blob: bcb394dfbb3587f8b4d40cc096b6ff7334ce6b12 [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 Torvalds268bb0c2011-05-20 12:50:29 -070015#include <linux/prefetch.h> /* prefetchw */
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
Ingo Molnara2bcd472009-03-29 23:47:48 +020017#include <asm/traps.h> /* dotraplinkage, ... */
18#include <asm/pgalloc.h> /* pgd_*(), ... */
Vegard Nossumf8561292008-04-04 00:53:23 +020019#include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Harvey Harrison33cb5242008-01-30 13:32:19 +010021/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010022 * Page fault error code bits:
23 *
24 * bit 0 == 0: no page found 1: protection fault
25 * bit 1 == 0: read access 1: write access
26 * bit 2 == 0: kernel-mode access 1: user-mode access
27 * bit 3 == 1: use of reserved bit detected
28 * bit 4 == 1: fault was an instruction fetch
Harvey Harrison33cb5242008-01-30 13:32:19 +010029 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +010030enum x86_pf_error_code {
31
32 PF_PROT = 1 << 0,
33 PF_WRITE = 1 << 1,
34 PF_USER = 1 << 2,
35 PF_RSVD = 1 << 3,
36 PF_INSTR = 1 << 4,
37};
Andi Kleen66c58152006-01-11 22:44:09 +010038
Ingo Molnarb814d412009-02-20 22:32:10 +010039/*
Ingo Molnarb319eed2009-02-22 10:24:18 +010040 * Returns 0 if mmiotrace is disabled, or if the fault is not
41 * handled by mmiotrace:
Ingo Molnarb814d412009-02-20 22:32:10 +010042 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -040043static inline int __kprobes
44kmmio_fault(struct pt_regs *regs, unsigned long addr)
Pekka Paalanen86069782008-05-12 21:20:56 +020045{
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020046 if (unlikely(is_kmmio_active()))
47 if (kmmio_handler(regs, addr) == 1)
48 return -1;
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020049 return 0;
Pekka Paalanen86069782008-05-12 21:20:56 +020050}
51
Masami Hiramatsu62c92952009-08-27 13:23:11 -040052static inline int __kprobes notify_page_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070053{
Christoph Hellwig74a0b572007-10-16 01:24:07 -070054 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070055
Christoph Hellwig74a0b572007-10-16 01:24:07 -070056 /* kprobe_running() needs smp_processor_id() */
Ingo Molnarb1801812009-02-20 22:42:57 +010057 if (kprobes_built_in() && !user_mode_vm(regs)) {
Christoph Hellwig74a0b572007-10-16 01:24:07 -070058 preempt_disable();
59 if (kprobe_running() && kprobe_fault_handler(regs, 14))
60 ret = 1;
61 preempt_enable();
62 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070063
Christoph Hellwig74a0b572007-10-16 01:24:07 -070064 return ret;
Harvey Harrison33cb5242008-01-30 13:32:19 +010065}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070066
Harvey Harrison1dc85be2008-01-30 13:32:35 +010067/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010068 * Prefetch quirks:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010069 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010070 * 32-bit mode:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010071 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010072 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
73 * Check that here and ignore it.
74 *
75 * 64-bit mode:
76 *
77 * Sometimes the CPU reports invalid exceptions on prefetch.
78 * Check that here and ignore it.
79 *
80 * Opcode checker based on code by Richard Brunner.
Harvey Harrison1dc85be2008-01-30 13:32:35 +010081 */
Ingo Molnar107a0362009-02-20 20:37:05 +010082static inline int
83check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
84 unsigned char opcode, int *prefetch)
85{
86 unsigned char instr_hi = opcode & 0xf0;
87 unsigned char instr_lo = opcode & 0x0f;
88
89 switch (instr_hi) {
90 case 0x20:
91 case 0x30:
92 /*
93 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
94 * In X86_64 long mode, the CPU will signal invalid
95 * opcode if some of these prefixes are present so
96 * X86_64 will never get here anyway
97 */
98 return ((instr_lo & 7) == 0x6);
99#ifdef CONFIG_X86_64
100 case 0x40:
101 /*
102 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
103 * Need to figure out under what instruction mode the
104 * instruction was issued. Could check the LDT for lm,
105 * but for now it's good enough to assume that long
106 * mode only uses well known segments or kernel.
