blob: c2f299fe9e04a88c8bc9675a5295141a4a0f82eb [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * MMU fault handling support.
4 *
5 * Copyright (C) 1998-2002 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 */
Ingo Molnar3f07c012017-02-08 18:51:30 +01008#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/kernel.h>
10#include <linux/mm.h>
Paul Gortmakere7088172016-07-23 14:01:45 -040011#include <linux/extable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/interrupt.h>
Prasanna S Panchamukhi1f7ad572005-09-06 15:19:30 -070013#include <linux/kprobes.h>
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -070014#include <linux/kdebug.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070015#include <linux/prefetch.h>
David Hildenbrand70ffdb92015-05-11 17:52:11 +020016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18#include <asm/pgtable.h>
19#include <asm/processor.h>
Paul Gortmaker82ed1ac92016-09-19 18:28:25 -040020#include <asm/exception.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Jan Beulich620de2f2008-02-04 23:43:03 -080022extern int die(char *, struct pt_regs *, long);
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
26 * (inside region 5, on ia64) and that page is present.
27 */
28static int
29mapped_kernel_page_is_present (unsigned long address)
30{
31 pgd_t *pgd;
32 pud_t *pud;
33 pmd_t *pmd;
34 pte_t *ptep, pte;
35
36 pgd = pgd_offset_k(address);
37 if (pgd_none(*pgd) || pgd_bad(*pgd))
38 return 0;
39
40 pud = pud_offset(pgd, address);
41 if (pud_none(*pud) || pud_bad(*pud))
42 return 0;
43
44 pmd = pmd_offset(pud, address);
45 if (pmd_none(*pmd) || pmd_bad(*pmd))
46 return 0;
47
48 ptep = pte_offset_kernel(pmd, address);
49 if (!ptep)
50 return 0;
51
52 pte = *ptep;
53 return pte_present(pte);
54}
55
Kautuk Consulf28fa722012-06-14 13:11:37 -070056# define VM_READ_BIT 0
57# define VM_WRITE_BIT 1
58# define VM_EXEC_BIT 2
59
Prasanna S Panchamukhi1f7ad572005-09-06 15:19:30 -070060void __kprobes
Linus Torvalds1da177e2005-04-16 15:20:36 -070061ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
62{
63 int signal = SIGSEGV, code = SEGV_MAPERR;
64 struct vm_area_struct *vma, *prev_vma;
65 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 unsigned long mask;
Souptick Joarder50a7ca32018-08-17 15:44:47 -070067 vm_fault_t fault;
Kautuk Consulf28fa722012-06-14 13:11:37 -070068 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
69
70 mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
71 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
72
Christoph Lameter0ffe9842006-03-28 22:54:38 -080073 /* mmap_sem is performance critical.... */
74 prefetchw(&mm->mmap_sem);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 /*
77 * If we're in an interrupt or have no user context, we must not take the fault..
78 */
David Hildenbrand70ffdb92015-05-11 17:52:11 +020079 if (faulthandler_disabled() || !mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 goto no_context;
81
82#ifdef CONFIG_VIRTUAL_MEM_MAP
83 /*
84 * If fault is in region 5 and we are in the kernel, we may already
85 * have the mmap_sem (pfn_valid macro is called during mmap). There
86 * is no vma for region 5 addr's anyway, so skip getting the semaphore
87 * and go directly to the exception handling code.
88 */
89
90 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
91 goto bad_area_no_up;
92#endif
93
Anil S Keshavamurthy7213b252005-06-23 00:09:27 -070094 /*
95 * This is to handle the kprobes on user space access instructions
96 */
Anshuman Khandualb98cca42019-07-16 16:28:00 -070097 if (kprobe_page_fault(regs, TRAP_BRKPT))
Anil S Keshavamurthy7213b252005-06-23 00:09:27 -070098 return;
99
Johannes Weiner759496b2013-09-12 15:13:39 -0700100 if (user_mode(regs))
101 flags |= FAULT_FLAG_USER;
102 if (mask & VM_WRITE)
103 flags |= FAULT_FLAG_WRITE;
Kautuk Consulf28fa722012-06-14 13:11:37 -0700104retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 down_read(&mm->mmap_sem);
106
107 vma = find_vma_prev(mm, address, &prev_vma);
Andrew Burgesse8c59c02007-08-16 10:30:46 -0700108 if (!vma && !prev_vma )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 goto bad_area;
110
Andrew Burgesse8c59c02007-08-16 10:30:46 -0700111 /*
112 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
113 *
114 * May find no vma, but could be that the last vm area is the
115 * register backing store that needs to expand upwards, in
116 * this case vma will be null, but prev_vma will ne non-null
117 */
118 if (( !vma && prev_vma ) || (address < vma->vm_start) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 goto check_expansion;
120
121 good_area:
122 code = SEGV_ACCERR;
123
124 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
127 || (1 << VM_EXEC_BIT) != VM_EXEC)
128# error File is out of sync with <linux/mm.h>. Please update.
129# endif
130
Jason Barondf67b3d2006-09-29 01:58:58 -0700131 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
132 goto bad_area;
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 if ((vma->vm_flags & mask) != mask)
135 goto bad_area;
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 /*
138 * If for any reason at all we couldn't handle the fault, make
139 * sure we exit gracefully rather than endlessly redo the
140 * fault.
