Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * MMU fault handling support. | |
4 | * | |
5 | * Copyright (C) 1998-2002 Hewlett-Packard Co | |
6 | * David Mosberger-Tang <davidm@hpl.hp.com> | |
7 | */ | |
3f07c014 | 8 | #include <linux/sched/signal.h> |
1da177e4 LT |
9 | #include <linux/kernel.h> |
10 | #include <linux/mm.h> | |
e7088170 | 11 | #include <linux/extable.h> |
1da177e4 | 12 | #include <linux/interrupt.h> |
1f7ad57b | 13 | #include <linux/kprobes.h> |
1eeb66a1 | 14 | #include <linux/kdebug.h> |
268bb0ce | 15 | #include <linux/prefetch.h> |
70ffdb93 | 16 | #include <linux/uaccess.h> |
1da177e4 LT |
17 | |
18 | #include <asm/pgtable.h> | |
19 | #include <asm/processor.h> | |
82ed1ac9 | 20 | #include <asm/exception.h> |
1da177e4 | 21 | |
620de2f5 | 22 | extern int die(char *, struct pt_regs *, long); |
1da177e4 | 23 | |
ae9a5b85 | 24 | #ifdef CONFIG_KPROBES |
576fe0bd | 25 | static inline int notify_page_fault(struct pt_regs *regs, int trap) |
ae9a5b85 | 26 | { |
576fe0bd | 27 | int ret = 0; |
ae9a5b85 | 28 | |
576fe0bd CH |
29 | if (!user_mode(regs)) { |
30 | /* kprobe_running() needs smp_processor_id() */ | |
31 | preempt_disable(); | |
45e18c22 | 32 | if (kprobe_running() && kprobe_fault_handler(regs, trap)) |
576fe0bd CH |
33 | ret = 1; |
34 | preempt_enable(); | |
35 | } | |
ae9a5b85 | 36 | |
576fe0bd | 37 | return ret; |
ae9a5b85 AK |
38 | } |
39 | #else | |
576fe0bd | 40 | static inline int notify_page_fault(struct pt_regs *regs, int trap) |
ae9a5b85 | 41 | { |
576fe0bd | 42 | return 0; |
ae9a5b85 AK |
43 | } |
44 | #endif | |
45 | ||
1da177e4 LT |
46 | /* |
47 | * Return TRUE if ADDRESS points at a page in the kernel's mapped segment | |
48 | * (inside region 5, on ia64) and that page is present. | |
49 | */ | |
50 | static int | |
51 | mapped_kernel_page_is_present (unsigned long address) | |
52 | { | |
53 | pgd_t *pgd; | |
54 | pud_t *pud; | |
55 | pmd_t *pmd; | |
56 | pte_t *ptep, pte; | |
57 | ||
58 | pgd = pgd_offset_k(address); | |
59 | if (pgd_none(*pgd) || pgd_bad(*pgd)) | |
60 | return 0; | |
61 | ||
62 | pud = pud_offset(pgd, address); | |
63 | if (pud_none(*pud) || pud_bad(*pud)) | |
64 | return 0; | |
65 | ||
66 | pmd = pmd_offset(pud, address); | |
67 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
68 | return 0; | |
69 | ||
70 | ptep = pte_offset_kernel(pmd, address); | |
71 | if (!ptep) | |
72 | return 0; | |
73 | ||
74 | pte = *ptep; | |
75 | return pte_present(pte); | |
76 | } | |
77 | ||
f28fa729 KC |
78 | # define VM_READ_BIT 0 |
79 | # define VM_WRITE_BIT 1 | |
80 | # define VM_EXEC_BIT 2 | |
81 | ||
1f7ad57b | 82 | void __kprobes |
1da177e4 LT |
83 | ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs) |
84 | { | |
85 | int signal = SIGSEGV, code = SEGV_MAPERR; | |
86 | struct vm_area_struct *vma, *prev_vma; | |
87 | struct mm_struct *mm = current->mm; | |
1da177e4 | 88 | unsigned long mask; |
50a7ca3c | 89 | vm_fault_t fault; |
f28fa729 KC |
90 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
91 | ||
92 | mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT) | |
93 | | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT)); | |
94 | ||
0ffe9849 CL |
95 | /* mmap_sem is performance critical.... */ |
96 | prefetchw(&mm->mmap_sem); | |
97 | ||
1da177e4 LT |
98 | /* |
99 | * If we're in an interrupt or have no user context, we must not take the fault.. | |
100 | */ | |
70ffdb93 | 101 | if (faulthandler_disabled() || !mm) |
1da177e4 LT |
102 | goto no_context; |
103 | ||
104 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
105 | /* | |
106 | * If fault is in region 5 and we are in the kernel, we may already | |
107 | * have the mmap_sem (pfn_valid macro is called during mmap). There | |
