blob: 9dc909841739bf24b01d7d2e574c4870cc409cfd [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/sched.h> /* test_thread_flag(), ... */
7#include <linux/kdebug.h> /* oops_begin/end, ... */
8#include <linux/module.h> /* search_exception_table */
9#include <linux/bootmem.h> /* max_low_pfn */
Masami Hiramatsu93266382014-04-17 17:18:14 +090010#include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */
Ingo Molnara2bcd472009-03-29 23:47:48 +020011#include <linux/mmiotrace.h> /* kmmio_handler, ... */
Ingo Molnarcdd6c482009-09-21 12:02:48 +020012#include <linux/perf_event.h> /* perf_sw_event */
Andi Kleenf672b492010-09-27 22:05:55 +020013#include <linux/hugetlb.h> /* hstate_index_to_shift */
Linus Torvalds268bb0c2011-05-20 12:50:29 -070014#include <linux/prefetch.h> /* prefetchw */
Frederic Weisbecker56dd9472013-02-24 00:23:25 +010015#include <linux/context_tracking.h> /* exception_enter(), ... */
David Hildenbrand70ffdb92015-05-11 17:52:11 +020016#include <linux/uaccess.h> /* faulthandler_disabled() */
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
Ingo Molnara2bcd472009-03-29 23:47:48 +020018#include <asm/traps.h> /* dotraplinkage, ... */
19#include <asm/pgalloc.h> /* pgd_*(), ... */
Vegard Nossumf8561292008-04-04 00:53:23 +020020#include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
Andy Lutomirskif40c3302014-05-05 12:19:36 -070021#include <asm/fixmap.h> /* VSYSCALL_ADDR */
22#include <asm/vsyscall.h> /* emulate_vsyscall */
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Seiji Aguchid34603b2013-10-30 16:39:03 -040024#define CREATE_TRACE_POINTS
25#include <asm/trace/exceptions.h>
26
Harvey Harrison33cb5242008-01-30 13:32:19 +010027/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010028 * Page fault error code bits:
29 *
30 * bit 0 == 0: no page found 1: protection fault
31 * bit 1 == 0: read access 1: write access
32 * bit 2 == 0: kernel-mode access 1: user-mode access
33 * bit 3 == 1: use of reserved bit detected
34 * bit 4 == 1: fault was an instruction fetch
Harvey Harrison33cb5242008-01-30 13:32:19 +010035 */
Ingo Molnar2d4a7162009-02-20 19:56:40 +010036enum x86_pf_error_code {
37
38 PF_PROT = 1 << 0,
39 PF_WRITE = 1 << 1,
40 PF_USER = 1 << 2,
41 PF_RSVD = 1 << 3,
42 PF_INSTR = 1 << 4,
43};
Andi Kleen66c58152006-01-11 22:44:09 +010044
Ingo Molnarb814d412009-02-20 22:32:10 +010045/*
Ingo Molnarb319eed2009-02-22 10:24:18 +010046 * Returns 0 if mmiotrace is disabled, or if the fault is not
47 * handled by mmiotrace:
Ingo Molnarb814d412009-02-20 22:32:10 +010048 */
Masami Hiramatsu93266382014-04-17 17:18:14 +090049static nokprobe_inline int
Masami Hiramatsu62c92952009-08-27 13:23:11 -040050kmmio_fault(struct pt_regs *regs, unsigned long addr)
Pekka Paalanen86069782008-05-12 21:20:56 +020051{
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020052 if (unlikely(is_kmmio_active()))
53 if (kmmio_handler(regs, addr) == 1)
54 return -1;
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +020055 return 0;
Pekka Paalanen86069782008-05-12 21:20:56 +020056}
57
Masami Hiramatsu93266382014-04-17 17:18:14 +090058static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070059{
Christoph Hellwig74a0b572007-10-16 01:24:07 -070060 int ret = 0;
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070061
Christoph Hellwig74a0b572007-10-16 01:24:07 -070062 /* kprobe_running() needs smp_processor_id() */
Andy Lutomirskif39b6f02015-03-18 18:33:33 -070063 if (kprobes_built_in() && !user_mode(regs)) {
Christoph Hellwig74a0b572007-10-16 01:24:07 -070064 preempt_disable();
65 if (kprobe_running() && kprobe_fault_handler(regs, 14))
66 ret = 1;
67 preempt_enable();
68 }
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070069
Christoph Hellwig74a0b572007-10-16 01:24:07 -070070 return ret;
Harvey Harrison33cb5242008-01-30 13:32:19 +010071}
Anil S Keshavamurthy1bd858a2006-06-26 00:25:25 -070072
Harvey Harrison1dc85be2008-01-30 13:32:35 +010073/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +010074 * Prefetch quirks:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010075 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010076 * 32-bit mode:
Harvey Harrison1dc85be2008-01-30 13:32:35 +010077 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +010078 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
79 * Check that here and ignore it.
80 *
81 * 64-bit mode:
82 *
83 * Sometimes the CPU reports invalid exceptions on prefetch.
84 * Check that here and ignore it.
85 *
86 * Opcode checker based on code by Richard Brunner.
Harvey Harrison1dc85be2008-01-30 13:32:35 +010087 */
Ingo Molnar107a0362009-02-20 20:37:05 +010088static inline int
89check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
90 unsigned char opcode, int *prefetch)
91{
92 unsigned char instr_hi = opcode & 0xf0;
93 unsigned char instr_lo = opcode & 0x0f;
94
95 switch (instr_hi) {
96 case 0x20:
97 case 0x30:
98 /*
99 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
100 * In X86_64 long mode, the CPU will signal invalid
101 * opcode if some of these prefixes are present so
102 * X86_64 will never get here anyway
103 */
104 return ((instr_lo & 7) == 0x6);
105#ifdef CONFIG_X86_64
106 case 0x40:
107 /*
108 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
109 * Need to figure out under what instruction mode the
110 * instruction was issued. Could check the LDT for lm,
111 * but for now it's good enough to assume that long
112 * mode only uses well known segments or kernel.