107 */
108 return (!user_mode(regs)) || (regs->cs == __USER_CS);
109#endif
110 case 0x60:
111 /* 0x64 thru 0x67 are valid prefixes in all modes. */
112 return (instr_lo & 0xC) == 0x4;
113 case 0xF0:
114 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
115 return !instr_lo || (instr_lo>>1) == 1;
116 case 0x00:
117 /* Prefetch instruction is 0x0F0D or 0x0F18 */
118 if (probe_kernel_address(instr, opcode))
119 return 0;
120
121 *prefetch = (instr_lo == 0xF) &&
122 (opcode == 0x0D || opcode == 0x18);
123 return 0;
124 default:
125 return 0;
126 }
127}
128
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100129static int
130is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
Harvey Harrison33cb5242008-01-30 13:32:19 +0100131{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100132 unsigned char *max_instr;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100133 unsigned char *instr;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100134 int prefetch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
Ingo Molnar30853542008-03-27 21:29:09 +0100136 /*
137 * If it was a exec (instruction fetch) fault on NX page, then
138 * do not ignore the fault:
139 */
Andi Kleen66c58152006-01-11 22:44:09 +0100140 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100142
Ingo Molnar107a0362009-02-20 20:37:05 +0100143 instr = (void *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700144 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Vincent Hanquez76381fe2005-06-23 00:08:46 -0700146 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return 0;
148
Ingo Molnar107a0362009-02-20 20:37:05 +0100149 while (instr < max_instr) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100150 unsigned char opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100152 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100153 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 instr++;
156
Ingo Molnar107a0362009-02-20 20:37:05 +0100157 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
160 return prefetch;
161}
162
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100163static void
164force_sig_info_fault(int si_signo, int si_code, unsigned long address,
Andi Kleenf672b492010-09-27 22:05:55 +0200165 struct task_struct *tsk, int fault)
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100166{
Andi Kleenf672b492010-09-27 22:05:55 +0200167 unsigned lsb = 0;
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100168 siginfo_t info;
169
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100170 info.si_signo = si_signo;
171 info.si_errno = 0;
172 info.si_code = si_code;
173 info.si_addr = (void __user *)address;
Andi Kleenf672b492010-09-27 22:05:55 +0200174 if (fault & VM_FAULT_HWPOISON_LARGE)
175 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
176 if (fault & VM_FAULT_HWPOISON)
177 lsb = PAGE_SHIFT;
178 info.si_addr_lsb = lsb;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100179
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100180 force_sig_info(si_signo, &info, tsk);
181}
182
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100183DEFINE_SPINLOCK(pgd_lock);
184LIST_HEAD(pgd_list);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100185
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100186#ifdef CONFIG_X86_32
187static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
188{
189 unsigned index = pgd_index(address);
190 pgd_t *pgd_k;
191 pud_t *pud, *pud_k;
192 pmd_t *pmd, *pmd_k;
193
194 pgd += index;
195 pgd_k = init_mm.pgd + index;
196
197 if (!pgd_present(*pgd_k))
198 return NULL;
199
200 /*
201 * set_pgd(pgd, *pgd_k); here would be useless on PAE
202 * and redundant with the set_pmd() on non-PAE. As would
203 * set_pud.
204 */
205 pud = pud_offset(pgd, address);
206 pud_k = pud_offset(pgd_k, address);
207 if (!pud_present(*pud_k))
208 return NULL;
209
210 pmd = pmd_offset(pud, address);
211 pmd_k = pmd_offset(pud_k, address);
212 if (!pmd_present(*pmd_k))
213 return NULL;
214
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800215 if (!pmd_present(*pmd))
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100216 set_pmd(pmd, *pmd_k);
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800217 else
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100218 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100219
220 return pmd_k;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100221}
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100222
223void vmalloc_sync_all(void)
224{
225 unsigned long address;
226
227 if (SHARED_KERNEL_PMD)
228 return;
229
230 for (address = VMALLOC_START & PMD_MASK;
231 address >= TASK_SIZE && address < FIXADDR_TOP;
232 address += PMD_SIZE) {
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100233 struct page *page;
234
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800235 spin_lock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100236 list_for_each_entry(page, &pgd_list, lru) {
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700237 spinlock_t *pgt_lock;
Borislav Petkovf01f7c52010-10-19 22:17:37 +0000238 pmd_t *ret;
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700239
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800240 /* the pgt_lock only for Xen */
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700241 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
242
243 spin_lock(pgt_lock);
244 ret = vmalloc_sync_one(page_address(page), address);
245 spin_unlock(pgt_lock);
246
247 if (!ret)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100248 break;
249 }
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800250 spin_unlock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100251 }
252}
253
254/*
255 * 32-bit:
256 *
257 * Handle a fault on the vmalloc or module mapping area
258 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400259static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100260{
261 unsigned long pgd_paddr;
262 pmd_t *pmd_k;
263 pte_t *pte_k;
264
265 /* Make sure we are in vmalloc area: */
266 if (!(address >= VMALLOC_START && address < VMALLOC_END))
267 return -1;
268
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200269 WARN_ON_ONCE(in_nmi());
270
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100271 /*
272 * Synchronize this task's top level page-table
273 * with the 'reference' page table.
274 *
275 * Do _not_ use "current" here. We might be inside
276 * an interrupt in the middle of a task switch..
277 */
278 pgd_paddr = read_cr3();
279 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
280 if (!pmd_k)
281 return -1;
282
283 pte_k = pte_offset_kernel(pmd_k, address);
284 if (!pte_present(*pte_k))
285 return -1;
286
287 return 0;
288}
289
290/*
291 * Did it hit the DOS screen memory VA from vm86 mode?