141 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -0700142 fault = handle_mm_fault(vma, address, flags);
Kautuk Consulf28fa722012-06-14 13:11:37 -0700143
144 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
145 return;
146
Nick Piggin83c54072007-07-19 01:47:05 -0700147 if (unlikely(fault & VM_FAULT_ERROR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 /*
149 * We ran out of memory, or some other thing happened
150 * to us that made us unable to handle the page fault
151 * gracefully.
152 */
Nick Piggin83c54072007-07-19 01:47:05 -0700153 if (fault & VM_FAULT_OOM) {
154 goto out_of_memory;
Linus Torvalds33692f22015-01-29 10:51:32 -0800155 } else if (fault & VM_FAULT_SIGSEGV) {
156 goto bad_area;
Nick Piggin83c54072007-07-19 01:47:05 -0700157 } else if (fault & VM_FAULT_SIGBUS) {
158 signal = SIGBUS;
159 goto bad_area;
160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 BUG();
162 }
Kautuk Consulf28fa722012-06-14 13:11:37 -0700163
164 if (flags & FAULT_FLAG_ALLOW_RETRY) {
165 if (fault & VM_FAULT_MAJOR)
166 current->maj_flt++;
167 else
168 current->min_flt++;
169 if (fault & VM_FAULT_RETRY) {
170 flags &= ~FAULT_FLAG_ALLOW_RETRY;
Shaohua Li45cac652012-10-08 16:32:19 -0700171 flags |= FAULT_FLAG_TRIED;
Kautuk Consulf28fa722012-06-14 13:11:37 -0700172
173 /* No need to up_read(&mm->mmap_sem) as we would
174 * have already released it in __lock_page_or_retry
175 * in mm/filemap.c.
176 */
177
178 goto retry;
179 }
180 }
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 up_read(&mm->mmap_sem);
183 return;
184
185 check_expansion:
186 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
Andrew Burgesse8c59c02007-08-16 10:30:46 -0700187 if (!vma)
188 goto bad_area;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 if (!(vma->vm_flags & VM_GROWSDOWN))
190 goto bad_area;
191 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
192 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
193 goto bad_area;
194 if (expand_stack(vma, address))
195 goto bad_area;
196 } else {
197 vma = prev_vma;
198 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
199 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
200 goto bad_area;
Hugh Dickins46dea3d2005-10-29 18:16:20 -0700201 /*
202 * Since the register backing store is accessed sequentially,
203 * we disallow growing it by more than a page at a time.
204 */
205 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
206 goto bad_area;
207 if (expand_upwards(vma, address))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 goto bad_area;
209 }
210 goto good_area;
211
212 bad_area:
213 up_read(&mm->mmap_sem);
214#ifdef CONFIG_VIRTUAL_MEM_MAP
215 bad_area_no_up:
216#endif
217 if ((isr & IA64_ISR_SP)
218 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
219 {
220 /*
221 * This fault was due to a speculative load or lfetch.fault, set the "ed"
222 * bit in the psr to ensure forward progress. (Target register will get a
223 * NaT for ld.s, lfetch will be canceled.)
224 */
225 ia64_psr(regs)->ed = 1;
226 return;
227 }
228 if (user_mode(regs)) {
Eric W. Biedermana618a272018-04-17 17:39:29 -0500229 force_sig_fault(signal, code, (void __user *) address,
Eric W. Biederman2e1661d22019-05-23 11:04:24 -0500230 0, __ISR_VALID, isr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return;
232 }
233
234 no_context:
Tony Luckf0a8d3c2005-04-25 13:22:44 -0700235 if ((isr & IA64_ISR_SP)
236 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
237 {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 /*
Tony Luckf0a8d3c2005-04-25 13:22:44 -0700239 * This fault was due to a speculative load or lfetch.fault, set the "ed"
240 * bit in the psr to ensure forward progress. (Target register will get a
241 * NaT for ld.s, lfetch will be canceled.)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
243 ia64_psr(regs)->ed = 1;
244 return;
245 }
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 /*
248 * Since we have no vma's for region 5, we might get here even if the address is
249 * valid, due to the VHPT walker inserting a non present translation that becomes
250 * stale. If that happens, the non present fault handler already purged the stale
251 * translation, which fixed the problem. So, we check to see if the translation is
252 * valid, and return if it is.
253 */
254 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
255 return;
256
Kiyoshi Ueda63028aa2005-08-24 18:03:43 -0400257 if (ia64_done_with_exception(regs))
258 return;
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 /*
261 * Oops. The kernel tried to access some bad page. We'll have to terminate things
262 * with extreme prejudice.
263 */
264 bust_spinlocks(1);
265
266 if (address < PAGE_SIZE)
267 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
268 else
269 printk(KERN_ALERT "Unable to handle kernel paging request at "
270 "virtual address %016lx\n", address);
Jan Beulich620de2f2008-02-04 23:43:03 -0800271 if (die("Oops", regs, isr))
272 regs = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 bust_spinlocks(0);
Jan Beulich620de2f2008-02-04 23:43:03 -0800274 if (regs)
275 do_exit(SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return;
277
278 out_of_memory:
279 up_read(&mm->mmap_sem);
npiggin@suse.de0c3b96e2010-05-07 14:34:33 -0700280 if (!user_mode(regs))
281 goto no_context;
282 pagefault_out_of_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}