113 */
Andy Lutomirski318f5a22011-08-03 09:31:53 -0400114 return (!user_mode(regs) || user_64bit_mode(regs));
Ingo Molnar107a0362009-02-20 20:37:05 +0100115#endif
116 case 0x60:
117 /* 0x64 thru 0x67 are valid prefixes in all modes. */
118 return (instr_lo & 0xC) == 0x4;
119 case 0xF0:
120 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
121 return !instr_lo || (instr_lo>>1) == 1;
122 case 0x00:
123 /* Prefetch instruction is 0x0F0D or 0x0F18 */
124 if (probe_kernel_address(instr, opcode))
125 return 0;
126
127 *prefetch = (instr_lo == 0xF) &&
128 (opcode == 0x0D || opcode == 0x18);
129 return 0;
130 default:
131 return 0;
132 }
133}
134
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100135static int
136is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
Harvey Harrison33cb5242008-01-30 13:32:19 +0100137{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100138 unsigned char *max_instr;
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100139 unsigned char *instr;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100140 int prefetch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Ingo Molnar30853542008-03-27 21:29:09 +0100142 /*
143 * If it was a exec (instruction fetch) fault on NX page, then
144 * do not ignore the fault:
145 */
Andi Kleen66c58152006-01-11 22:44:09 +0100146 if (error_code & PF_INSTR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return 0;
Harvey Harrison1dc85be2008-01-30 13:32:35 +0100148
Ingo Molnar107a0362009-02-20 20:37:05 +0100149 instr = (void *)convert_ip_to_linear(current, regs);
Andi Kleenf1290ec2005-04-16 15:24:59 -0700150 max_instr = instr + 15;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Andy Lutomirskid31bf072015-03-18 18:33:27 -0700152 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 return 0;
154
Ingo Molnar107a0362009-02-20 20:37:05 +0100155 while (instr < max_instr) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100156 unsigned char opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Andi Kleenab2bf0c2006-12-07 02:14:06 +0100158 if (probe_kernel_address(instr, opcode))
Harvey Harrison33cb5242008-01-30 13:32:19 +0100159 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 instr++;
162
Ingo Molnar107a0362009-02-20 20:37:05 +0100163 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
166 return prefetch;
167}
168
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100169static void
170force_sig_info_fault(int si_signo, int si_code, unsigned long address,
Andi Kleenf672b492010-09-27 22:05:55 +0200171 struct task_struct *tsk, int fault)
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100172{
Andi Kleenf672b492010-09-27 22:05:55 +0200173 unsigned lsb = 0;
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100174 siginfo_t info;
175
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100176 info.si_signo = si_signo;
177 info.si_errno = 0;
178 info.si_code = si_code;
179 info.si_addr = (void __user *)address;
Andi Kleenf672b492010-09-27 22:05:55 +0200180 if (fault & VM_FAULT_HWPOISON_LARGE)
181 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
182 if (fault & VM_FAULT_HWPOISON)
183 lsb = PAGE_SHIFT;
184 info.si_addr_lsb = lsb;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100185
Harvey Harrisonc4aba4a2008-01-30 13:32:35 +0100186 force_sig_info(si_signo, &info, tsk);
187}
188
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100189DEFINE_SPINLOCK(pgd_lock);
190LIST_HEAD(pgd_list);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100191
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100192#ifdef CONFIG_X86_32
193static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
194{
195 unsigned index = pgd_index(address);
196 pgd_t *pgd_k;
197 pud_t *pud, *pud_k;
198 pmd_t *pmd, *pmd_k;
199
200 pgd += index;
201 pgd_k = init_mm.pgd + index;
202
203 if (!pgd_present(*pgd_k))
204 return NULL;
205
206 /*
207 * set_pgd(pgd, *pgd_k); here would be useless on PAE
208 * and redundant with the set_pmd() on non-PAE. As would
209 * set_pud.
210 */
211 pud = pud_offset(pgd, address);
212 pud_k = pud_offset(pgd_k, address);
213 if (!pud_present(*pud_k))
214 return NULL;
215
216 pmd = pmd_offset(pud, address);
217 pmd_k = pmd_offset(pud_k, address);
218 if (!pmd_present(*pmd_k))
219 return NULL;
220
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800221 if (!pmd_present(*pmd))
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100222 set_pmd(pmd, *pmd_k);
Jeremy Fitzhardingeb8bcfe92009-02-17 23:05:19 -0800223 else
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100224 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100225
226 return pmd_k;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100227}
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100228
229void vmalloc_sync_all(void)
230{
231 unsigned long address;
232
233 if (SHARED_KERNEL_PMD)
234 return;
235
236 for (address = VMALLOC_START & PMD_MASK;
237 address >= TASK_SIZE && address < FIXADDR_TOP;
238 address += PMD_SIZE) {
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100239 struct page *page;
240
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800241 spin_lock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100242 list_for_each_entry(page, &pgd_list, lru) {
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700243 spinlock_t *pgt_lock;
Borislav Petkovf01f7c52010-10-19 22:17:37 +0000244 pmd_t *ret;
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700245
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800246 /* the pgt_lock only for Xen */
Jeremy Fitzhardinge617d34d2010-09-21 12:01:51 -0700247 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
248
249 spin_lock(pgt_lock);
250 ret = vmalloc_sync_one(page_address(page), address);
251 spin_unlock(pgt_lock);
252
253 if (!ret)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100254 break;
255 }
Andrea Arcangelia79e53d2011-02-16 15:45:22 -0800256 spin_unlock(&pgd_lock);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100257 }
258}
259
260/*
261 * 32-bit:
262 *
263 * Handle a fault on the vmalloc or module mapping area
264 */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900265static noinline int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100266{
267 unsigned long pgd_paddr;
268 pmd_t *pmd_k;
269 pte_t *pte_k;
270
271 /* Make sure we are in vmalloc area: */
272 if (!(address >= VMALLOC_START && address < VMALLOC_END))
273 return -1;
274
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200275 WARN_ON_ONCE(in_nmi());
276
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100277 /*
278 * Synchronize this task's top level page-table
279 * with the 'reference' page table.
280 *
281 * Do _not_ use "current" here. We might be inside
282 * an interrupt in the middle of a task switch..
283 */
284 pgd_paddr = read_cr3();
285 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
286 if (!pmd_k)
287 return -1;
288
289 pte_k = pte_offset_kernel(pmd_k, address);
290 if (!pte_present(*pte_k))
291 return -1;
292
293 return 0;
294}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900295NOKPROBE_SYMBOL(vmalloc_fault);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100296
297/*
298 * Did it hit the DOS screen memory VA from vm86 mode?
299 */
300static inline void
301check_v8086_mode(struct pt_regs *regs, unsigned long address,
302 struct task_struct *tsk)
303{
304 unsigned long bit;
305
306 if (!v8086_mode(regs))
307 return;
308
309 bit = (address - 0xA0000) >> PAGE_SHIFT;
310 if (bit < 32)
311 tsk->thread.screen_bitmap |= 1 << bit;
312}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Akinobu Mita087975b2009-06-27 15:35:15 +0900314static bool low_pfn(unsigned long pfn)
315{
316 return pfn < max_low_pfn;
317}
318
Adrian Bunkcae30f822008-02-13 23:31:31 +0200319static void dump_pagetable(unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
Akinobu Mita087975b2009-06-27 15:35:15 +0900321 pgd_t *base = __va(read_cr3());
322 pgd_t *pgd = &base[pgd_index(address)];
323 pmd_t *pmd;
324 pte_t *pte;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100325
Harvey Harrison1156e092008-01-30 13:34:10 +0100326#ifdef CONFIG_X86_PAE
Akinobu Mita087975b2009-06-27 15:35:15 +0900327 printk("*pdpt = %016Lx ", pgd_val(*pgd));
328 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
329 goto out;
Harvey Harrison1156e092008-01-30 13:34:10 +0100330#endif
Akinobu Mita087975b2009-06-27 15:35:15 +0900331 pmd = pmd_offset(pud_offset(pgd, address), address);
332 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
Harvey Harrison1156e092008-01-30 13:34:10 +0100333
334 /*
335 * We must not directly access the pte in the highpte
336 * case if the page table is located in highmem.
