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#include <linux/perf_event.h>
18
19#include <asm/processor.h>
20#include <asm/exception.h>
21
22extern int die(char *, struct pt_regs *, long);
23
24/*
25 * 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 p4d_t *p4d;
33 pud_t *pud;
34 pmd_t *pmd;
35 pte_t *ptep, pte;
36
37 pgd = pgd_offset_k(address);
38 if (pgd_none(*pgd) || pgd_bad(*pgd))
39 return 0;
40
41 p4d = p4d_offset(pgd, address);
42 if (p4d_none(*p4d) || p4d_bad(*p4d))
43 return 0;
44
45 pud = pud_offset(p4d, address);
46 if (pud_none(*pud) || pud_bad(*pud))
47 return 0;
48
49 pmd = pmd_offset(pud, address);
50 if (pmd_none(*pmd) || pmd_bad(*pmd))
51 return 0;
52
53 ptep = pte_offset_kernel(pmd, address);
54 if (!ptep)
55 return 0;
56
57 pte = *ptep;
58 return pte_present(pte);
59}
60
61# define VM_READ_BIT 0
62# define VM_WRITE_BIT 1
63# define VM_EXEC_BIT 2
64
65void __kprobes
66ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
67{
68 int signal = SIGSEGV, code = SEGV_MAPERR;
69 struct vm_area_struct *vma, *prev_vma;
70 struct mm_struct *mm = current->mm;
71 unsigned long mask;
72 vm_fault_t fault;
73 unsigned int flags = FAULT_FLAG_DEFAULT;
74
75 mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
76 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
77
78 /* mmap_lock is performance critical.... */
79 prefetchw(&mm->mmap_lock);
80
81 /*
82 * If we're in an interrupt or have no user context, we must not take the fault..
83 */
84 if (faulthandler_disabled() || !mm)
85 goto no_context;
86
87 /*
88 * This is to handle the kprobes on user space access instructions
89 */
90 if (kprobe_page_fault(regs, TRAP_BRKPT))
91 return;
92
93 if (user_mode(regs))
94 flags |= FAULT_FLAG_USER;
95 if (mask & VM_WRITE)
96 flags |= FAULT_FLAG_WRITE;
97
98 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
99retry:
100 mmap_read_lock(mm);
101
102 vma = find_vma_prev(mm, address, &prev_vma);
103 if (!vma && !prev_vma )
104 goto bad_area;
105
106 /*
107 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
108 *
109 * May find no vma, but could be that the last vm area is the
110 * register backing store that needs to expand upwards, in
111 * this case vma will be null, but prev_vma will ne non-null
112 */
113 if (( !vma && prev_vma ) || (address < vma->vm_start) )
114 goto check_expansion;
115
116 good_area:
117 code = SEGV_ACCERR;
118
119 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
120
121# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
122 || (1 << VM_EXEC_BIT) != VM_EXEC)
123# error File is out of sync with <linux/mm.h>. Please update.
124# endif
125
126 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
127 goto bad_area;
128
129 if ((vma->vm_flags & mask) != mask)
130 goto bad_area;
131
132 /*
133 * If for any reason at all we couldn't handle the fault, make
134 * sure we exit gracefully rather than endlessly redo the
135 * fault.
136 */
137 fault = handle_mm_fault(vma, address, flags, regs);
138
139 if (fault_signal_pending(fault, regs))
140 return;
141
142 /* The fault is fully completed (including releasing mmap lock) */
143 if (fault & VM_FAULT_COMPLETED)
144 return;
145
146 if (unlikely(fault & VM_FAULT_ERROR)) {
147 /*
148 * We ran out of memory, or some other thing happened
149 * to us that made us unable to handle the page fault
150 * gracefully.
151 */
152 if (fault & VM_FAULT_OOM) {
153 goto out_of_memory;
154 } else if (fault & VM_FAULT_SIGSEGV) {
155 goto bad_area;
156 } else if (fault & VM_FAULT_SIGBUS) {
157 signal = SIGBUS;
158 goto bad_area;
159 }
160 BUG();
161 }
162
163 if (fault & VM_FAULT_RETRY) {
164 flags |= FAULT_FLAG_TRIED;
165
166 /* No need to mmap_read_unlock(mm) as we would
167 * have already released it in __lock_page_or_retry
168 * in mm/filemap.c.
169 */
170
171 goto retry;
172 }
173
174 mmap_read_unlock(mm);
175 return;
176
177 check_expansion:
178 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
179 if (!vma)
180 goto bad_area;
181 if (!(vma->vm_flags & VM_GROWSDOWN))
182 goto bad_area;
183 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
184 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
185 goto bad_area;
186 if (expand_stack(vma, address))
187 goto bad_area;
188 } else {
189 vma = prev_vma;
190 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
191 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
192 goto bad_area;
193 /*
194 * Since the register backing store is accessed sequentially,
195 * we disallow growing it by more than a page at a time.
