piggin92181f12009-01-20 04:24:26 +0100784 */
785 up_read(&current->mm->mmap_sem);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100786
Nick Piggin92181f12009-01-20 04:24:26 +0100787 pagefault_out_of_memory();
788}
789
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100790static void
Andi Kleena6e04aa2009-09-16 11:50:09 +0200791do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
792 unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100793{
794 struct task_struct *tsk = current;
795 struct mm_struct *mm = tsk->mm;
Andi Kleena6e04aa2009-09-16 11:50:09 +0200796 int code = BUS_ADRERR;
Nick Piggin92181f12009-01-20 04:24:26 +0100797
798 up_read(&mm->mmap_sem);
799
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100800 /* Kernel mode? Handle exceptions or die: */
Linus Torvalds96054562010-08-13 09:49:20 -0700801 if (!(error_code & PF_USER)) {
Nick Piggin92181f12009-01-20 04:24:26 +0100802 no_context(regs, error_code, address);
Linus Torvalds96054562010-08-13 09:49:20 -0700803 return;
804 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100805
Ingo Molnarcd1b68f2009-02-20 23:39:02 +0100806 /* User-space => ok to do another page fault: */
Nick Piggin92181f12009-01-20 04:24:26 +0100807 if (is_prefetch(regs, error_code, address))
808 return;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100809
810 tsk->thread.cr2 = address;
811 tsk->thread.error_code = error_code;
812 tsk->thread.trap_no = 14;
813
Andi Kleena6e04aa2009-09-16 11:50:09 +0200814#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenf672b492010-09-27 22:05:55 +0200815 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
Andi Kleena6e04aa2009-09-16 11:50:09 +0200816 printk(KERN_ERR
817 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
818 tsk->comm, tsk->pid, address);
819 code = BUS_MCEERR_AR;
820 }
821#endif
Andi Kleenf672b492010-09-27 22:05:55 +0200822 force_sig_info_fault(SIGBUS, code, address, tsk, fault);
Nick Piggin92181f12009-01-20 04:24:26 +0100823}
824
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100825static noinline void
826mm_fault_error(struct pt_regs *regs, unsigned long error_code,
827 unsigned long address, unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100828{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100829 if (fault & VM_FAULT_OOM) {
Andrey Vaginf8626852011-03-09 15:22:23 -0800830 /* Kernel mode? Handle exceptions or die: */
831 if (!(error_code & PF_USER)) {
832 up_read(&current->mm->mmap_sem);
833 no_context(regs, error_code, address);
834 return;
835 }
836
Nick Piggin92181f12009-01-20 04:24:26 +0100837 out_of_memory(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100838 } else {
Andi Kleenf672b492010-09-27 22:05:55 +0200839 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
840 VM_FAULT_HWPOISON_LARGE))
Andi Kleena6e04aa2009-09-16 11:50:09 +0200841 do_sigbus(regs, error_code, address, fault);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100842 else
843 BUG();
844 }
Nick Piggin92181f12009-01-20 04:24:26 +0100845}
846
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100847static int spurious_fault_check(unsigned long error_code, pte_t *pte)
848{
849 if ((error_code & PF_WRITE) && !pte_write(*pte))
850 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100851
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100852 if ((error_code & PF_INSTR) && !pte_exec(*pte))
853 return 0;
854
855 return 1;
856}
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100859 * Handle a spurious fault caused by a stale TLB entry.
860 *
861 * This allows us to lazily refresh the TLB when increasing the
862 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
863 * eagerly is very expensive since that implies doing a full
864 * cross-processor TLB flush, even if no stale TLB entries exist
865 * on other processors.
866 *
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100867 * There are no security implications to leaving a stale TLB when
868 * increasing the permissions on a page.