292 */
293static inline void
294check_v8086_mode(struct pt_regs *regs, unsigned long address,
295 struct task_struct *tsk)
296{
297 unsigned long bit;
298
299 if (!v8086_mode(regs))
300 return;
301
302 bit = (address - 0xA0000) >> PAGE_SHIFT;
303 if (bit < 32)
304 tsk->thread.screen_bitmap |= 1 << bit;
305}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Akinobu Mita087975b2009-06-27 15:35:15 +0900307static bool low_pfn(unsigned long pfn)
308{
309 return pfn < max_low_pfn;
310}
311
Adrian Bunkcae30f822008-02-13 23:31:31 +0200312static void dump_pagetable(unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Akinobu Mita087975b2009-06-27 15:35:15 +0900314 pgd_t *base = __va(read_cr3());
315 pgd_t *pgd = &base[pgd_index(address)];
316 pmd_t *pmd;
317 pte_t *pte;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100318
Harvey Harrison1156e092008-01-30 13:34:10 +0100319#ifdef CONFIG_X86_PAE
Akinobu Mita087975b2009-06-27 15:35:15 +0900320 printk("*pdpt = %016Lx ", pgd_val(*pgd));
321 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
322 goto out;
Harvey Harrison1156e092008-01-30 13:34:10 +0100323#endif
Akinobu Mita087975b2009-06-27 15:35:15 +0900324 pmd = pmd_offset(pud_offset(pgd, address), address);
325 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
Harvey Harrison1156e092008-01-30 13:34:10 +0100326
327 /*
328 * We must not directly access the pte in the highpte
329 * case if the page table is located in highmem.
330 * And let's rather not kmap-atomic the pte, just in case
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100331 * it's allocated already:
Harvey Harrison1156e092008-01-30 13:34:10 +0100332 */
Akinobu Mita087975b2009-06-27 15:35:15 +0900333 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
334 goto out;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100335
Akinobu Mita087975b2009-06-27 15:35:15 +0900336 pte = pte_offset_kernel(pmd, address);
337 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
338out:
Harvey Harrison1156e092008-01-30 13:34:10 +0100339 printk("\n");
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100340}
341
342#else /* CONFIG_X86_64: */
343
344void vmalloc_sync_all(void)
345{
Haicheng Li6afb5152010-05-19 17:42:14 +0800346 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100347}
348
349/*
350 * 64-bit:
351 *
352 * Handle a fault on the vmalloc area
353 *
354 * This assumes no large pages in there.
355 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400356static noinline __kprobes int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100357{
358 pgd_t *pgd, *pgd_ref;
359 pud_t *pud, *pud_ref;
360 pmd_t *pmd, *pmd_ref;
361 pte_t *pte, *pte_ref;
362
363 /* Make sure we are in vmalloc area: */
364 if (!(address >= VMALLOC_START && address < VMALLOC_END))
365 return -1;
366
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200367 WARN_ON_ONCE(in_nmi());
368
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100369 /*
370 * Copy kernel mappings over when needed. This can also
371 * happen within a race in page table update. In the later
372 * case just flush:
373 */
374 pgd = pgd_offset(current->active_mm, address);
375 pgd_ref = pgd_offset_k(address);
376 if (pgd_none(*pgd_ref))
377 return -1;
378
379 if (pgd_none(*pgd))
380 set_pgd(pgd, *pgd_ref);
381 else
382 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
383
384 /*
385 * Below here mismatches are bugs because these lower tables
386 * are shared:
387 */
388
389 pud = pud_offset(pgd, address);
390 pud_ref = pud_offset(pgd_ref, address);
391 if (pud_none(*pud_ref))
392 return -1;
393
394 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
395 BUG();
396
397 pmd = pmd_offset(pud, address);
398 pmd_ref = pmd_offset(pud_ref, address);
399 if (pmd_none(*pmd_ref))
400 return -1;
401
402 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
403 BUG();
404
405 pte_ref = pte_offset_kernel(pmd_ref, address);
406 if (!pte_present(*pte_ref))
407 return -1;
408
409 pte = pte_offset_kernel(pmd, address);
410
411 /*
412 * Don't use pte_page here, because the mappings can point
413 * outside mem_map, and the NUMA hash lookup cannot handle
414 * that:
415 */
416 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
417 BUG();
418
419 return 0;
420}
421
422static const char errata93_warning[] =
Joe Perchesad361c92009-07-06 13:05:40 -0700423KERN_ERR
424"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
425"******* Working around it, but it may cause SEGVs or burn power.\n"
426"******* Please consider a BIOS update.\n"
427"******* Disabling USB legacy in the BIOS may also help.\n";
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100428
429/*
430 * No vm86 mode in 64-bit mode:
431 */
432static inline void
433check_v8086_mode(struct pt_regs *regs, unsigned long address,
434 struct task_struct *tsk)
435{
436}
437
438static int bad_address(void *p)
439{
440 unsigned long dummy;
441
442 return probe_kernel_address((unsigned long *)p, dummy);
443}
444
445static void dump_pagetable(unsigned long address)
446{
Akinobu Mita087975b2009-06-27 15:35:15 +0900447 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
448 pgd_t *pgd = base + pgd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 pud_t *pud;
450 pmd_t *pmd;
451 pte_t *pte;
452
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100453 if (bad_address(pgd))
454 goto bad;
455
Jan Beulichd646bce2006-02-03 21:51:47 +0100456 printk("PGD %lx ", pgd_val(*pgd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100457
458 if (!pgd_present(*pgd))
459 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Andi Kleend2ae5b52006-06-26 13:57:56 +0200461 pud = pud_offset(pgd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100462 if (bad_address(pud))
463 goto bad;
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 printk("PUD %lx ", pud_val(*pud));
Andi Kleenb5360222008-02-04 16:48:09 +0100466 if (!pud_present(*pud) || pud_large(*pud))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100467 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 pmd = pmd_offset(pud, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100470 if (bad_address(pmd))
471 goto bad;
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 printk("PMD %lx ", pmd_val(*pmd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100474 if (!pmd_present(*pmd) || pmd_large(*pmd))
475 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477 pte = pte_offset_kernel(pmd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100478 if (bad_address(pte))
479 goto bad;
480
Harvey Harrison33cb5242008-01-30 13:32:19 +0100481 printk("PTE %lx", pte_val(*pte));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100482out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 printk("\n");
484 return;
485bad:
486 printk("BAD\n");
487}
488
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100489#endif /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100491/*
492 * Workaround for K8 erratum #93 & buggy BIOS.