108 | * is no vma for region 5 addr's anyway, so skip getting the semaphore | |
109 | * and go directly to the exception handling code. | |
110 | */ | |
111 | ||
112 | if ((REGION_NUMBER(address) == 5) && !user_mode(regs)) | |
113 | goto bad_area_no_up; | |
114 | #endif | |
115 | ||
7213b252 AK |
116 | /* |
117 | * This is to handle the kprobes on user space access instructions | |
118 | */ | |
576fe0bd | 119 | if (notify_page_fault(regs, TRAP_BRKPT)) |
7213b252 AK |
120 | return; |
121 | ||
759496ba JW |
122 | if (user_mode(regs)) |
123 | flags |= FAULT_FLAG_USER; | |
124 | if (mask & VM_WRITE) | |
125 | flags |= FAULT_FLAG_WRITE; | |
f28fa729 | 126 | retry: |
1da177e4 LT |
127 | down_read(&mm->mmap_sem); |
128 | ||
129 | vma = find_vma_prev(mm, address, &prev_vma); | |
e8c59c0c | 130 | if (!vma && !prev_vma ) |
1da177e4 LT |
131 | goto bad_area; |
132 | ||
e8c59c0c AB |
133 | /* |
134 | * find_vma_prev() returns vma such that address < vma->vm_end or NULL | |
135 | * | |
136 | * May find no vma, but could be that the last vm area is the | |
137 | * register backing store that needs to expand upwards, in | |
138 | * this case vma will be null, but prev_vma will ne non-null | |
139 | */ | |
140 | if (( !vma && prev_vma ) || (address < vma->vm_start) ) | |
1da177e4 LT |
141 | goto check_expansion; |
142 | ||
143 | good_area: | |
144 | code = SEGV_ACCERR; | |
145 | ||
146 | /* OK, we've got a good vm_area for this memory area. Check the access permissions: */ | |
147 | ||
1da177e4 LT |
148 | # if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \ |
149 | || (1 << VM_EXEC_BIT) != VM_EXEC) | |
150 | # error File is out of sync with <linux/mm.h>. Please update. | |
151 | # endif | |
152 | ||
df67b3da JB |
153 | if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE)))) |
154 | goto bad_area; | |
155 | ||
1da177e4 LT |
156 | if ((vma->vm_flags & mask) != mask) |
157 | goto bad_area; | |
158 | ||
1da177e4 LT |
159 | /* |
160 | * If for any reason at all we couldn't handle the fault, make | |
161 | * sure we exit gracefully rather than endlessly redo the | |
162 | * fault. | |
163 | */ | |
dcddffd4 | 164 | fault = handle_mm_fault(vma, address, flags); |
f28fa729 KC |
165 | |
166 | if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) | |
167 | return; | |
168 | ||
83c54070 | 169 | if (unlikely(fault & VM_FAULT_ERROR)) { |
1da177e4 LT |
170 | /* |
171 | * We ran out of memory, or some other thing happened | |
172 | * to us that made us unable to handle the page fault | |
173 | * gracefully. | |
174 | */ | |
83c54070 NP |
175 | if (fault & VM_FAULT_OOM) { |
176 | goto out_of_memory; | |
33692f27 LT |
177 | } else if (fault & VM_FAULT_SIGSEGV) { |
178 | goto bad_area; | |
83c54070 NP |
179 | } else if (fault & VM_FAULT_SIGBUS) { |
180 | signal = SIGBUS; | |
181 | goto bad_area; | |
182 | } | |
1da177e4 LT |
183 | BUG(); |
184 | } | |
f28fa729 KC |
185 | |
186 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
187 | if (fault & VM_FAULT_MAJOR) | |
188 | current->maj_flt++; | |
189 | else | |
190 | current->min_flt++; | |
191 | if (fault & VM_FAULT_RETRY) { | |
192 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 193 | flags |= FAULT_FLAG_TRIED; |
f28fa729 KC |
194 | |
195 | /* No need to up_read(&mm->mmap_sem) as we would | |
196 | * have already released it in __lock_page_or_retry | |
197 | * in mm/filemap.c. | |
198 | */ | |
199 | ||
200 | goto retry; | |
201 | } | |
202 | } | |
203 | ||
1da177e4 LT |
204 | up_read(&mm->mmap_sem); |
205 | return; | |
206 | ||
207 | check_expansion: | |
208 | if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) { | |
e8c59c0c AB |