869 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400870static noinline __kprobes int
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100871spurious_fault(unsigned long error_code, unsigned long address)
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100872{
873 pgd_t *pgd;
874 pud_t *pud;
875 pmd_t *pmd;
876 pte_t *pte;
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500877 int ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100878
879 /* Reserved-bit violation or user access to kernel space? */
880 if (error_code & (PF_USER | PF_RSVD))
881 return 0;
882
883 pgd = init_mm.pgd + pgd_index(address);
884 if (!pgd_present(*pgd))
885 return 0;
886
887 pud = pud_offset(pgd, address);
888 if (!pud_present(*pud))
889 return 0;
890
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100891 if (pud_large(*pud))
892 return spurious_fault_check(error_code, (pte_t *) pud);
893
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100894 pmd = pmd_offset(pud, address);
895 if (!pmd_present(*pmd))
896 return 0;
897
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100898 if (pmd_large(*pmd))
899 return spurious_fault_check(error_code, (pte_t *) pmd);
900
Shaohua Li660a2932010-07-27 16:06:28 +0800901 /*
902 * Note: don't use pte_present() here, since it returns true
903 * if the _PAGE_PROTNONE bit is set. However, this aliases the
904 * _PAGE_GLOBAL bit, which for kernel pages give false positives
905 * when CONFIG_DEBUG_PAGEALLOC is used.
906 */
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100907 pte = pte_offset_kernel(pmd, address);
Shaohua Li660a2932010-07-27 16:06:28 +0800908 if (!(pte_flags(*pte) & _PAGE_PRESENT))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100909 return 0;
910
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500911 ret = spurious_fault_check(error_code, pte);
912 if (!ret)
913 return 0;
914
915 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100916 * Make sure we have permissions in PMD.
917 * If not, then there's a bug in the page tables:
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500918 */
919 ret = spurious_fault_check(error_code, (pte_t *) pmd);
920 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100921
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500922 return ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100923}
924
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200925int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100927static inline int
Michel Lespinasse68da3362010-10-26 14:21:58 -0700928access_error(unsigned long error_code, struct vm_area_struct *vma)
Nick Piggin92181f12009-01-20 04:24:26 +0100929{
Michel Lespinasse68da3362010-10-26 14:21:58 -0700930 if (error_code & PF_WRITE) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100931 /* write, present and write, not present: */
Nick Piggin92181f12009-01-20 04:24:26 +0100932 if (unlikely(!(vma->vm_flags & VM_WRITE)))
933 return 1;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100934 return 0;
Nick Piggin92181f12009-01-20 04:24:26 +0100935 }
936
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100937 /* read, present: */
938 if (unlikely(error_code & PF_PROT))
939 return 1;
940
941 /* read, not present: */
942 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
943 return 1;
944
Nick Piggin92181f12009-01-20 04:24:26 +0100945 return 0;
946}
947
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800948static int fault_in_kernel_space(unsigned long address)
949{
Ingo Molnard9517342009-02-20 23:32:28 +0100950 return address >= TASK_SIZE_MAX;
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800951}
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953/*
954 * This routine handles page faults. It determines the address,
955 * and the problem, and then passes it off to one of the appropriate
956 * routines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 */
Ingo Molnarc3731c62009-02-20 23:22:34 +0100958dotraplinkage void __kprobes
959do_page_fault(struct pt_regs *regs, unsigned long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Harvey Harrison33cb5242008-01-30 13:32:19 +0100961 struct vm_area_struct *vma;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100962 struct task_struct *tsk;
963 unsigned long address;
964 struct mm_struct *mm;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100965 int fault;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700966 int write = error_code & PF_WRITE;
967 unsigned int flags = FAULT_FLAG_ALLOW_RETRY |
968 (write ? FAULT_FLAG_WRITE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100970 tsk = current;
971 mm = tsk->mm;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100972
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100973 /* Get the faulting address: */
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200974 address = read_cr2();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Vegard Nossumf8561292008-04-04 00:53:23 +0200976 /*
977 * Detect and handle instructions that would cause a page fault for
978 * both a tracked kernel page and a userspace page.