337 * And let's rather not kmap-atomic the pte, just in case
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100338 * it's allocated already:
Harvey Harrison1156e092008-01-30 13:34:10 +0100339 */
Akinobu Mita087975b2009-06-27 15:35:15 +0900340 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
341 goto out;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100342
Akinobu Mita087975b2009-06-27 15:35:15 +0900343 pte = pte_offset_kernel(pmd, address);
344 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
345out:
Harvey Harrison1156e092008-01-30 13:34:10 +0100346 printk("\n");
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100347}
348
349#else /* CONFIG_X86_64: */
350
351void vmalloc_sync_all(void)
352{
Yasuaki Ishimatsu9661d5b2014-08-22 13:27:34 -0700353 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100354}
355
356/*
357 * 64-bit:
358 *
359 * Handle a fault on the vmalloc area
360 *
361 * This assumes no large pages in there.
362 */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900363static noinline int vmalloc_fault(unsigned long address)
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100364{
365 pgd_t *pgd, *pgd_ref;
366 pud_t *pud, *pud_ref;
367 pmd_t *pmd, *pmd_ref;
368 pte_t *pte, *pte_ref;
369
370 /* Make sure we are in vmalloc area: */
371 if (!(address >= VMALLOC_START && address < VMALLOC_END))
372 return -1;
373
Frederic Weisbeckerebc88272010-09-27 18:50:51 +0200374 WARN_ON_ONCE(in_nmi());
375
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100376 /*
377 * Copy kernel mappings over when needed. This can also
378 * happen within a race in page table update. In the later
379 * case just flush:
380 */
381 pgd = pgd_offset(current->active_mm, address);
382 pgd_ref = pgd_offset_k(address);
383 if (pgd_none(*pgd_ref))
384 return -1;
385
Samu Kallio1160c272013-03-23 09:36:35 -0400386 if (pgd_none(*pgd)) {
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100387 set_pgd(pgd, *pgd_ref);
Samu Kallio1160c272013-03-23 09:36:35 -0400388 arch_flush_lazy_mmu_mode();
389 } else {
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100390 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
Samu Kallio1160c272013-03-23 09:36:35 -0400391 }
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100392
393 /*
394 * Below here mismatches are bugs because these lower tables
395 * are shared:
396 */
397
398 pud = pud_offset(pgd, address);
399 pud_ref = pud_offset(pgd_ref, address);
400 if (pud_none(*pud_ref))
401 return -1;
402
403 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
404 BUG();
405
406 pmd = pmd_offset(pud, address);
407 pmd_ref = pmd_offset(pud_ref, address);
408 if (pmd_none(*pmd_ref))
409 return -1;
410
411 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
412 BUG();
413
414 pte_ref = pte_offset_kernel(pmd_ref, address);
415 if (!pte_present(*pte_ref))
416 return -1;
417
418 pte = pte_offset_kernel(pmd, address);
419
420 /*
421 * Don't use pte_page here, because the mappings can point
422 * outside mem_map, and the NUMA hash lookup cannot handle
423 * that:
424 */
425 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
426 BUG();
427
428 return 0;
429}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900430NOKPROBE_SYMBOL(vmalloc_fault);
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100431
Jan Beuliche05139f2011-09-28 16:56:51 +0100432#ifdef CONFIG_CPU_SUP_AMD
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100433static const char errata93_warning[] =
Joe Perchesad361c92009-07-06 13:05:40 -0700434KERN_ERR
435"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
436"******* Working around it, but it may cause SEGVs or burn power.\n"
437"******* Please consider a BIOS update.\n"
438"******* Disabling USB legacy in the BIOS may also help.\n";
Jan Beuliche05139f2011-09-28 16:56:51 +0100439#endif
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100440
441/*
442 * No vm86 mode in 64-bit mode:
443 */
444static inline void
445check_v8086_mode(struct pt_regs *regs, unsigned long address,
446 struct task_struct *tsk)
447{
448}
449
450static int bad_address(void *p)
451{
452 unsigned long dummy;
453
454 return probe_kernel_address((unsigned long *)p, dummy);
455}
456
457static void dump_pagetable(unsigned long address)
458{
Akinobu Mita087975b2009-06-27 15:35:15 +0900459 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
460 pgd_t *pgd = base + pgd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 pud_t *pud;
462 pmd_t *pmd;
463 pte_t *pte;
464
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100465 if (bad_address(pgd))
466 goto bad;
467
Jan Beulichd646bce2006-02-03 21:51:47 +0100468 printk("PGD %lx ", pgd_val(*pgd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100469
470 if (!pgd_present(*pgd))
471 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Andi Kleend2ae5b52006-06-26 13:57:56 +0200473 pud = pud_offset(pgd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100474 if (bad_address(pud))
475 goto bad;
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 printk("PUD %lx ", pud_val(*pud));
Andi Kleenb5360222008-02-04 16:48:09 +0100478 if (!pud_present(*pud) || pud_large(*pud))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100479 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 pmd = pmd_offset(pud, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100482 if (bad_address(pmd))
483 goto bad;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 printk("PMD %lx ", pmd_val(*pmd));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100486 if (!pmd_present(*pmd) || pmd_large(*pmd))
487 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 pte = pte_offset_kernel(pmd, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100490 if (bad_address(pte))
491 goto bad;
492
Harvey Harrison33cb5242008-01-30 13:32:19 +0100493 printk("PTE %lx", pte_val(*pte));
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100494out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 printk("\n");
496 return;
497bad:
498 printk("BAD\n");
499}
500
Ingo Molnarf2f13a82009-02-20 22:50:24 +0100501#endif /* CONFIG_X86_64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100503/*
504 * Workaround for K8 erratum #93 & buggy BIOS.
505 *
506 * BIOS SMM functions are required to use a specific workaround
507 * to avoid corruption of the 64bit RIP register on C stepping K8.
508 *
509 * A lot of BIOS that didn't get tested properly miss this.
510 *
511 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
512 * Try to work around it here.
513 *
514 * Note we only handle faults in kernel here.
515 * Does nothing on 32-bit.
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100516 */
Harvey Harrison33cb5242008-01-30 13:32:19 +0100517static int is_errata93(struct pt_regs *regs, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
Jan Beuliche05139f2011-09-28 16:56:51 +0100519#if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
520 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
521 || boot_cpu_data.x86 != 0xf)
522 return 0;
523
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100524 if (address != regs->ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100526
Harvey Harrison33cb5242008-01-30 13:32:19 +0100527 if ((address >> 32) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 address |= 0xffffffffUL << 32;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100531 if ((address >= (u64)_stext && address <= (u64)_etext) ||
532 (address >= MODULES_VADDR && address <= MODULES_END)) {
Ingo Molnara454ab32009-05-03 10:09:03 +0200533 printk_once(errata93_warning);
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100534 regs->ip = address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return 1;
536 }
Harvey Harrisonfdfe8aa2008-01-30 13:33:13 +0100537#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return 0;
Harvey Harrison33cb5242008-01-30 13:32:19 +0100539}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Harvey Harrison35f32662008-01-30 13:34:09 +0100541/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100542 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
543 * to illegal addresses >4GB.
544 *
545 * We catch this in the page fault handler because these addresses
546 * are not reachable. Just detect this case and return. Any code
Harvey Harrison35f32662008-01-30 13:34:09 +0100547 * segment in LDT is compatibility mode.