493 *
494 * BIOS SMM functions are required to use a specific workaround
495 * to avoid corruption of the 64bit RIP register on C stepping K8.
496 *
497 * A lot of BIOS that didn't get tested properly miss this.
498 *
499 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
500 * Try to work around it here.
501 *
502 * Note we only handle faults in kernel here.
503 * Does nothing on 32-bit.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100504 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100505static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100507#ifdef CONFIG_X86_64
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100508 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100510
Harvey Harrison33cb5242008-01-30 13:32:19 +0100511 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100515 if ((address >= (u64)_stext && address <= (u64)_etext) ||
516 (address >= MODULES_VADDR && address <= MODULES_END)) {
Ingo Molnara454ab32009-05-03 10:09:03 +0200517 printk_once(errata93_warning);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100518 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return 1;
520 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100521#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100523}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Harvey Harrison35f32662008-01-30 13:34:09 +0100525/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100526 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
527 * to illegal addresses >4GB.
528 *
529 * We catch this in the page fault handler because these addresses
530 * are not reachable. Just detect this case and return. Any code
Harvey Harrison35f32662008-01-30 13:34:09 +0100531 * segment in LDT is compatibility mode.
532 */
533static int is_errata100(struct pt_regs *regs, unsigned long address)
534{
535#ifdef CONFIG_X86_64
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100536 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
Harvey Harrison35f32662008-01-30 13:34:09 +0100537 return 1;
538#endif
539 return 0;
540}
541
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100542static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
543{
544#ifdef CONFIG_X86_F00F_BUG
545 unsigned long nr;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100546
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100547 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100548 * Pentium F0 0F C7 C8 bug workaround:
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100549 */
550 if (boot_cpu_data.f00f_bug) {
551 nr = (address - idt_descr.address) >> 3;
552
553 if (nr == 6) {
554 do_invalid_op(regs, 0);
555 return 1;
556 }
557 }
558#endif
559 return 0;
560}
561
Ingo Molnar8f766142009-02-20 23:00:29 +0100562static const char nx_warning[] = KERN_CRIT
563"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
564
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100565static void
566show_fault_oops(struct pt_regs *regs, unsigned long error_code,
567 unsigned long address)
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100568{
Harvey Harrison1156e092008-01-30 13:34:10 +0100569 if (!oops_may_print())
570 return;
571
Harvey Harrison1156e092008-01-30 13:34:10 +0100572 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100573 unsigned int level;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100574
Harvey Harrison1156e092008-01-30 13:34:10 +0100575 pte_t *pte = lookup_address(address, &level);
576
Ingo Molnar8f766142009-02-20 23:00:29 +0100577 if (pte && pte_present(*pte) && !pte_exec(*pte))
578 printk(nx_warning, current_uid());
Harvey Harrison1156e092008-01-30 13:34:10 +0100579 }
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100580
Harvey Harrison1156e092008-01-30 13:34:10 +0100581 printk(KERN_ALERT "BUG: unable to handle kernel ");
582 if (address < PAGE_SIZE)
583 printk(KERN_CONT "NULL pointer dereference");
584 else
585 printk(KERN_CONT "paging request");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100586
Vegard Nossumf294a8c2008-07-01 15:38:13 +0200587 printk(KERN_CONT " at %p\n", (void *) address);
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100588 printk(KERN_ALERT "IP:");
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100589 printk_address(regs->ip, 1);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100590
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100591 dump_pagetable(address);
592}
593
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100594static noinline void
595pgtable_bad(struct pt_regs *regs, unsigned long error_code,
596 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100598 struct task_struct *tsk;
599 unsigned long flags;
600 int sig;
601
602 flags = oops_begin();
603 tsk = current;
604 sig = SIGKILL;
Jan Beulich12091402005-09-12 18:49:24 +0200605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
Nick Piggin92181f12009-01-20 04:24:26 +0100607 tsk->comm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 dump_pagetable(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100609
610 tsk->thread.cr2 = address;
611 tsk->thread.trap_no = 14;
612 tsk->thread.error_code = error_code;
613
Jan Beulich22f59912008-01-30 13:31:23 +0100614 if (__die("Bad pagetable", regs, error_code))
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200615 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100616
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200617 oops_end(flags, regs, sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100620static noinline void
621no_context(struct pt_regs *regs, unsigned long error_code,
622 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100623{
624 struct task_struct *tsk = current;
Ingo Molnar19803072009-01-21 10:39:51 +0100625 unsigned long *stackend;
Nick Piggin92181f12009-01-20 04:24:26 +0100626 unsigned long flags;
627 int sig;
Nick Piggin92181f12009-01-20 04:24:26 +0100628
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100629 /* Are we prepared to handle this kernel fault? */
Nick Piggin92181f12009-01-20 04:24:26 +0100630 if (fixup_exception(regs))
631 return;
632
633 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100634 * 32-bit:
Nick Piggin92181f12009-01-20 04:24:26 +0100635 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100636 * Valid to do another page fault here, because if this fault
637 * had been triggered by is_prefetch fixup_exception would have
638 * handled it.
