Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * MMU fault handling support.
4 *
5 * Copyright (C) 1998-2002 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 */
8#include <linux/sched/signal.h>
9#include <linux/kernel.h>
10#include <linux/mm.h>
11#include <linux/extable.h>
12#include <linux/interrupt.h>
13#include <linux/kprobes.h>
14#include <linux/kdebug.h>
15#include <linux/prefetch.h>
16#include <linux/uaccess.h>
17
18#include <asm/pgtable.h>
19#include <asm/processor.h>
20#include <asm/exception.h>
21
22extern int die(char *, struct pt_regs *, long);
23
24#ifdef CONFIG_KPROBES
25static inline int notify_page_fault(struct pt_regs *regs, int trap)
26{
27 int ret = 0;
28
29 if (!user_mode(regs)) {
30 /* kprobe_running() needs smp_processor_id() */
31 preempt_disable();
32 if (kprobe_running() && kprobe_fault_handler(regs, trap))
33 ret = 1;
34 preempt_enable();
35 }
36
37 return ret;
38}
39#else
40static inline int notify_page_fault(struct pt_regs *regs, int trap)
41{
42 return 0;
43}
44#endif
45
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 */
50static int
51mapped_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
78# define VM_READ_BIT 0
79# define VM_WRITE_BIT 1
80# define VM_EXEC_BIT 2
81
82void __kprobes
83ia64_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;
88 struct siginfo si;
89 unsigned long mask;
90 int fault;
91 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
92
93 mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
94 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
95
96 /* mmap_sem is performance critical.... */
97 prefetchw(&mm->mmap_sem);
98
99 /*
100 * If we're in an interrupt or have no user context, we must not take the fault..
101 */
102 if (faulthandler_disabled() || !mm)
103 goto no_context;
104
105#ifdef CONFIG_VIRTUAL_MEM_MAP
106 /*
107 * If fault is in region 5 and we are in the kernel, we may already
108 * have the mmap_sem (pfn_valid macro is called during mmap). There
109 * is no vma for region 5 addr's anyway, so skip getting the semaphore
110 * and go directly to the exception handling code.
111 */
112
113 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
114 goto bad_area_no_up;
115#endif
116
117 /*
118 * This is to handle the kprobes on user space access instructions
119 */
120 if (notify_page_fault(regs, TRAP_BRKPT))
121 return;
122
123 if (user_mode(regs))
124 flags |= FAULT_FLAG_USER;
125 if (mask & VM_WRITE)
126 flags |= FAULT_FLAG_WRITE;
127retry:
128 down_read(&mm->mmap_sem);
129
130 vma = find_vma_prev(mm, address, &prev_vma);
131 if (!vma && !prev_vma )
132 goto bad_area;
133
134 /*
135 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
136 *
137 * May find no vma, but could be that the last vm area is the
138 * register backing store that needs to expand upwards, in
139 * this case vma will be null, but prev_vma will ne non-null
140 */
141 if (( !vma && prev_vma ) || (address < vma->vm_start) )
142 goto check_expansion;
143
144 good_area:
145 code = SEGV_ACCERR;
146
147 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
148
149# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
150 || (1 << VM_EXEC_BIT) != VM_EXEC)
151# error File is out of sync with <linux/mm.h>. Please update.
152# endif
153
154 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
155 goto bad_area;
156
157 if ((vma->vm_flags & mask) != mask)
158 goto bad_area;
159
160 /*
161 * If for any reason at all we couldn't handle the fault, make
162 * sure we exit gracefully rather than endlessly redo the
163 * fault.
164 */
165 fault = handle_mm_fault(vma, address, flags);
166
167 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
168 return;
169
170 if (unlikely(fault & VM_FAULT_ERROR)) {
171 /*
172 * We ran out of memory, or some other thing happened
173 * to us that made us unable to handle the page fault
174 * gracefully.