979 */
980 if (kmemcheck_active(regs))
981 kmemcheck_hide(regs);
Ingo Molnar5dfaf902009-06-16 10:23:32 +0200982 prefetchw(&mm->mmap_sem);
Vegard Nossumf8561292008-04-04 00:53:23 +0200983
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +0200984 if (unlikely(kmmio_fault(regs, address)))
Pekka Paalanen86069782008-05-12 21:20:56 +0200985 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
987 /*
988 * We fault-in kernel-space virtual memory on-demand. The
989 * 'reference' page table is init_mm.pgd.
990 *
991 * NOTE! We MUST NOT take any locks for this case. We may
992 * be in an interrupt or a critical region, and should
993 * only copy the information from the master page table,
994 * nothing more.
995 *
996 * This verifies that the fault happens in kernel space
997 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +0100998 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 */
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -08001000 if (unlikely(fault_in_kernel_space(address))) {
Vegard Nossumf8561292008-04-04 00:53:23 +02001001 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1002 if (vmalloc_fault(address) >= 0)
1003 return;
1004
1005 if (kmemcheck_fault(regs, address, error_code))
1006 return;
1007 }
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001008
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001009 /* Can handle a stale RO->RW TLB: */
Nick Piggin92181f12009-01-20 04:24:26 +01001010 if (spurious_fault(error_code, address))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001011 return;
1012
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001013 /* kprobes don't want to hook the spurious faults: */
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001014 if (notify_page_fault(regs))
1015 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001016 /*
1017 * Don't take the mm semaphore here. If we fixup a prefetch
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001018 * fault we could otherwise deadlock:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001019 */
Nick Piggin92181f12009-01-20 04:24:26 +01001020 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001021
Nick Piggin92181f12009-01-20 04:24:26 +01001022 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001023 }
1024
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001025 /* kprobes don't want to hook the spurious faults: */
Ingo Molnarf8a6b2b2009-02-13 09:44:22 +01001026 if (unlikely(notify_page_fault(regs)))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001027 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001028 /*
Linus Torvalds891cffb2008-10-12 13:16:12 -07001029 * It's safe to allow irq's after cr2 has been saved and the
1030 * vmalloc fault has been handled.
1031 *
1032 * User-mode registers count as a user access even for any
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001033 * potential system fault or CPU buglet:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001034 */
Linus Torvalds891cffb2008-10-12 13:16:12 -07001035 if (user_mode_vm(regs)) {
1036 local_irq_enable();
1037 error_code |= PF_USER;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001038 } else {
1039 if (regs->flags & X86_EFLAGS_IF)
1040 local_irq_enable();
1041 }
Jan Beulich8c914cb2006-03-25 16:29:40 +01001042
Andi Kleen66c58152006-01-11 22:44:09 +01001043 if (unlikely(error_code & PF_RSVD))
Nick Piggin92181f12009-01-20 04:24:26 +01001044 pgtable_bad(regs, error_code, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001046 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
Peter Zijlstra7dd1fcc2009-03-13 12:21:33 +01001047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001049 * If we're in an interrupt, have no user context or are running
1050 * in an atomic region then we must not take the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
Nick Piggin92181f12009-01-20 04:24:26 +01001052 if (unlikely(in_atomic() || !mm)) {
1053 bad_area_nosemaphore(regs, error_code, address);
1054 return;
1055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Ingo Molnar3a1dfe62008-10-13 17:49:02 +02001057 /*
1058 * When running in the kernel we expect faults to occur only to
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001059 * addresses in user space. All other faults represent errors in
1060 * the kernel and should generate an OOPS. Unfortunately, in the
1061 * case of an erroneous fault occurring in a code path which already
1062 * holds mmap_sem we will deadlock attempting to validate the fault
1063 * against the address space. Luckily the kernel only validly
1064 * references user space from well defined areas of code, which are
1065 * listed in the exceptions table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 *