639 *
640 * 64-bit:
641 *
642 * Hall of shame of CPU/BIOS bugs.
Nick Piggin92181f12009-01-20 04:24:26 +0100643 */
644 if (is_prefetch(regs, error_code, address))
645 return;
646
647 if (is_errata93(regs, address))
648 return;
649
650 /*
651 * Oops. The kernel tried to access some bad page. We'll have to
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100652 * terminate things with extreme prejudice:
Nick Piggin92181f12009-01-20 04:24:26 +0100653 */
Nick Piggin92181f12009-01-20 04:24:26 +0100654 flags = oops_begin();
Nick Piggin92181f12009-01-20 04:24:26 +0100655
656 show_fault_oops(regs, error_code, address);
657
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100658 stackend = end_of_stack(tsk);
Jan Beulich0e7810b2009-11-20 14:00:14 +0000659 if (tsk != &init_task && *stackend != STACK_END_MAGIC)
Ingo Molnar19803072009-01-21 10:39:51 +0100660 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
661
Ingo Molnar1cc99542009-02-20 23:07:48 +0100662 tsk->thread.cr2 = address;
663 tsk->thread.trap_no = 14;
664 tsk->thread.error_code = error_code;
Nick Piggin92181f12009-01-20 04:24:26 +0100665
Nick Piggin92181f12009-01-20 04:24:26 +0100666 sig = SIGKILL;
667 if (__die("Oops", regs, error_code))
668 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100669
Nick Piggin92181f12009-01-20 04:24:26 +0100670 /* Executive summary in case the body of the oops scrolled away */
671 printk(KERN_EMERG "CR2: %016lx\n", address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100672
Nick Piggin92181f12009-01-20 04:24:26 +0100673 oops_end(flags, regs, sig);
Nick Piggin92181f12009-01-20 04:24:26 +0100674}
675
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100676/*
677 * Print out info about fatal segfaults, if the show_unhandled_signals
678 * sysctl is set:
679 */
680static inline void
681show_signal_msg(struct pt_regs *regs, unsigned long error_code,
682 unsigned long address, struct task_struct *tsk)
683{
684 if (!unhandled_signal(tsk, SIGSEGV))
685 return;
686
687 if (!printk_ratelimit())
688 return;
689
Roland Dreiera1a08d12009-07-11 00:10:04 -0700690 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100691 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
692 tsk->comm, task_pid_nr(tsk), address,
693 (void *)regs->ip, (void *)regs->sp, error_code);
694
695 print_vma_addr(KERN_CONT " in ", regs->ip);
696
697 printk(KERN_CONT "\n");
698}
699
700static void
701__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
702 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100703{
704 struct task_struct *tsk = current;
705
706 /* User mode accesses just cause a SIGSEGV */
707 if (error_code & PF_USER) {
708 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100709 * It's possible to have interrupts off here:
Nick Piggin92181f12009-01-20 04:24:26 +0100710 */
711 local_irq_enable();
712
713 /*
714 * Valid to do another page fault here because this one came
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100715 * from user space:
Nick Piggin92181f12009-01-20 04:24:26 +0100716 */
717 if (is_prefetch(regs, error_code, address))
718 return;
719
720 if (is_errata100(regs, address))
721 return;
722
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100723 if (unlikely(show_unhandled_signals))
724 show_signal_msg(regs, error_code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100725
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100726 /* Kernel addresses are always protection faults: */
727 tsk->thread.cr2 = address;
728 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
729 tsk->thread.trap_no = 14;
730
Andi Kleenf672b492010-09-27 22:05:55 +0200731 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100732
Nick Piggin92181f12009-01-20 04:24:26 +0100733 return;
734 }
735
736 if (is_f00f_bug(regs, address))
737 return;
738
739 no_context(regs, error_code, address);
740}
741
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100742static noinline void
743bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
744 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100745{
746 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
747}
748
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100749static void
750__bad_area(struct pt_regs *regs, unsigned long error_code,
751 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100752{
753 struct mm_struct *mm = current->mm;
754
755 /*
756 * Something tried to access memory that isn't in our memory map..
757 * Fix it, but check if it's kernel or user first..