175 */
176 if (fault & VM_FAULT_OOM) {
177 goto out_of_memory;
178 } else if (fault & VM_FAULT_SIGSEGV) {
179 goto bad_area;
180 } else if (fault & VM_FAULT_SIGBUS) {
181 signal = SIGBUS;
182 goto bad_area;
183 }
184 BUG();
185 }
186
187 if (flags & FAULT_FLAG_ALLOW_RETRY) {
188 if (fault & VM_FAULT_MAJOR)
189 current->maj_flt++;
190 else
191 current->min_flt++;
192 if (fault & VM_FAULT_RETRY) {
193 flags &= ~FAULT_FLAG_ALLOW_RETRY;
194 flags |= FAULT_FLAG_TRIED;
195
196 /* No need to up_read(&mm->mmap_sem) as we would
197 * have already released it in __lock_page_or_retry
198 * in mm/filemap.c.
199 */
200
201 goto retry;
202 }
203 }
204
205 up_read(&mm->mmap_sem);
206 return;
207
208 check_expansion:
209 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
210 if (!vma)
211 goto bad_area;
212 if (!(vma->vm_flags & VM_GROWSDOWN))
213 goto bad_area;
214 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
215 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
216 goto bad_area;
217 if (expand_stack(vma, address))
218 goto bad_area;
219 } else {
220 vma = prev_vma;
221 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
222 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
223 goto bad_area;
224 /*
225 * Since the register backing store is accessed sequentially,
226 * we disallow growing it by more than a page at a time.
227 */
228 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
229 goto bad_area;
230 if (expand_upwards(vma, address))
231 goto bad_area;
232 }
233 goto good_area;
234
235 bad_area:
236 up_read(&mm->mmap_sem);
237#ifdef CONFIG_VIRTUAL_MEM_MAP
238 bad_area_no_up:
239#endif
240 if ((isr & IA64_ISR_SP)
241 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
242 {
243 /*
244 * This fault was due to a speculative load or lfetch.fault, set the "ed"
245 * bit in the psr to ensure forward progress. (Target register will get a
246 * NaT for ld.s, lfetch will be canceled.)
247 */
248 ia64_psr(regs)->ed = 1;
249 return;
250 }
251 if (user_mode(regs)) {
252 si.si_signo = signal;
253 si.si_errno = 0;
254 si.si_code = code;
255 si.si_addr = (void __user *) address;
256 si.si_isr = isr;
257 si.si_flags = __ISR_VALID;
258 force_sig_info(signal, &si, current);
259 return;
260 }
261
262 no_context:
263 if ((isr & IA64_ISR_SP)
264 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
265 {
266 /*
267 * This fault was due to a speculative load or lfetch.fault, set the "ed"
268 * bit in the psr to ensure forward progress. (Target register will get a
269 * NaT for ld.s, lfetch will be canceled.)
270 */
271 ia64_psr(regs)->ed = 1;
272 return;
273 }
274
275 /*
276 * Since we have no vma's for region 5, we might get here even if the address is
277 * valid, due to the VHPT walker inserting a non present translation that becomes
278 * stale. If that happens, the non present fault handler already purged the stale
279 * translation, which fixed the problem. So, we check to see if the translation is
280 * valid, and return if it is.
281 */
282 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
283 return;
284
285 if (ia64_done_with_exception(regs))
286 return;
287
288 /*
289 * Oops. The kernel tried to access some bad page. We'll have to terminate things
290 * with extreme prejudice.
291 */
292 bust_spinlocks(1);
293
294 if (address < PAGE_SIZE)
295 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
296 else
297 printk(KERN_ALERT "Unable to handle kernel paging request at "
298 "virtual address %016lx\n", address);
299 if (die("Oops", regs, isr))
300 regs = NULL;
301 bust_spinlocks(0);
302 if (regs)
303 do_exit(SIGKILL);
304 return;
305
306 out_of_memory:
307 up_read(&mm->mmap_sem);
308 if (!user_mode(regs))
309 goto no_context;
310 pagefault_out_of_memory();
311}
1/*
2 * MMU fault handling support.