196 */
197 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
198 goto bad_area;
199 if (expand_upwards(vma, address))
200 goto bad_area;
201 }
202 goto good_area;
203
204 bad_area:
205 mmap_read_unlock(mm);
206 if ((isr & IA64_ISR_SP)
207 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
208 {
209 /*
210 * This fault was due to a speculative load or lfetch.fault, set the "ed"
211 * bit in the psr to ensure forward progress. (Target register will get a
212 * NaT for ld.s, lfetch will be canceled.)
213 */
214 ia64_psr(regs)->ed = 1;
215 return;
216 }
217 if (user_mode(regs)) {
218 force_sig_fault(signal, code, (void __user *) address,
219 0, __ISR_VALID, isr);
220 return;
221 }
222
223 no_context:
224 if ((isr & IA64_ISR_SP)
225 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
226 {
227 /*
228 * This fault was due to a speculative load or lfetch.fault, set the "ed"
229 * bit in the psr to ensure forward progress. (Target register will get a
230 * NaT for ld.s, lfetch will be canceled.)
231 */
232 ia64_psr(regs)->ed = 1;
233 return;
234 }
235
236 /*
237 * Since we have no vma's for region 5, we might get here even if the address is
238 * valid, due to the VHPT walker inserting a non present translation that becomes
239 * stale. If that happens, the non present fault handler already purged the stale
240 * translation, which fixed the problem. So, we check to see if the translation is
241 * valid, and return if it is.
242 */
243 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
244 return;
245
246 if (ia64_done_with_exception(regs))
247 return;
248
249 /*
250 * Oops. The kernel tried to access some bad page. We'll have to terminate things
251 * with extreme prejudice.
252 */
253 bust_spinlocks(1);
254
255 if (address < PAGE_SIZE)
256 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
257 else
258 printk(KERN_ALERT "Unable to handle kernel paging request at "
259 "virtual address %016lx\n", address);
260 if (die("Oops", regs, isr))
261 regs = NULL;
262 bust_spinlocks(0);
263 if (regs)
264 make_task_dead(SIGKILL);
265 return;
266
267 out_of_memory:
268 mmap_read_unlock(mm);
269 if (!user_mode(regs))
270 goto no_context;
271 pagefault_out_of_memory();
272}
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/system.h>
18#include <asm/uaccess.h>
19
20extern int die(char *, struct pt_regs *, long);
21
22#ifdef CONFIG_KPROBES
23static inline int notify_page_fault(struct pt_regs *regs, int trap)
24{
25 int ret = 0;
26
27 if (!user_mode(regs)) {
28 /* kprobe_running() needs smp_processor_id() */
29 preempt_disable();
30 if (kprobe_running() && kprobe_fault_handler(regs, trap))
31 ret = 1;
32 preempt_enable();
33 }
34
35 return ret;
36}
37#else
38static inline int notify_page_fault(struct pt_regs *regs, int trap)
39{
40 return 0;
41}
42#endif
43
44/*
45 * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
46 * (inside region 5, on ia64) and that page is present.
47 */
48static int
49mapped_kernel_page_is_present (unsigned long address)
50{
51 pgd_t *pgd;
52 pud_t *pud;
53 pmd_t *pmd;
54 pte_t *ptep, pte;
55
56 pgd = pgd_offset_k(address);
57 if (pgd_none(*pgd) || pgd_bad(*pgd))
58 return 0;
59
60 pud = pud_offset(pgd, address);
61 if (pud_none(*pud) || pud_bad(*pud))
62 return 0;
63
64 pmd = pmd_offset(pud, address);
65 if (pmd_none(*pmd) || pmd_bad(*pmd))
66 return 0;
67
68 ptep = pte_offset_kernel(pmd, address);
69 if (!ptep)
70 return 0;
71
72 pte = *ptep;
73 return pte_present(pte);
74}
75
76void __kprobes
77ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
78{
79 int signal = SIGSEGV, code = SEGV_MAPERR;
80 struct vm_area_struct *vma, *prev_vma;
81 struct mm_struct *mm = current->mm;
82 struct siginfo si;
83 unsigned long mask;
84 int fault;
85
86 /* mmap_sem is performance critical.... */
87 prefetchw(&mm->mmap_sem);
88
89 /*
90 * If we're in an interrupt or have no user context, we must not take the fault..
91 */
92 if (in_atomic() || !mm)
93 goto no_context;
94
95#ifdef CONFIG_VIRTUAL_MEM_MAP
96 /*
97 * If fault is in region 5 and we are in the kernel, we may already
98 * have the mmap_sem (pfn_valid macro is called during mmap). There
99 * is no vma for region 5 addr's anyway, so skip getting the semaphore
100 * and go directly to the exception handling code.