758 */
759 up_read(&mm->mmap_sem);
760
761 __bad_area_nosemaphore(regs, error_code, address, si_code);
762}
763
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100764static noinline void
765bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100766{
767 __bad_area(regs, error_code, address, SEGV_MAPERR);
768}
769
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100770static noinline void
771bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
772 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100773{
774 __bad_area(regs, error_code, address, SEGV_ACCERR);
775}
776
777/* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100778static void
779out_of_memory(struct pt_regs *regs, unsigned long error_code,
780 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100781{
782 /*
783 * We ran out of memory, call the OOM killer, and return the userspace
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100784 * (which will retry the fault, or kill us if we got oom-killed):
Nick Piggin92181f12009-01-20 04:24:26 +0100785 */
786 up_read(&current->mm->mmap_sem);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100787
Nick Piggin92181f12009-01-20 04:24:26 +0100788 pagefault_out_of_memory();
789}
790
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100791static void
Andi Kleena6e04aa2009-09-16 11:50:09 +0200792do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
793 unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100794{
795 struct task_struct *tsk = current;
796 struct mm_struct *mm = tsk->mm;
Andi Kleena6e04aa2009-09-16 11:50:09 +0200797 int code = BUS_ADRERR;
Nick Piggin92181f12009-01-20 04:24:26 +0100798
799 up_read(&mm->mmap_sem);
800
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100801 /* Kernel mode? Handle exceptions or die: */
Linus Torvalds96054562010-08-13 09:49:20 -0700802 if (!(error_code & PF_USER)) {
Nick Piggin92181f12009-01-20 04:24:26 +0100803 no_context(regs, error_code, address);
Linus Torvalds96054562010-08-13 09:49:20 -0700804 return;
805 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100806
Ingo Molnarcd1b68f2009-02-20 23:39:02 +0100807 /* User-space => ok to do another page fault: */
Nick Piggin92181f12009-01-20 04:24:26 +0100808 if (is_prefetch(regs, error_code, address))
809 return;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100810
811 tsk->thread.cr2 = address;
812 tsk->thread.error_code = error_code;
813 tsk->thread.trap_no = 14;
814
Andi Kleena6e04aa2009-09-16 11:50:09 +0200815#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenf672b492010-09-27 22:05:55 +0200816 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
Andi Kleena6e04aa2009-09-16 11:50:09 +0200817 printk(KERN_ERR
818 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
819 tsk->comm, tsk->pid, address);
820 code = BUS_MCEERR_AR;
821 }
822#endif
Andi Kleenf672b492010-09-27 22:05:55 +0200823 force_sig_info_fault(SIGBUS, code, address, tsk, fault);
Nick Piggin92181f12009-01-20 04:24:26 +0100824}
825
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100826static noinline void
827mm_fault_error(struct pt_regs *regs, unsigned long error_code,
828 unsigned long address, unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100829{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100830 if (fault & VM_FAULT_OOM) {
Andrey Vaginf8626852011-03-09 15:22:23 -0800831 /* Kernel mode? Handle exceptions or die: */
832 if (!(error_code & PF_USER)) {
833 up_read(&current->mm->mmap_sem);
834 no_context(regs, error_code, address);
835 return;
836 }
837
Nick Piggin92181f12009-01-20 04:24:26 +0100838 out_of_memory(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100839 } else {
Andi Kleenf672b492010-09-27 22:05:55 +0200840 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
841 VM_FAULT_HWPOISON_LARGE))
Andi Kleena6e04aa2009-09-16 11:50:09 +0200842 do_sigbus(regs, error_code, address, fault);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100843 else
844 BUG();
845 }
Nick Piggin92181f12009-01-20 04:24:26 +0100846}
847
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100848static int spurious_fault_check(unsigned long error_code, pte_t *pte)
849{
850 if ((error_code & PF_WRITE) && !pte_write(*pte))
851 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100852
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100853 if ((error_code & PF_INSTR) && !pte_exec(*pte))
854 return 0;
855
856 return 1;
857}
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100860 * Handle a spurious fault caused by a stale TLB entry.
861 *
862 * This allows us to lazily refresh the TLB when increasing the
863 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
864 * eagerly is very expensive since that implies doing a full
865 * cross-processor TLB flush, even if no stale TLB entries exist
866 * on other processors.
867 *
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100868 * There are no security implications to leaving a stale TLB when
869 * increasing the permissions on a page.
870 */
Masami Hiramatsu62c92952009-08-27 13:23:11 -0400871static noinline __kprobes int
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100872spurious_fault(unsigned long error_code, unsigned long address)
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100873{
874 pgd_t *pgd;
875 pud_t *pud;
876 pmd_t *pmd;
877 pte_t *pte;
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500878 int ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100879
880 /* Reserved-bit violation or user access to kernel space? */
881 if (error_code & (PF_USER | PF_RSVD))
882 return 0;
883
884 pgd = init_mm.pgd + pgd_index(address);
885 if (!pgd_present(*pgd))
886 return 0;
887
888 pud = pud_offset(pgd, address);
889 if (!pud_present(*pud))
890 return 0;
891
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100892 if (pud_large(*pud))
893 return spurious_fault_check(error_code, (pte_t *) pud);
894
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100895 pmd = pmd_offset(pud, address);
896 if (!pmd_present(*pmd))
897 return 0;
898
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100899 if (pmd_large(*pmd))
900 return spurious_fault_check(error_code, (pte_t *) pmd);
901
Shaohua Li660a2932010-07-27 16:06:28 +0800902 /*
903 * Note: don't use pte_present() here, since it returns true
904 * if the _PAGE_PROTNONE bit is set. However, this aliases the
905 * _PAGE_GLOBAL bit, which for kernel pages give false positives
906 * when CONFIG_DEBUG_PAGEALLOC is used.