548 */
549static int is_errata100(struct pt_regs *regs, unsigned long address)
550{
551#ifdef CONFIG_X86_64
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100552 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
Harvey Harrison35f32662008-01-30 13:34:09 +0100553 return 1;
554#endif
555 return 0;
556}
557
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100558static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
559{
560#ifdef CONFIG_X86_F00F_BUG
561 unsigned long nr;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100562
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100563 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100564 * Pentium F0 0F C7 C8 bug workaround:
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100565 */
Borislav Petkove2604b42013-03-20 15:07:24 +0100566 if (boot_cpu_has_bug(X86_BUG_F00F)) {
Harvey Harrison29caf2f2008-01-30 13:34:09 +0100567 nr = (address - idt_descr.address) >> 3;
568
569 if (nr == 6) {
570 do_invalid_op(regs, 0);
571 return 1;
572 }
573 }
574#endif
575 return 0;
576}
577
Ingo Molnar8f766142009-02-20 23:00:29 +0100578static const char nx_warning[] = KERN_CRIT
579"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
Jiri Kosinaeff50c32014-06-10 22:49:31 +0200580static const char smep_warning[] = KERN_CRIT
581"unable to execute userspace code (SMEP?) (uid: %d)\n";
Ingo Molnar8f766142009-02-20 23:00:29 +0100582
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100583static void
584show_fault_oops(struct pt_regs *regs, unsigned long error_code,
585 unsigned long address)
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100586{
Harvey Harrison1156e092008-01-30 13:34:10 +0100587 if (!oops_may_print())
588 return;
589
Harvey Harrison1156e092008-01-30 13:34:10 +0100590 if (error_code & PF_INSTR) {
Harvey Harrison93809be2008-02-01 17:49:43 +0100591 unsigned int level;
Matt Fleming426e34c2013-12-06 21:13:04 +0000592 pgd_t *pgd;
593 pte_t *pte;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100594
Matt Fleming426e34c2013-12-06 21:13:04 +0000595 pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK);
596 pgd += pgd_index(address);
597
598 pte = lookup_address_in_pgd(pgd, address, &level);
Harvey Harrison1156e092008-01-30 13:34:10 +0100599
Ingo Molnar8f766142009-02-20 23:00:29 +0100600 if (pte && pte_present(*pte) && !pte_exec(*pte))
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800601 printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
Jiri Kosinaeff50c32014-06-10 22:49:31 +0200602 if (pte && pte_present(*pte) && pte_exec(*pte) &&
603 (pgd_flags(*pgd) & _PAGE_USER) &&
Andy Lutomirski1e02ce42014-10-24 15:58:08 -0700604 (__read_cr4() & X86_CR4_SMEP))
Jiri Kosinaeff50c32014-06-10 22:49:31 +0200605 printk(smep_warning, from_kuid(&init_user_ns, current_uid()));
Harvey Harrison1156e092008-01-30 13:34:10 +0100606 }
Harvey Harrisonfd40d6e2008-01-30 13:34:11 +0100607
Harvey Harrison1156e092008-01-30 13:34:10 +0100608 printk(KERN_ALERT "BUG: unable to handle kernel ");
609 if (address < PAGE_SIZE)
610 printk(KERN_CONT "NULL pointer dereference");
611 else
612 printk(KERN_CONT "paging request");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100613
Vegard Nossumf294a8c2008-07-01 15:38:13 +0200614 printk(KERN_CONT " at %p\n", (void *) address);
Harvey Harrison19f0dda2008-01-30 13:34:10 +0100615 printk(KERN_ALERT "IP:");
Jiri Slaby5f01c982013-10-25 15:06:58 +0200616 printk_address(regs->ip);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100617
Harvey Harrisonb3279c72008-01-30 13:34:10 +0100618 dump_pagetable(address);
619}
620
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100621static noinline void
622pgtable_bad(struct pt_regs *regs, unsigned long error_code,
623 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100625 struct task_struct *tsk;
626 unsigned long flags;
627 int sig;
628
629 flags = oops_begin();
630 tsk = current;
631 sig = SIGKILL;
Jan Beulich12091402005-09-12 18:49:24 +0200632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
Nick Piggin92181f12009-01-20 04:24:26 +0100634 tsk->comm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 dump_pagetable(address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100636
637 tsk->thread.cr2 = address;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530638 tsk->thread.trap_nr = X86_TRAP_PF;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100639 tsk->thread.error_code = error_code;
640
Jan Beulich22f59912008-01-30 13:31:23 +0100641 if (__die("Bad pagetable", regs, error_code))
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200642 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100643
Alexander van Heukelum874d93d2008-10-22 12:00:09 +0200644 oops_end(flags, regs, sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100647static noinline void
648no_context(struct pt_regs *regs, unsigned long error_code,
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800649 unsigned long address, int signal, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100650{
651 struct task_struct *tsk = current;
Nick Piggin92181f12009-01-20 04:24:26 +0100652 unsigned long flags;
653 int sig;
Nick Piggin92181f12009-01-20 04:24:26 +0100654
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100655 /* Are we prepared to handle this kernel fault? */
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800656 if (fixup_exception(regs)) {
Peter Zijlstrac026b352014-01-10 21:06:03 +0100657 /*
658 * Any interrupt that takes a fault gets the fixup. This makes
659 * the below recursive fault logic only apply to a faults from
660 * task context.
661 */
662 if (in_interrupt())
663 return;
664
665 /*
666 * Per the above we're !in_interrupt(), aka. task context.
667 *
668 * In this case we need to make sure we're not recursively
669 * faulting through the emulate_vsyscall() logic.
670 */
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800671 if (current_thread_info()->sig_on_uaccess_error && signal) {
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530672 tsk->thread.trap_nr = X86_TRAP_PF;
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800673 tsk->thread.error_code = error_code | PF_USER;
674 tsk->thread.cr2 = address;
675
676 /* XXX: hwpoison faults will set the wrong code. */
677 force_sig_info_fault(signal, si_code, address, tsk, 0);
678 }
Peter Zijlstrac026b352014-01-10 21:06:03 +0100679
680 /*
681 * Barring that, we can do the fixup and be happy.
682 */
Nick Piggin92181f12009-01-20 04:24:26 +0100683 return;
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800684 }
Nick Piggin92181f12009-01-20 04:24:26 +0100685
686 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100687 * 32-bit:
Nick Piggin92181f12009-01-20 04:24:26 +0100688 *
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100689 * Valid to do another page fault here, because if this fault
690 * had been triggered by is_prefetch fixup_exception would have
691 * handled it.
692 *
693 * 64-bit:
694 *
695 * Hall of shame of CPU/BIOS bugs.