101 */
102
103 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
104 goto bad_area_no_up;
105#endif
106
107 /*
108 * This is to handle the kprobes on user space access instructions
109 */
110 if (notify_page_fault(regs, TRAP_BRKPT))
111 return;
112
113 down_read(&mm->mmap_sem);
114
115 vma = find_vma_prev(mm, address, &prev_vma);
116 if (!vma && !prev_vma )
117 goto bad_area;
118
119 /*
120 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
121 *
122 * May find no vma, but could be that the last vm area is the
123 * register backing store that needs to expand upwards, in
124 * this case vma will be null, but prev_vma will ne non-null
125 */
126 if (( !vma && prev_vma ) || (address < vma->vm_start) )
127 goto check_expansion;
128
129 good_area:
130 code = SEGV_ACCERR;
131
132 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
133
134# define VM_READ_BIT 0
135# define VM_WRITE_BIT 1
136# define VM_EXEC_BIT 2
137
138# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
139 || (1 << VM_EXEC_BIT) != VM_EXEC)
140# error File is out of sync with <linux/mm.h>. Please update.
141# endif
142
143 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
144 goto bad_area;
145
146 mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
147 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
148
149 if ((vma->vm_flags & mask) != mask)
150 goto bad_area;
151
152 /*
153 * If for any reason at all we couldn't handle the fault, make
154 * sure we exit gracefully rather than endlessly redo the
155 * fault.
156 */
157 fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0);
158 if (unlikely(fault & VM_FAULT_ERROR)) {
159 /*
160 * We ran out of memory, or some other thing happened
161 * to us that made us unable to handle the page fault
162 * gracefully.
163 */
164 if (fault & VM_FAULT_OOM) {
165 goto out_of_memory;
166 } else if (fault & VM_FAULT_SIGBUS) {
167 signal = SIGBUS;
168 goto bad_area;
169 }
170 BUG();
171 }
172 if (fault & VM_FAULT_MAJOR)
173 current->maj_flt++;
174 else
175 current->min_flt++;
176 up_read(&mm->mmap_sem);
177 return;
178
179 check_expansion:
180 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
181 if (!vma)
182 goto bad_area;
183 if (!(vma->vm_flags & VM_GROWSDOWN))
184 goto bad_area;
185 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
186 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
187 goto bad_area;
188 if (expand_stack(vma, address))
189 goto bad_area;
190 } else {
191 vma = prev_vma;
192 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
193 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
194 goto bad_area;
195 /*
196 * Since the register backing store is accessed sequentially,
197 * we disallow growing it by more than a page at a time.
198 */
199 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
200 goto bad_area;
201 if (expand_upwards(vma, address))
202 goto bad_area;
203 }
204 goto good_area;
205
206 bad_area:
207 up_read(&mm->mmap_sem);
208#ifdef CONFIG_VIRTUAL_MEM_MAP
209 bad_area_no_up:
210#endif
211 if ((isr & IA64_ISR_SP)
212 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
213 {
214 /*
215 * This fault was due to a speculative load or lfetch.fault, set the "ed"
216 * bit in the psr to ensure forward progress. (Target register will get a
217 * NaT for ld.s, lfetch will be canceled.)
218 */
219 ia64_psr(regs)->ed = 1;
220 return;
221 }
222 if (user_mode(regs)) {
223 si.si_signo = signal;
224 si.si_errno = 0;
225 si.si_code = code;
226 si.si_addr = (void __user *) address;
227 si.si_isr = isr;
228 si.si_flags = __ISR_VALID;
229 force_sig_info(signal, &si, current);
230 return;
231 }
232
233 no_context:
234 if ((isr & IA64_ISR_SP)
235 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
236 {
237 /*
238 * This fault was due to a speculative load or lfetch.fault, set the "ed"
239 * bit in the psr to ensure forward progress. (Target register will get a
240 * NaT for ld.s, lfetch will be canceled.)
241 */
242 ia64_psr(regs)->ed = 1;
243 return;
244 }
245
246 /*
247 * Since we have no vma's for region 5, we might get here even if the address is
248 * valid, due to the VHPT walker inserting a non present translation that becomes
249 * stale. If that happens, the non present fault handler already purged the stale
250 * translation, which fixed the problem. So, we check to see if the translation is
251 * valid, and return if it is.
252 */
253 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
254 return;
255
256 if (ia64_done_with_exception(regs))
257 return;
258
259 /*
260 * Oops. The kernel tried to access some bad page. We'll have to terminate things
261 * with extreme prejudice.
262 */
263 bust_spinlocks(1);
264
265 if (address < PAGE_SIZE)
266 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
267 else
268 printk(KERN_ALERT "Unable to handle kernel paging request at "
269 "virtual address %016lx\n", address);
270 if (die("Oops", regs, isr))
271 regs = NULL;
272 bust_spinlocks(0);
273 if (regs)
274 do_exit(SIGKILL);
275 return;
276
277 out_of_memory:
278 up_read(&mm->mmap_sem);
279 if (!user_mode(regs))
280 goto no_context;
281 pagefault_out_of_memory();
282}