907 */
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100908 pte = pte_offset_kernel(pmd, address);
Shaohua Li660a2932010-07-27 16:06:28 +0800909 if (!(pte_flags(*pte) & _PAGE_PRESENT))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100910 return 0;
911
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500912 ret = spurious_fault_check(error_code, pte);
913 if (!ret)
914 return 0;
915
916 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100917 * Make sure we have permissions in PMD.
918 * If not, then there's a bug in the page tables:
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500919 */
920 ret = spurious_fault_check(error_code, (pte_t *) pmd);
921 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100922
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500923 return ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100924}
925
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +0200926int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100928static inline int
Michel Lespinasse68da3362010-10-26 14:21:58 -0700929access_error(unsigned long error_code, struct vm_area_struct *vma)
Nick Piggin92181f12009-01-20 04:24:26 +0100930{
Michel Lespinasse68da3362010-10-26 14:21:58 -0700931 if (error_code & PF_WRITE) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100932 /* write, present and write, not present: */
Nick Piggin92181f12009-01-20 04:24:26 +0100933 if (unlikely(!(vma->vm_flags & VM_WRITE)))
934 return 1;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100935 return 0;
Nick Piggin92181f12009-01-20 04:24:26 +0100936 }
937
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100938 /* read, present: */
939 if (unlikely(error_code & PF_PROT))
940 return 1;
941
942 /* read, not present: */
943 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
944 return 1;
945
Nick Piggin92181f12009-01-20 04:24:26 +0100946 return 0;
947}
948
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800949static int fault_in_kernel_space(unsigned long address)
950{
Ingo Molnard9517342009-02-20 23:32:28 +0100951 return address >= TASK_SIZE_MAX;
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -0800952}
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954/*
955 * This routine handles page faults. It determines the address,
956 * and the problem, and then passes it off to one of the appropriate
957 * routines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 */
Ingo Molnarc3731c62009-02-20 23:22:34 +0100959dotraplinkage void __kprobes
960do_page_fault(struct pt_regs *regs, unsigned long error_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961{
Harvey Harrison33cb5242008-01-30 13:32:19 +0100962 struct vm_area_struct *vma;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100963 struct task_struct *tsk;
964 unsigned long address;
965 struct mm_struct *mm;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +0100966 int fault;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700967 int write = error_code & PF_WRITE;
968 unsigned int flags = FAULT_FLAG_ALLOW_RETRY |
969 (write ? FAULT_FLAG_WRITE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Arjan van de Vena9ba9a32006-03-25 16:30:10 +0100971 tsk = current;
972 mm = tsk->mm;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100973
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100974 /* Get the faulting address: */
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200975 address = read_cr2();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Vegard Nossumf8561292008-04-04 00:53:23 +0200977 /*
978 * Detect and handle instructions that would cause a page fault for
979 * both a tracked kernel page and a userspace page.
980 */
981 if (kmemcheck_active(regs))
982 kmemcheck_hide(regs);
Ingo Molnar5dfaf902009-06-16 10:23:32 +0200983 prefetchw(&mm->mmap_sem);
Vegard Nossumf8561292008-04-04 00:53:23 +0200984
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +0200985 if (unlikely(kmmio_fault(regs, address)))
Pekka Paalanen86069782008-05-12 21:20:56 +0200986 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 /*
989 * We fault-in kernel-space virtual memory on-demand. The
990 * 'reference' page table is init_mm.pgd.
991 *
992 * NOTE! We MUST NOT take any locks for this case. We may
993 * be in an interrupt or a critical region, and should
994 * only copy the information from the master page table,
995 * nothing more.
996 *
997 * This verifies that the fault happens in kernel space
998 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +0100999 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 */
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -08001001 if (unlikely(fault_in_kernel_space(address))) {
Vegard Nossumf8561292008-04-04 00:53:23 +02001002 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1003 if (vmalloc_fault(address) >= 0)
1004 return;
1005
1006 if (kmemcheck_fault(regs, address, error_code))
1007 return;
1008 }
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001009
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001010 /* Can handle a stale RO->RW TLB: */
Nick Piggin92181f12009-01-20 04:24:26 +01001011 if (spurious_fault(error_code, address))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001012 return;
1013
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001014 /* kprobes don't want to hook the spurious faults: */
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001015 if (notify_page_fault(regs))
1016 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001017 /*
1018 * Don't take the mm semaphore here. If we fixup a prefetch
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001019 * fault we could otherwise deadlock:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001020 */
Nick Piggin92181f12009-01-20 04:24:26 +01001021 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001022
Nick Piggin92181f12009-01-20 04:24:26 +01001023 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001024 }
1025
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001026 /* kprobes don't want to hook the spurious faults: */
Ingo Molnarf8a6b2b2009-02-13 09:44:22 +01001027 if (unlikely(notify_page_fault(regs)))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001028 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001029 /*
Linus Torvalds891cffb2008-10-12 13:16:12 -07001030 * It's safe to allow irq's after cr2 has been saved and the
1031 * vmalloc fault has been handled.