209 | if (!vma) |
210 | goto bad_area; | |
1da177e4 LT |
211 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
212 | goto bad_area; | |
213 | if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) | |
214 | || REGION_OFFSET(address) >= RGN_MAP_LIMIT) | |
215 | goto bad_area; | |
216 | if (expand_stack(vma, address)) | |
217 | goto bad_area; | |
218 | } else { | |
219 | vma = prev_vma; | |
220 | if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) | |
221 | || REGION_OFFSET(address) >= RGN_MAP_LIMIT) | |
222 | goto bad_area; | |
46dea3d0 HD |
223 | /* |
224 | * Since the register backing store is accessed sequentially, | |
225 | * we disallow growing it by more than a page at a time. | |
226 | */ | |
227 | if (address > vma->vm_end + PAGE_SIZE - sizeof(long)) | |
228 | goto bad_area; | |
229 | if (expand_upwards(vma, address)) | |
1da177e4 LT |
230 | goto bad_area; |
231 | } | |
232 | goto good_area; | |
233 | ||
234 | bad_area: | |
235 | up_read(&mm->mmap_sem); | |
236 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
237 | bad_area_no_up: | |
238 | #endif | |
239 | if ((isr & IA64_ISR_SP) | |
240 | || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) | |
241 | { | |
242 | /* | |
243 | * This fault was due to a speculative load or lfetch.fault, set the "ed" | |
244 | * bit in the psr to ensure forward progress. (Target register will get a | |
245 | * NaT for ld.s, lfetch will be canceled.) | |
246 | */ | |
247 | ia64_psr(regs)->ed = 1; | |
248 | return; | |
249 | } | |
250 | if (user_mode(regs)) { | |
a618a275 | 251 | force_sig_fault(signal, code, (void __user *) address, |
2e1661d2 | 252 | 0, __ISR_VALID, isr); |
1da177e4 LT |
253 | return; |
254 | } | |
255 | ||
256 | no_context: | |
f0a8d3c9 TL |
257 | if ((isr & IA64_ISR_SP) |
258 | || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) | |
259 | { | |
1da177e4 | 260 | /* |
f0a8d3c9 TL |
261 | * This fault was due to a speculative load or lfetch.fault, set the "ed" |
262 | * bit in the psr to ensure forward progress. (Target register will get a | |
263 | * NaT for ld.s, lfetch will be canceled.) | |
1da177e4 LT |
264 | */ |
265 | ia64_psr(regs)->ed = 1; | |
266 | return; | |
267 | } | |
268 | ||
1da177e4 LT |
269 | /* |
270 | * Since we have no vma's for region 5, we might get here even if the address is | |
271 | * valid, due to the VHPT walker inserting a non present translation that becomes | |
272 | * stale. If that happens, the non present fault handler already purged the stale | |
273 | * translation, which fixed the problem. So, we check to see if the translation is | |
274 | * valid, and return if it is. | |
275 | */ | |
276 | if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address)) | |
277 | return; | |
278 | ||
63028aa7 KU |
279 | if (ia64_done_with_exception(regs)) |
280 | return; | |
281 | ||
1da177e4 LT |
282 | /* |
283 | * Oops. The kernel tried to access some bad page. We'll have to terminate things | |
284 | * with extreme prejudice. | |
285 | */ | |
286 | bust_spinlocks(1); | |
287 | ||
288 | if (address < PAGE_SIZE) | |
289 | printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address); | |
290 | else | |
291 | printk(KERN_ALERT "Unable to handle kernel paging request at " | |
292 | "virtual address %016lx\n", address); | |
620de2f5 JB |
293 | if (die("Oops", regs, isr)) |
294 | regs = NULL; | |
1da177e4 | 295 | bust_spinlocks(0); |
620de2f5 JB |
296 | if (regs) |
297 | do_exit(SIGKILL); | |
1da177e4 LT |
298 | return; |
299 | ||
300 | out_of_memory: | |
301 | up_read(&mm->mmap_sem); | |
0c3b96e4 | 302 | if (!user_mode(regs)) |
303 | goto no_context; | |
304 | pagefault_out_of_memory(); | |
1da177e4 | 305 | } |