Nick Piggin92181f12009-01-20 04:24:26 +0100696 */
697 if (is_prefetch(regs, error_code, address))
698 return;
699
700 if (is_errata93(regs, address))
701 return;
702
703 /*
704 * Oops. The kernel tried to access some bad page. We'll have to
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100705 * terminate things with extreme prejudice:
Nick Piggin92181f12009-01-20 04:24:26 +0100706 */
Nick Piggin92181f12009-01-20 04:24:26 +0100707 flags = oops_begin();
Nick Piggin92181f12009-01-20 04:24:26 +0100708
709 show_fault_oops(regs, error_code, address);
710
Aaron Tomlina70857e2014-09-12 14:16:18 +0100711 if (task_stack_end_corrupted(tsk))
Prarit Bhargavab0f4c4b2012-01-26 08:55:34 -0500712 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
Ingo Molnar19803072009-01-21 10:39:51 +0100713
Ingo Molnar1cc99542009-02-20 23:07:48 +0100714 tsk->thread.cr2 = address;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530715 tsk->thread.trap_nr = X86_TRAP_PF;
Ingo Molnar1cc99542009-02-20 23:07:48 +0100716 tsk->thread.error_code = error_code;
Nick Piggin92181f12009-01-20 04:24:26 +0100717
Nick Piggin92181f12009-01-20 04:24:26 +0100718 sig = SIGKILL;
719 if (__die("Oops", regs, error_code))
720 sig = 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100721
Nick Piggin92181f12009-01-20 04:24:26 +0100722 /* Executive summary in case the body of the oops scrolled away */
Prarit Bhargavab0f4c4b2012-01-26 08:55:34 -0500723 printk(KERN_DEFAULT "CR2: %016lx\n", address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100724
Nick Piggin92181f12009-01-20 04:24:26 +0100725 oops_end(flags, regs, sig);
Nick Piggin92181f12009-01-20 04:24:26 +0100726}
727
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100728/*
729 * Print out info about fatal segfaults, if the show_unhandled_signals
730 * sysctl is set:
731 */
732static inline void
733show_signal_msg(struct pt_regs *regs, unsigned long error_code,
734 unsigned long address, struct task_struct *tsk)
735{
736 if (!unhandled_signal(tsk, SIGSEGV))
737 return;
738
739 if (!printk_ratelimit())
740 return;
741
Roland Dreiera1a08d12009-07-11 00:10:04 -0700742 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100743 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
744 tsk->comm, task_pid_nr(tsk), address,
745 (void *)regs->ip, (void *)regs->sp, error_code);
746
747 print_vma_addr(KERN_CONT " in ", regs->ip);
748
749 printk(KERN_CONT "\n");
750}
751
752static void
753__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
754 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100755{
756 struct task_struct *tsk = current;
757
758 /* User mode accesses just cause a SIGSEGV */
759 if (error_code & PF_USER) {
760 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100761 * It's possible to have interrupts off here:
Nick Piggin92181f12009-01-20 04:24:26 +0100762 */
763 local_irq_enable();
764
765 /*
766 * Valid to do another page fault here because this one came
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100767 * from user space:
Nick Piggin92181f12009-01-20 04:24:26 +0100768 */
769 if (is_prefetch(regs, error_code, address))
770 return;
771
772 if (is_errata100(regs, address))
773 return;
774
Andy Lutomirski3ae36652011-08-10 11:15:32 -0400775#ifdef CONFIG_X86_64
776 /*
777 * Instruction fetch faults in the vsyscall page might need
778 * emulation.
779 */
780 if (unlikely((error_code & PF_INSTR) &&
Andy Lutomirskif40c3302014-05-05 12:19:36 -0700781 ((address & ~0xfff) == VSYSCALL_ADDR))) {
Andy Lutomirski3ae36652011-08-10 11:15:32 -0400782 if (emulate_vsyscall(regs, address))
783 return;
784 }
785#endif
Kees Cooke575a862013-02-07 09:44:13 -0800786 /* Kernel addresses are always protection faults: */
787 if (address >= TASK_SIZE)
788 error_code |= PF_PROT;
Andy Lutomirski3ae36652011-08-10 11:15:32 -0400789
Kees Cooke575a862013-02-07 09:44:13 -0800790 if (likely(show_unhandled_signals))
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100791 show_signal_msg(regs, error_code, address, tsk);
Nick Piggin92181f12009-01-20 04:24:26 +0100792
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100793 tsk->thread.cr2 = address;
Kees Cooke575a862013-02-07 09:44:13 -0800794 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530795 tsk->thread.trap_nr = X86_TRAP_PF;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100796
Andi Kleenf672b492010-09-27 22:05:55 +0200797 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100798
Nick Piggin92181f12009-01-20 04:24:26 +0100799 return;
800 }
801
802 if (is_f00f_bug(regs, address))
803 return;
804
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800805 no_context(regs, error_code, address, SIGSEGV, si_code);
Nick Piggin92181f12009-01-20 04:24:26 +0100806}
807
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100808static noinline void
809bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
810 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100811{
812 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
813}
814
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100815static void
816__bad_area(struct pt_regs *regs, unsigned long error_code,
817 unsigned long address, int si_code)
Nick Piggin92181f12009-01-20 04:24:26 +0100818{
819 struct mm_struct *mm = current->mm;
820
821 /*
822 * Something tried to access memory that isn't in our memory map..
823 * Fix it, but check if it's kernel or user first..
824 */
825 up_read(&mm->mmap_sem);
826
827 __bad_area_nosemaphore(regs, error_code, address, si_code);
828}
829
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100830static noinline void
831bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100832{
833 __bad_area(regs, error_code, address, SEGV_MAPERR);
834}
835
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100836static noinline void
837bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
838 unsigned long address)
Nick Piggin92181f12009-01-20 04:24:26 +0100839{
840 __bad_area(regs, error_code, address, SEGV_ACCERR);
841}
842
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100843static void
Andi Kleena6e04aa2009-09-16 11:50:09 +0200844do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
845 unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100846{
847 struct task_struct *tsk = current;
Andi Kleena6e04aa2009-09-16 11:50:09 +0200848 int code = BUS_ADRERR;
Nick Piggin92181f12009-01-20 04:24:26 +0100849
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100850 /* Kernel mode? Handle exceptions or die: */
Linus Torvalds96054562010-08-13 09:49:20 -0700851 if (!(error_code & PF_USER)) {
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800852 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
Linus Torvalds96054562010-08-13 09:49:20 -0700853 return;
854 }
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100855
Ingo Molnarcd1b68f2009-02-20 23:39:02 +0100856 /* User-space => ok to do another page fault: */
Nick Piggin92181f12009-01-20 04:24:26 +0100857 if (is_prefetch(regs, error_code, address))
858 return;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100859
860 tsk->thread.cr2 = address;
861 tsk->thread.error_code = error_code;
Srikar Dronamraju51e7dc72012-03-12 14:55:55 +0530862 tsk->thread.trap_nr = X86_TRAP_PF;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100863
Andi Kleena6e04aa2009-09-16 11:50:09 +0200864#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenf672b492010-09-27 22:05:55 +0200865 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
Andi Kleena6e04aa2009-09-16 11:50:09 +0200866 printk(KERN_ERR
867 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
868 tsk->comm, tsk->pid, address);
869 code = BUS_MCEERR_AR;
870 }
871#endif
Andi Kleenf672b492010-09-27 22:05:55 +0200872 force_sig_info_fault(SIGBUS, code, address, tsk, fault);
Nick Piggin92181f12009-01-20 04:24:26 +0100873}
874
Johannes Weiner3a13c4d2013-09-12 15:13:40 -0700875static noinline void
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100876mm_fault_error(struct pt_regs *regs, unsigned long error_code,
877 unsigned long address, unsigned int fault)
Nick Piggin92181f12009-01-20 04:24:26 +0100878{
Johannes Weiner3a13c4d2013-09-12 15:13:40 -0700879 if (fatal_signal_pending(current) && !(error_code & PF_USER)) {
Johannes Weiner3a13c4d2013-09-12 15:13:40 -0700880 no_context(regs, error_code, address, 0, 0);
881 return;
KOSAKI Motohirob80ef102011-05-26 17:12:12 +0900882 }
KOSAKI Motohirob80ef102011-05-26 17:12:12 +0900883
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100884 if (fault & VM_FAULT_OOM) {
Andrey Vaginf8626852011-03-09 15:22:23 -0800885 /* Kernel mode? Handle exceptions or die: */
886 if (!(error_code & PF_USER)) {
Andy Lutomirski4fc34902011-11-07 16:33:40 -0800887 no_context(regs, error_code, address,
888 SIGSEGV, SEGV_MAPERR);
Johannes Weiner3a13c4d2013-09-12 15:13:40 -0700889 return;
Andrey Vaginf8626852011-03-09 15:22:23 -0800890 }
891
David Rientjesc2d23f92012-12-12 13:52:10 -0800892 /*
893 * We ran out of memory, call the OOM killer, and return the
894 * userspace (which will retry the fault, or kill us if we got
895 * oom-killed):
896 */
897 pagefault_out_of_memory();
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100898 } else {
Andi Kleenf672b492010-09-27 22:05:55 +0200899 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
900 VM_FAULT_HWPOISON_LARGE))
Andi Kleena6e04aa2009-09-16 11:50:09 +0200901 do_sigbus(regs, error_code, address, fault);
Linus Torvalds33692f22015-01-29 10:51:32 -0800902 else if (fault & VM_FAULT_SIGSEGV)
903 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100904 else
905 BUG();
906 }
Nick Piggin92181f12009-01-20 04:24:26 +0100907}
908
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100909static int spurious_fault_check(unsigned long error_code, pte_t *pte)
910{
911 if ((error_code & PF_WRITE) && !pte_write(*pte))
912 return 0;
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100913
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100914 if ((error_code & PF_INSTR) && !pte_exec(*pte))
915 return 0;
916
917 return 1;
918}
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920/*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100921 * Handle a spurious fault caused by a stale TLB entry.