1067 * As the vast majority of faults will be valid we will only perform
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001068 * the source reference check when there is a possibility of a
1069 * deadlock. Attempt to lock the address space, if we cannot we then
1070 * validate the source. If this is invalid we can skip the address
1071 * space check, thus avoiding the deadlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 */
Nick Piggin92181f12009-01-20 04:24:26 +01001073 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
Andi Kleen66c58152006-01-11 22:44:09 +01001074 if ((error_code & PF_USER) == 0 &&
Nick Piggin92181f12009-01-20 04:24:26 +01001075 !search_exception_tables(regs->ip)) {
1076 bad_area_nosemaphore(regs, error_code, address);
1077 return;
1078 }
Michel Lespinassed065bd82010-10-26 14:21:57 -07001079retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 down_read(&mm->mmap_sem);
Peter Zijlstra01006072009-01-29 16:02:12 +01001081 } else {
1082 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001083 * The above down_read_trylock() might have succeeded in
1084 * which case we'll have missed the might_sleep() from
1085 * down_read():
Peter Zijlstra01006072009-01-29 16:02:12 +01001086 */
1087 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 }
1089
1090 vma = find_vma(mm, address);
Nick Piggin92181f12009-01-20 04:24:26 +01001091 if (unlikely(!vma)) {
1092 bad_area(regs, error_code, address);
1093 return;
1094 }
1095 if (likely(vma->vm_start <= address))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 goto good_area;
Nick Piggin92181f12009-01-20 04:24:26 +01001097 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1098 bad_area(regs, error_code, address);
1099 return;
1100 }
Harvey Harrison33cb5242008-01-30 13:32:19 +01001101 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001102 /*
1103 * Accessing the stack below %sp is always a bug.
1104 * The large cushion allows instructions like enter
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001105 * and pusha to work. ("enter $65535, $31" pushes
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001106 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +02001107 */
Nick Piggin92181f12009-01-20 04:24:26 +01001108 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1109 bad_area(regs, error_code, address);
1110 return;
1111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
Nick Piggin92181f12009-01-20 04:24:26 +01001113 if (unlikely(expand_stack(vma, address))) {
1114 bad_area(regs, error_code, address);
1115 return;
1116 }
1117
1118 /*
1119 * Ok, we have a good vm_area for this memory access, so
1120 * we can handle it..
1121 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122good_area:
Michel Lespinasse68da3362010-10-26 14:21:58 -07001123 if (unlikely(access_error(error_code, vma))) {
Nick Piggin92181f12009-01-20 04:24:26 +01001124 bad_area_access_error(regs, error_code, address);
1125 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
1127
1128 /*
1129 * If for any reason at all we couldn't handle the fault,
1130 * make sure we exit gracefully rather than endlessly redo
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001131 * the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001133 fault = handle_mm_fault(mm, vma, address, flags);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001134
Nick Piggin83c54072007-07-19 01:47:05 -07001135 if (unlikely(fault & VM_FAULT_ERROR)) {
Nick Piggin92181f12009-01-20 04:24:26 +01001136 mm_fault_error(regs, error_code, address, fault);
1137 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001139
Michel Lespinassed065bd82010-10-26 14:21:57 -07001140 /*
1141 * Major/minor page fault accounting is only done on the
1142 * initial attempt. If we go through a retry, it is extremely
1143 * likely that the page will be found in page cache at that point.
1144 */
1145 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1146 if (fault & VM_FAULT_MAJOR) {
1147 tsk->maj_flt++;
1148 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
1149 regs, address);
1150 } else {
1151 tsk->min_flt++;
1152 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
1153 regs, address);
1154 }
1155 if (fault & VM_FAULT_RETRY) {
1156 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
1157 * of starvation. */
1158 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1159 goto retry;
1160 }
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001161 }
Harvey Harrisond729ab32008-01-30 13:33:23 +01001162
Ingo Molnar8c938f92009-02-20 22:12:18 +01001163 check_v8086_mode(regs, address, tsk);
1164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 up_read(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166}