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
22 extern 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 */
28 static int
mapped_kernel_page_is_present(unsigned long address)29 mapped_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
65 void __kprobes
ia64_do_page_fault(unsigned long address,unsigned long isr,struct pt_regs * regs)66 ia64_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 #ifdef CONFIG_VIRTUAL_MEM_MAP
88 /*
89 * If fault is in region 5 and we are in the kernel, we may already
90 * have the mmap_lock (pfn_valid macro is called during mmap). There
91 * is no vma for region 5 addr's anyway, so skip getting the semaphore
92 * and go directly to the exception handling code.
93 */
94
95 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
96 goto bad_area_no_up;
97 #endif
98
99 /*
100 * This is to handle the kprobes on user space access instructions
101 */
102 if (kprobe_page_fault(regs, TRAP_BRKPT))
103 return;
104
105 if (user_mode(regs))
106 flags |= FAULT_FLAG_USER;
107 if (mask & VM_WRITE)
108 flags |= FAULT_FLAG_WRITE;
109
110 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
111 retry:
112 mmap_read_lock(mm);
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 # if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
134 || (1 << VM_EXEC_BIT) != VM_EXEC)
135 # error File is out of sync with <linux/mm.h>. Please update.
136 # endif
137
138 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
139 goto bad_area;
140
141 if ((vma->vm_flags & mask) != mask)
142 goto bad_area;
143
144 /*
145 * If for any reason at all we couldn't handle the fault, make
146 * sure we exit gracefully rather than endlessly redo the
147 * fault.
148 */
149 fault = handle_mm_fault(vma, address, flags, regs);
150
151 if (fault_signal_pending(fault, regs))
152 return;
153
154 if (unlikely(fault & VM_FAULT_ERROR)) {
155 /*
156 * We ran out of memory, or some other thing happened
157 * to us that made us unable to handle the page fault
158 * gracefully.
159 */
160 if (fault & VM_FAULT_OOM) {
161 goto out_of_memory;
162 } else if (fault & VM_FAULT_SIGSEGV) {
163 goto bad_area;
164 } else if (fault & VM_FAULT_SIGBUS) {
165 signal = SIGBUS;
166 goto bad_area;
167 }
168 BUG();
169 }
170
171 if (flags & FAULT_FLAG_ALLOW_RETRY) {
172 if (fault & VM_FAULT_RETRY) {
173 flags |= FAULT_FLAG_TRIED;
174
175 /* No need to mmap_read_unlock(mm) as we would
176 * have already released it in __lock_page_or_retry
177 * in mm/filemap.c.
178 */
179
180 goto retry;
181 }
182 }
183
184 mmap_read_unlock(mm);
185 return;
186
187 check_expansion:
188 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
189 if (!vma)
190 goto bad_area;
191 if (!(vma->vm_flags & VM_GROWSDOWN))
192 goto bad_area;
193 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
194 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
195 goto bad_area;
196 if (expand_stack(vma, address))
197 goto bad_area;
198 } else {
199 vma = prev_vma;
200 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
201 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
202 goto bad_area;
203 /*
204 * Since the register backing store is accessed sequentially,
205 * we disallow growing it by more than a page at a time.
206 */
207 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
208 goto bad_area;
209 if (expand_upwards(vma, address))
210 goto bad_area;
211 }
212 goto good_area;
213
214 bad_area:
215 mmap_read_unlock(mm);
216 #ifdef CONFIG_VIRTUAL_MEM_MAP
217 bad_area_no_up:
218 #endif
219 if ((isr & IA64_ISR_SP)
220 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
221 {
222 /*
223 * This fault was due to a speculative load or lfetch.fault, set the "ed"
224 * bit in the psr to ensure forward progress. (Target register will get a
225 * NaT for ld.s, lfetch will be canceled.)
226 */
227 ia64_psr(regs)->ed = 1;
228 return;
229 }
230 if (user_mode(regs)) {
231 force_sig_fault(signal, code, (void __user *) address,
232 0, __ISR_VALID, isr);
233 return;
234 }
235
236 no_context:
237 if ((isr & IA64_ISR_SP)
238 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
239 {
240 /*
241 * This fault was due to a speculative load or lfetch.fault, set the "ed"
242 * bit in the psr to ensure forward progress. (Target register will get a
243 * NaT for ld.s, lfetch will be canceled.)
244 */
245 ia64_psr(regs)->ed = 1;
246 return;
247 }
248
249 /*
250 * Since we have no vma's for region 5, we might get here even if the address is
251 * valid, due to the VHPT walker inserting a non present translation that becomes
252 * stale. If that happens, the non present fault handler already purged the stale
253 * translation, which fixed the problem. So, we check to see if the translation is
254 * valid, and return if it is.
255 */
256 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
257 return;
258
259 if (ia64_done_with_exception(regs))
260 return;
261
262 /*
263 * Oops. The kernel tried to access some bad page. We'll have to terminate things
264 * with extreme prejudice.
265 */
266 bust_spinlocks(1);
267
268 if (address < PAGE_SIZE)
269 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
270 else
271 printk(KERN_ALERT "Unable to handle kernel paging request at "
272 "virtual address %016lx\n", address);
273 if (die("Oops", regs, isr))
274 regs = NULL;
275 bust_spinlocks(0);
276 if (regs)
277 make_task_dead(SIGKILL);
278 return;
279
280 out_of_memory:
281 mmap_read_unlock(mm);
282 if (!user_mode(regs))
283 goto no_context;
284 pagefault_out_of_memory();
285 }
286