922 *
923 * This allows us to lazily refresh the TLB when increasing the
924 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
925 * eagerly is very expensive since that implies doing a full
926 * cross-processor TLB flush, even if no stale TLB entries exist
927 * on other processors.
928 *
David Vrabel31668512014-04-10 18:46:45 +0100929 * Spurious faults may only occur if the TLB contains an entry with
930 * fewer permission than the page table entry. Non-present (P = 0)
931 * and reserved bit (R = 1) faults are never spurious.
932 *
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100933 * There are no security implications to leaving a stale TLB when
934 * increasing the permissions on a page.
David Vrabel31668512014-04-10 18:46:45 +0100935 *
936 * Returns non-zero if a spurious fault was handled, zero otherwise.
937 *
938 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
939 * (Optional Invalidation).
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100940 */
Masami Hiramatsu93266382014-04-17 17:18:14 +0900941static noinline int
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100942spurious_fault(unsigned long error_code, unsigned long address)
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100943{
944 pgd_t *pgd;
945 pud_t *pud;
946 pmd_t *pmd;
947 pte_t *pte;
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500948 int ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100949
David Vrabel31668512014-04-10 18:46:45 +0100950 /*
951 * Only writes to RO or instruction fetches from NX may cause
952 * spurious faults.
953 *
954 * These could be from user or supervisor accesses but the TLB
955 * is only lazily flushed after a kernel mapping protection
956 * change, so user accesses are not expected to cause spurious
957 * faults.
958 */
959 if (error_code != (PF_WRITE | PF_PROT)
960 && error_code != (PF_INSTR | PF_PROT))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100961 return 0;
962
963 pgd = init_mm.pgd + pgd_index(address);
964 if (!pgd_present(*pgd))
965 return 0;
966
967 pud = pud_offset(pgd, address);
968 if (!pud_present(*pud))
969 return 0;
970
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100971 if (pud_large(*pud))
972 return spurious_fault_check(error_code, (pte_t *) pud);
973
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100974 pmd = pmd_offset(pud, address);
975 if (!pmd_present(*pmd))
976 return 0;
977
Thomas Gleixnerd8b57bb2008-02-06 22:39:43 +0100978 if (pmd_large(*pmd))
979 return spurious_fault_check(error_code, (pte_t *) pmd);
980
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100981 pte = pte_offset_kernel(pmd, address);
Andrea Arcangeli954f8572013-02-22 15:11:49 -0800982 if (!pte_present(*pte))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100983 return 0;
984
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500985 ret = spurious_fault_check(error_code, pte);
986 if (!ret)
987 return 0;
988
989 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100990 * Make sure we have permissions in PMD.
991 * If not, then there's a bug in the page tables:
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500992 */
993 ret = spurious_fault_check(error_code, (pte_t *) pmd);
994 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
Ingo Molnar2d4a7162009-02-20 19:56:40 +0100995
Steven Rostedt3c3e5692009-02-19 11:46:36 -0500996 return ret;
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100997}
Masami Hiramatsu93266382014-04-17 17:18:14 +0900998NOKPROBE_SYMBOL(spurious_fault);
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +0100999
Masoud Asgharifard Sharbianiabd4f752007-07-22 11:12:28 +02001000int show_unhandled_signals = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001002static inline int
Michel Lespinasse68da3362010-10-26 14:21:58 -07001003access_error(unsigned long error_code, struct vm_area_struct *vma)
Nick Piggin92181f12009-01-20 04:24:26 +01001004{
Michel Lespinasse68da3362010-10-26 14:21:58 -07001005 if (error_code & PF_WRITE) {
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001006 /* write, present and write, not present: */
Nick Piggin92181f12009-01-20 04:24:26 +01001007 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1008 return 1;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001009 return 0;
Nick Piggin92181f12009-01-20 04:24:26 +01001010 }
1011
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001012 /* read, present: */
1013 if (unlikely(error_code & PF_PROT))
1014 return 1;
1015
1016 /* read, not present: */
1017 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1018 return 1;
1019
Nick Piggin92181f12009-01-20 04:24:26 +01001020 return 0;
1021}
1022
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -08001023static int fault_in_kernel_space(unsigned long address)
1024{
Ingo Molnard9517342009-02-20 23:32:28 +01001025 return address >= TASK_SIZE_MAX;
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -08001026}
1027
H. Peter Anvin40d3cd62012-09-21 12:43:14 -07001028static inline bool smap_violation(int error_code, struct pt_regs *regs)
1029{
H. Peter Anvin4640c7e2014-02-13 07:46:04 -08001030 if (!IS_ENABLED(CONFIG_X86_SMAP))
1031 return false;
1032
1033 if (!static_cpu_has(X86_FEATURE_SMAP))
1034 return false;
1035
H. Peter Anvin40d3cd62012-09-21 12:43:14 -07001036 if (error_code & PF_USER)
1037 return false;
1038
Andy Lutomirskif39b6f02015-03-18 18:33:33 -07001039 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC))
H. Peter Anvin40d3cd62012-09-21 12:43:14 -07001040 return false;
1041
1042 return true;
1043}
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045/*
1046 * This routine handles page faults. It determines the address,
1047 * and the problem, and then passes it off to one of the appropriate
1048 * routines.