3 *
4 * Copyright (C) 1998-2002 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
6 */
7#include <linux/sched.h>
8#include <linux/kernel.h>
9#include <linux/mm.h>
10#include <linux/interrupt.h>
11#include <linux/kprobes.h>
12#include <linux/kdebug.h>
13#include <linux/prefetch.h>
14
15#include <asm/pgtable.h>
16#include <asm/processor.h>
17#include <asm/uaccess.h>
18
19extern int die(char *, struct pt_regs *, long);
20
21#ifdef CONFIG_KPROBES
22static inline int notify_page_fault(struct pt_regs *regs, int trap)
23{
24 int ret = 0;
25
26 if (!user_mode(regs)) {
27 /* kprobe_running() needs smp_processor_id() */
28 preempt_disable();
29 if (kprobe_running() && kprobe_fault_handler(regs, trap))
30 ret = 1;
31 preempt_enable();
32 }
33
34 return ret;
35}
36#else
37static inline int notify_page_fault(struct pt_regs *regs, int trap)
38{
39 return 0;
40}
41#endif
42
43/*
44 * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
45 * (inside region 5, on ia64) and that page is present.
46 */
47static int
48mapped_kernel_page_is_present (unsigned long address)
49{
50 pgd_t *pgd;
51 pud_t *pud;
52 pmd_t *pmd;
53 pte_t *ptep, pte;
54
55 pgd = pgd_offset_k(address);
56 if (pgd_none(*pgd) || pgd_bad(*pgd))
57 return 0;
58
59 pud = pud_offset(pgd, address);
60 if (pud_none(*pud) || pud_bad(*pud))
61 return 0;
62
63 pmd = pmd_offset(pud, address);
64 if (pmd_none(*pmd) || pmd_bad(*pmd))
65 return 0;
66
67 ptep = pte_offset_kernel(pmd, address);
68 if (!ptep)
69 return 0;
70
71 pte = *ptep;
72 return pte_present(pte);
73}
74
75void __kprobes
76ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
77{
78 int signal = SIGSEGV, code = SEGV_MAPERR;
79 struct vm_area_struct *vma, *prev_vma;
80 struct mm_struct *mm = current->mm;
81 struct siginfo si;
82 unsigned long mask;
83 int fault;
84
85 /* mmap_sem is performance critical.... */
86 prefetchw(&mm->mmap_sem);
87
88 /*
89 * If we're in an interrupt or have no user context, we must not take the fault..
90 */
91 if (in_atomic() || !mm)
92 goto no_context;
93
94#ifdef CONFIG_VIRTUAL_MEM_MAP
95 /*
96 * If fault is in region 5 and we are in the kernel, we may already
97 * have the mmap_sem (pfn_valid macro is called during mmap). There
98 * is no vma for region 5 addr's anyway, so skip getting the semaphore
99 * and go directly to the exception handling code.
100 */
101
102 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
103 goto bad_area_no_up;
104#endif
105
106 /*
107 * This is to handle the kprobes on user space access instructions
108 */
109 if (notify_page_fault(regs, TRAP_BRKPT))
110 return;
111
112 down_read(&mm->mmap_sem);
113
114 vma = find_vma_prev(mm, address, &prev_vma);
115 if (!vma && !prev_vma )
116 goto bad_area;
117
118 /*
119 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
120 *
121 * May find no vma, but could be that the last vm area is the
122 * register backing store that needs to expand upwards, in
123 * this case vma will be null, but prev_vma will ne non-null
124 */
125 if (( !vma && prev_vma ) || (address < vma->vm_start) )
126 goto check_expansion;
127
128 good_area:
129 code = SEGV_ACCERR;
130
131 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
132
133# define VM_READ_BIT 0
134# define VM_WRITE_BIT 1
135# define VM_EXEC_BIT 2
136
137# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
138 || (1 << VM_EXEC_BIT) != VM_EXEC)
139# error File is out of sync with <linux/mm.h>. Please update.