1032 *
1033 * User-mode registers count as a user access even for any
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001034 * potential system fault or CPU buglet:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001035 */
Linus Torvalds891cffb2008-10-12 13:16:12 -07001036 if (user_mode_vm(regs)) {
1037 local_irq_enable();
1038 error_code |= PF_USER;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001039 } else {
1040 if (regs->flags & X86_EFLAGS_IF)
1041 local_irq_enable();
1042 }
Jan Beulich8c914cb2006-03-25 16:29:40 +01001043
Andi Kleen66c58152006-01-11 22:44:09 +01001044 if (unlikely(error_code & PF_RSVD))
Nick Piggin92181f12009-01-20 04:24:26 +01001045 pgtable_bad(regs, error_code, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001047 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
Peter Zijlstra7dd1fcc2009-03-13 12:21:33 +01001048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001050 * If we're in an interrupt, have no user context or are running
1051 * in an atomic region then we must not take the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 */
Nick Piggin92181f12009-01-20 04:24:26 +01001053 if (unlikely(in_atomic() || !mm)) {
1054 bad_area_nosemaphore(regs, error_code, address);
1055 return;
1056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Ingo Molnar3a1dfe62008-10-13 17:49:02 +02001058 /*
1059 * When running in the kernel we expect faults to occur only to
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001060 * addresses in user space. All other faults represent errors in
1061 * the kernel and should generate an OOPS. Unfortunately, in the
1062 * case of an erroneous fault occurring in a code path which already
1063 * holds mmap_sem we will deadlock attempting to validate the fault
1064 * against the address space. Luckily the kernel only validly
1065 * references user space from well defined areas of code, which are
1066 * listed in the exceptions table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 *
1068 * As the vast majority of faults will be valid we will only perform
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001069 * the source reference check when there is a possibility of a
1070 * deadlock. Attempt to lock the address space, if we cannot we then
1071 * validate the source. If this is invalid we can skip the address
1072 * space check, thus avoiding the deadlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 */
Nick Piggin92181f12009-01-20 04:24:26 +01001074 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
Andi Kleen66c58152006-01-11 22:44:09 +01001075 if ((error_code & PF_USER) == 0 &&
Nick Piggin92181f12009-01-20 04:24:26 +01001076 !search_exception_tables(regs->ip)) {
1077 bad_area_nosemaphore(regs, error_code, address);
1078 return;
1079 }
Michel Lespinassed065bd82010-10-26 14:21:57 -07001080retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 down_read(&mm->mmap_sem);
Peter Zijlstra01006072009-01-29 16:02:12 +01001082 } else {
1083 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001084 * The above down_read_trylock() might have succeeded in
1085 * which case we'll have missed the might_sleep() from
1086 * down_read():
Peter Zijlstra01006072009-01-29 16:02:12 +01001087 */
1088 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 }
1090
1091 vma = find_vma(mm, address);
Nick Piggin92181f12009-01-20 04:24:26 +01001092 if (unlikely(!vma)) {
1093 bad_area(regs, error_code, address);
1094 return;
1095 }
1096 if (likely(vma->vm_start <= address))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 goto good_area;
Nick Piggin92181f12009-01-20 04:24:26 +01001098 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1099 bad_area(regs, error_code, address);
1100 return;
1101 }
Harvey Harrison33cb5242008-01-30 13:32:19 +01001102 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001103 /*
1104 * Accessing the stack below %sp is always a bug.
1105 * The large cushion allows instructions like enter
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001106 * and pusha to work. ("enter $65535, $31" pushes
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001107 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +02001108 */
Nick Piggin92181f12009-01-20 04:24:26 +01001109 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1110 bad_area(regs, error_code, address);
1111 return;
1112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 }
Nick Piggin92181f12009-01-20 04:24:26 +01001114 if (unlikely(expand_stack(vma, address))) {
1115 bad_area(regs, error_code, address);
1116 return;
1117 }
1118
1119 /*
1120 * Ok, we have a good vm_area for this memory access, so
1121 * we can handle it..
1122 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123good_area:
Michel Lespinasse68da3362010-10-26 14:21:58 -07001124 if (unlikely(access_error(error_code, vma))) {
Nick Piggin92181f12009-01-20 04:24:26 +01001125 bad_area_access_error(regs, error_code, address);
1126 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128
1129 /*
1130 * If for any reason at all we couldn't handle the fault,
1131 * make sure we exit gracefully rather than endlessly redo
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001132 * the fault:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001134 fault = handle_mm_fault(mm, vma, address, flags);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001135
Nick Piggin83c54072007-07-19 01:47:05 -07001136 if (unlikely(fault & VM_FAULT_ERROR)) {
Nick Piggin92181f12009-01-20 04:24:26 +01001137 mm_fault_error(regs, error_code, address, fault);
1138 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001140
Michel Lespinassed065bd82010-10-26 14:21:57 -07001141 /*
1142 * Major/minor page fault accounting is only done on the
1143 * initial attempt. If we go through a retry, it is extremely
1144 * likely that the page will be found in page cache at that point.
1145 */
1146 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1147 if (fault & VM_FAULT_MAJOR) {
1148 tsk->maj_flt++;
1149 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
1150 regs, address);
1151 } else {
1152 tsk->min_flt++;
1153 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
1154 regs, address);
1155 }
1156 if (fault & VM_FAULT_RETRY) {
1157 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
1158 * of starvation. */
1159 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1160 goto retry;
1161 }
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001162 }
Harvey Harrisond729ab32008-01-30 13:33:23 +01001163
Ingo Molnar8c938f92009-02-20 22:12:18 +01001164 check_v8086_mode(regs, address, tsk);
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 up_read(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}