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001049 *
1050 * This function must have noinline because both callers
1051 * {,trace_}do_page_fault() have notrace on. Having this an actual function
1052 * guarantees there's a function trace entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 */
Masami Hiramatsu93266382014-04-17 17:18:14 +09001054static noinline void
Jiri Olsa0ac09f92014-02-28 17:05:26 +01001055__do_page_fault(struct pt_regs *regs, unsigned long error_code,
1056 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
Harvey Harrison33cb5242008-01-30 13:32:19 +01001058 struct vm_area_struct *vma;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001059 struct task_struct *tsk;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001060 struct mm_struct *mm;
Linus Torvalds26178ec2014-12-15 15:07:33 -08001061 int fault, major = 0;
Johannes Weiner759496b2013-09-12 15:13:39 -07001062 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Arjan van de Vena9ba9a32006-03-25 16:30:10 +01001064 tsk = current;
1065 mm = tsk->mm;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001066
Vegard Nossumf8561292008-04-04 00:53:23 +02001067 /*
1068 * Detect and handle instructions that would cause a page fault for
1069 * both a tracked kernel page and a userspace page.
1070 */
1071 if (kmemcheck_active(regs))
1072 kmemcheck_hide(regs);
Ingo Molnar5dfaf902009-06-16 10:23:32 +02001073 prefetchw(&mm->mmap_sem);
Vegard Nossumf8561292008-04-04 00:53:23 +02001074
Pekka Paalanen0fd0e3d2008-05-12 21:20:57 +02001075 if (unlikely(kmmio_fault(regs, address)))
Pekka Paalanen86069782008-05-12 21:20:56 +02001076 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 /*
1079 * We fault-in kernel-space virtual memory on-demand. The
1080 * 'reference' page table is init_mm.pgd.
1081 *
1082 * NOTE! We MUST NOT take any locks for this case. We may
1083 * be in an interrupt or a critical region, and should
1084 * only copy the information from the master page table,
1085 * nothing more.
1086 *
1087 * This verifies that the fault happens in kernel space
1088 * (error_code & 4) == 0, and that the fault was not a
Jan Beulich8b1bde92006-01-11 22:42:23 +01001089 * protection error (error_code & 9) == 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 */
Hiroshi Shimamoto0973a062009-02-04 15:24:09 -08001091 if (unlikely(fault_in_kernel_space(address))) {
Vegard Nossumf8561292008-04-04 00:53:23 +02001092 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1093 if (vmalloc_fault(address) >= 0)
1094 return;
1095
1096 if (kmemcheck_fault(regs, address, error_code))
1097 return;
1098 }
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001099
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001100 /* Can handle a stale RO->RW TLB: */
Nick Piggin92181f12009-01-20 04:24:26 +01001101 if (spurious_fault(error_code, address))
Jeremy Fitzhardinge5b727a32008-01-30 13:34:11 +01001102 return;
1103
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001104 /* kprobes don't want to hook the spurious faults: */
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001105 if (kprobes_fault(regs))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001106 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001107 /*
1108 * Don't take the mm semaphore here. If we fixup a prefetch
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001109 * fault we could otherwise deadlock:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001110 */
Nick Piggin92181f12009-01-20 04:24:26 +01001111 bad_area_nosemaphore(regs, error_code, address);
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001112
Nick Piggin92181f12009-01-20 04:24:26 +01001113 return;
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001114 }
1115
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001116 /* kprobes don't want to hook the spurious faults: */
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001117 if (unlikely(kprobes_fault(regs)))
Masami Hiramatsu9be260a2009-02-05 17:12:39 -05001118 return;
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001119
1120 if (unlikely(error_code & PF_RSVD))
1121 pgtable_bad(regs, error_code, address);
1122
H. Peter Anvin4640c7e2014-02-13 07:46:04 -08001123 if (unlikely(smap_violation(error_code, regs))) {
1124 bad_area_nosemaphore(regs, error_code, address);
1125 return;
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001126 }
1127
1128 /*
1129 * If we're in an interrupt, have no user context or are running
David Hildenbrand70ffdb92015-05-11 17:52:11 +02001130 * in a region with pagefaults disabled then we must not take the fault
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001131 */
David Hildenbrand70ffdb92015-05-11 17:52:11 +02001132 if (unlikely(faulthandler_disabled() || !mm)) {
Peter Zijlstrae00b12e2013-10-24 12:52:06 +02001133 bad_area_nosemaphore(regs, error_code, address);
1134 return;
1135 }
1136
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001137 /*
Linus Torvalds891cffb2008-10-12 13:16:12 -07001138 * It's safe to allow irq's after cr2 has been saved and the
1139 * vmalloc fault has been handled.
1140 *
1141 * User-mode registers count as a user access even for any
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001142 * potential system fault or CPU buglet:
Harvey Harrisonf8c2ee22008-01-30 13:34:10 +01001143 */
Andy Lutomirskif39b6f02015-03-18 18:33:33 -07001144 if (user_mode(regs)) {
Linus Torvalds891cffb2008-10-12 13:16:12 -07001145 local_irq_enable();
1146 error_code |= PF_USER;
Johannes Weiner759496b2013-09-12 15:13:39 -07001147 flags |= FAULT_FLAG_USER;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001148 } else {
1149 if (regs->flags & X86_EFLAGS_IF)
1150 local_irq_enable();
1151 }
Jan Beulich8c914cb2006-03-25 16:29:40 +01001152
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +02001153 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
Peter Zijlstra7dd1fcc2009-03-13 12:21:33 +01001154
Johannes Weiner759496b2013-09-12 15:13:39 -07001155 if (error_code & PF_WRITE)
1156 flags |= FAULT_FLAG_WRITE;
1157
Ingo Molnar3a1dfe62008-10-13 17:49:02 +02001158 /*
1159 * When running in the kernel we expect faults to occur only to
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001160 * addresses in user space. All other faults represent errors in
1161 * the kernel and should generate an OOPS. Unfortunately, in the
1162 * case of an erroneous fault occurring in a code path which already
1163 * holds mmap_sem we will deadlock attempting to validate the fault
1164 * against the address space. Luckily the kernel only validly
1165 * references user space from well defined areas of code, which are
1166 * listed in the exceptions table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 *
1168 * As the vast majority of faults will be valid we will only perform
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001169 * the source reference check when there is a possibility of a
1170 * deadlock. Attempt to lock the address space, if we cannot we then
1171 * validate the source. If this is invalid we can skip the address
1172 * space check, thus avoiding the deadlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 */
Nick Piggin92181f12009-01-20 04:24:26 +01001174 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
Andi Kleen66c58152006-01-11 22:44:09 +01001175 if ((error_code & PF_USER) == 0 &&
Nick Piggin92181f12009-01-20 04:24:26 +01001176 !search_exception_tables(regs->ip)) {
1177 bad_area_nosemaphore(regs, error_code, address);
1178 return;
1179 }
Michel Lespinassed065bd82010-10-26 14:21:57 -07001180retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 down_read(&mm->mmap_sem);
Peter Zijlstra01006072009-01-29 16:02:12 +01001182 } else {
1183 /*
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001184 * The above down_read_trylock() might have succeeded in
1185 * which case we'll have missed the might_sleep() from
1186 * down_read():
Peter Zijlstra01006072009-01-29 16:02:12 +01001187 */
1188 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
1190
1191 vma = find_vma(mm, address);
Nick Piggin92181f12009-01-20 04:24:26 +01001192 if (unlikely(!vma)) {
1193 bad_area(regs, error_code, address);
1194 return;
1195 }
1196 if (likely(vma->vm_start <= address))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 goto good_area;
Nick Piggin92181f12009-01-20 04:24:26 +01001198 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1199 bad_area(regs, error_code, address);
1200 return;
1201 }
Harvey Harrison33cb5242008-01-30 13:32:19 +01001202 if (error_code & PF_USER) {
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001203 /*
1204 * Accessing the stack below %sp is always a bug.