140# endif
141
142 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
143 goto bad_area;
144
145 mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
146 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
147
148 if ((vma->vm_flags & mask) != mask)
149 goto bad_area;
150
151 /*
152 * If for any reason at all we couldn't handle the fault, make
153 * sure we exit gracefully rather than endlessly redo the
154 * fault.
155 */
156 fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0);
157 if (unlikely(fault & VM_FAULT_ERROR)) {
158 /*
159 * We ran out of memory, or some other thing happened
160 * to us that made us unable to handle the page fault
161 * gracefully.
162 */
163 if (fault & VM_FAULT_OOM) {
164 goto out_of_memory;
165 } else if (fault & VM_FAULT_SIGBUS) {
166 signal = SIGBUS;
167 goto bad_area;
168 }
169 BUG();
170 }
171 if (fault & VM_FAULT_MAJOR)
172 current->maj_flt++;
173 else
174 current->min_flt++;
175 up_read(&mm->mmap_sem);
176 return;
177
178 check_expansion:
179 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
180 if (!vma)
181 goto bad_area;
182 if (!(vma->vm_flags & VM_GROWSDOWN))
183 goto bad_area;
184 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
185 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
186 goto bad_area;
187 if (expand_stack(vma, address))
188 goto bad_area;
189 } else {
190 vma = prev_vma;
191 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
192 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
193 goto bad_area;
194 /*
195 * Since the register backing store is accessed sequentially,
196 * we disallow growing it by more than a page at a time.
197 */
198 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
199 goto bad_area;
200 if (expand_upwards(vma, address))
201 goto bad_area;
202 }
203 goto good_area;
204
205 bad_area:
206 up_read(&mm->mmap_sem);
207#ifdef CONFIG_VIRTUAL_MEM_MAP
208 bad_area_no_up:
209#endif
210 if ((isr & IA64_ISR_SP)
211 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
212 {
213 /*
214 * This fault was due to a speculative load or lfetch.fault, set the "ed"
215 * bit in the psr to ensure forward progress. (Target register will get a
216 * NaT for ld.s, lfetch will be canceled.)
217 */
218 ia64_psr(regs)->ed = 1;
219 return;
220 }
221 if (user_mode(regs)) {
222 si.si_signo = signal;
223 si.si_errno = 0;
224 si.si_code = code;
225 si.si_addr = (void __user *) address;
226 si.si_isr = isr;
227 si.si_flags = __ISR_VALID;
228 force_sig_info(signal, &si, current);
229 return;
230 }
231
232 no_context:
233 if ((isr & IA64_ISR_SP)
234 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
235 {
236 /*
237 * This fault was due to a speculative load or lfetch.fault, set the "ed"
238 * bit in the psr to ensure forward progress. (Target register will get a
239 * NaT for ld.s, lfetch will be canceled.)
240 */
241 ia64_psr(regs)->ed = 1;
242 return;
243 }
244
245 /*
246 * Since we have no vma's for region 5, we might get here even if the address is
247 * valid, due to the VHPT walker inserting a non present translation that becomes
248 * stale. If that happens, the non present fault handler already purged the stale
249 * translation, which fixed the problem. So, we check to see if the translation is
250 * valid, and return if it is.
251 */
252 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
253 return;
254
255 if (ia64_done_with_exception(regs))
256 return;
257
258 /*
259 * Oops. The kernel tried to access some bad page. We'll have to terminate things
260 * with extreme prejudice.
261 */
262 bust_spinlocks(1);
263
264 if (address < PAGE_SIZE)
265 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
266 else
267 printk(KERN_ALERT "Unable to handle kernel paging request at "
268 "virtual address %016lx\n", address);
269 if (die("Oops", regs, isr))
270 regs = NULL;
271 bust_spinlocks(0);
272 if (regs)
273 do_exit(SIGKILL);
274 return;
275
276 out_of_memory:
277 up_read(&mm->mmap_sem);
278 if (!user_mode(regs))
279 goto no_context;
280 pagefault_out_of_memory();
281}