1205 * The large cushion allows instructions like enter
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001206 * and pusha to work. ("enter $65535, $31" pushes
Harvey Harrison6f4d3682008-01-30 13:33:13 +01001207 * 32 pointers and then decrements %sp by 65535.)
Chuck Ebbert03fdc2c2006-06-26 13:59:50 +02001208 */
Nick Piggin92181f12009-01-20 04:24:26 +01001209 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1210 bad_area(regs, error_code, address);
1211 return;
1212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Nick Piggin92181f12009-01-20 04:24:26 +01001214 if (unlikely(expand_stack(vma, address))) {
1215 bad_area(regs, error_code, address);
1216 return;
1217 }
1218
1219 /*
1220 * Ok, we have a good vm_area for this memory access, so
1221 * we can handle it..
1222 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223good_area:
Michel Lespinasse68da3362010-10-26 14:21:58 -07001224 if (unlikely(access_error(error_code, vma))) {
Nick Piggin92181f12009-01-20 04:24:26 +01001225 bad_area_access_error(regs, error_code, address);
1226 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 }
1228
1229 /*
1230 * If for any reason at all we couldn't handle the fault,
1231 * make sure we exit gracefully rather than endlessly redo
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001232 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1233 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001235 fault = handle_mm_fault(mm, vma, address, flags);
Linus Torvalds26178ec2014-12-15 15:07:33 -08001236 major |= fault & VM_FAULT_MAJOR;
Ingo Molnar2d4a7162009-02-20 19:56:40 +01001237
Johannes Weiner3a13c4d2013-09-12 15:13:40 -07001238 /*
Linus Torvalds26178ec2014-12-15 15:07:33 -08001239 * If we need to retry the mmap_sem has already been released,
1240 * and if there is a fatal signal pending there is no guarantee
1241 * that we made any progress. Handle this case first.
Johannes Weiner3a13c4d2013-09-12 15:13:40 -07001242 */
Linus Torvalds26178ec2014-12-15 15:07:33 -08001243 if (unlikely(fault & VM_FAULT_RETRY)) {
1244 /* Retry at most once */
1245 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1246 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1247 flags |= FAULT_FLAG_TRIED;
1248 if (!fatal_signal_pending(tsk))
1249 goto retry;
1250 }
Johannes Weiner3a13c4d2013-09-12 15:13:40 -07001251
Linus Torvalds26178ec2014-12-15 15:07:33 -08001252 /* User mode? Just return to handle the fatal exception */
Linus Torvaldscf3c0a12014-12-17 11:52:37 -08001253 if (flags & FAULT_FLAG_USER)
Linus Torvalds26178ec2014-12-15 15:07:33 -08001254 return;
1255
1256 /* Not returning to user mode? Handle exceptions or die: */
1257 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1258 return;
1259 }
1260
1261 up_read(&mm->mmap_sem);
Johannes Weiner3a13c4d2013-09-12 15:13:40 -07001262 if (unlikely(fault & VM_FAULT_ERROR)) {
1263 mm_fault_error(regs, error_code, address, fault);
1264 return;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001265 }
1266
1267 /*
Linus Torvalds26178ec2014-12-15 15:07:33 -08001268 * Major/minor page fault accounting. If any of the events
1269 * returned VM_FAULT_MAJOR, we account it as a major fault.
Michel Lespinassed065bd82010-10-26 14:21:57 -07001270 */
Linus Torvalds26178ec2014-12-15 15:07:33 -08001271 if (major) {
1272 tsk->maj_flt++;
1273 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1274 } else {
1275 tsk->min_flt++;
1276 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
Peter Zijlstraac17dc82009-03-13 12:21:34 +01001277 }
Harvey Harrisond729ab32008-01-30 13:33:23 +01001278
Ingo Molnar8c938f92009-02-20 22:12:18 +01001279 check_v8086_mode(regs, address, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280}
Masami Hiramatsu93266382014-04-17 17:18:14 +09001281NOKPROBE_SYMBOL(__do_page_fault);
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +02001282
Masami Hiramatsu93266382014-04-17 17:18:14 +09001283dotraplinkage void notrace
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +02001284do_page_fault(struct pt_regs *regs, unsigned long error_code)
1285{
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001286 unsigned long address = read_cr2(); /* Get the faulting address */
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +01001287 enum ctx_state prev_state;
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001288
1289 /*
1290 * We must have this function tagged with __kprobes, notrace and call
1291 * read_cr2() before calling anything else. To avoid calling any kind
1292 * of tracing machinery before we've observed the CR2 value.
1293 *
1294 * exception_{enter,exit}() contain all sorts of tracepoints.
1295 */
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +01001296
1297 prev_state = exception_enter();
Jiri Olsa0ac09f92014-02-28 17:05:26 +01001298 __do_page_fault(regs, error_code, address);
Frederic Weisbecker6c1e0252013-02-24 01:19:14 +01001299 exception_exit(prev_state);
Frederic Weisbecker6ba3c972012-07-11 20:26:35 +02001300}
Masami Hiramatsu93266382014-04-17 17:18:14 +09001301NOKPROBE_SYMBOL(do_page_fault);
Seiji Aguchi25c74b12013-10-30 16:37:00 -04001302
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001303#ifdef CONFIG_TRACING
Masami Hiramatsu93266382014-04-17 17:18:14 +09001304static nokprobe_inline void
1305trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1306 unsigned long error_code)
Seiji Aguchid34603b2013-10-30 16:39:03 -04001307{
1308 if (user_mode(regs))
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001309 trace_page_fault_user(address, regs, error_code);
Seiji Aguchid34603b2013-10-30 16:39:03 -04001310 else
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001311 trace_page_fault_kernel(address, regs, error_code);
Seiji Aguchid34603b2013-10-30 16:39:03 -04001312}
1313
Masami Hiramatsu93266382014-04-17 17:18:14 +09001314dotraplinkage void notrace
Seiji Aguchi25c74b12013-10-30 16:37:00 -04001315trace_do_page_fault(struct pt_regs *regs, unsigned long error_code)
1316{
Jiri Olsa0ac09f92014-02-28 17:05:26 +01001317 /*
1318 * The exception_enter and tracepoint processing could
1319 * trigger another page faults (user space callchain
1320 * reading) and destroy the original cr2 value, so read
1321 * the faulting address now.
1322 */
1323 unsigned long address = read_cr2();
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001324 enum ctx_state prev_state;
Seiji Aguchi25c74b12013-10-30 16:37:00 -04001325
1326 prev_state = exception_enter();
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001327 trace_page_fault_entries(address, regs, error_code);
Jiri Olsa0ac09f92014-02-28 17:05:26 +01001328 __do_page_fault(regs, error_code, address);
Seiji Aguchi25c74b12013-10-30 16:37:00 -04001329 exception_exit(prev_state);
1330}
Masami Hiramatsu93266382014-04-17 17:18:14 +09001331NOKPROBE_SYMBOL(trace_do_page_fault);
Peter Zijlstrad4078e22014-03-05 14:07:49 +01001332#endif /* CONFIG_TRACING */