1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * OpenRISC fault.c
4 *
5 * Linux architectural port borrowing liberally from similar works of
6 * others. All original copyrights apply as per the original source
7 * declaration.
8 *
9 * Modifications for the OpenRISC architecture:
10 * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
11 * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
12 */
13
14 #include <linux/mm.h>
15 #include <linux/interrupt.h>
16 #include <linux/extable.h>
17 #include <linux/sched/signal.h>
18 #include <linux/perf_event.h>
19
20 #include <linux/uaccess.h>
21 #include <asm/mmu_context.h>
22 #include <asm/siginfo.h>
23 #include <asm/signal.h>
24
25 #define NUM_TLB_ENTRIES 64
26 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
27
28 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
29 * - also look into include/asm/mmu_context.h
30 */
31 volatile pgd_t *current_pgd[NR_CPUS];
32
33 extern void __noreturn die(char *, struct pt_regs *, long);
34
35 /*
36 * This routine handles page faults. It determines the address,
37 * and the problem, and then passes it off to one of the appropriate
38 * routines.
39 *
40 * If this routine detects a bad access, it returns 1, otherwise it
41 * returns 0.
42 */
43
do_page_fault(struct pt_regs * regs,unsigned long address,unsigned long vector,int write_acc)44 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
45 unsigned long vector, int write_acc)
46 {
47 struct task_struct *tsk;
48 struct mm_struct *mm;
49 struct vm_area_struct *vma;
50 int si_code;
51 vm_fault_t fault;
52 unsigned int flags = FAULT_FLAG_DEFAULT;
53
54 tsk = current;
55
56 /*
57 * We fault-in kernel-space virtual memory on-demand. The
58 * 'reference' page table is init_mm.pgd.
59 *
60 * NOTE! We MUST NOT take any locks for this case. We may
61 * be in an interrupt or a critical region, and should
62 * only copy the information from the master page table,
63 * nothing more.
64 *
65 * NOTE2: This is done so that, when updating the vmalloc
66 * mappings we don't have to walk all processes pgdirs and
67 * add the high mappings all at once. Instead we do it as they
68 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
69 * bit set so sometimes the TLB can use a lingering entry.
70 *
71 * This verifies that the fault happens in kernel space
72 * and that the fault was not a protection error.
73 */
74
75 if (address >= VMALLOC_START &&
76 (vector != 0x300 && vector != 0x400) &&
77 !user_mode(regs))
78 goto vmalloc_fault;
79
80 /* If exceptions were enabled, we can reenable them here */
81 if (user_mode(regs)) {
82 /* Exception was in userspace: reenable interrupts */
83 local_irq_enable();
84 flags |= FAULT_FLAG_USER;
85 } else {
86 /* If exception was in a syscall, then IRQ's may have
87 * been enabled or disabled. If they were enabled,
88 * reenable them.
89 */
90 if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
91 local_irq_enable();
92 }
93
94 mm = tsk->mm;
95 si_code = SEGV_MAPERR;
96
97 /*
98 * If we're in an interrupt or have no user
99 * context, we must not take the fault..
100 */
101
102 if (in_interrupt() || !mm)
103 goto no_context;
104
105 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
106
107 retry:
108 mmap_read_lock(mm);
109 vma = find_vma(mm, address);
110
111 if (!vma)
112 goto bad_area;
113
114 if (vma->vm_start <= address)
115 goto good_area;
116
117 if (!(vma->vm_flags & VM_GROWSDOWN))
118 goto bad_area;
119
120 if (user_mode(regs)) {
121 /*
122 * accessing the stack below usp is always a bug.
123 * we get page-aligned addresses so we can only check
124 * if we're within a page from usp, but that might be
125 * enough to catch brutal errors at least.
126 */
127 if (address + PAGE_SIZE < regs->sp)
128 goto bad_area;
129 }
130 if (expand_stack(vma, address))
131 goto bad_area;
132
133 /*
134 * Ok, we have a good vm_area for this memory access, so
135 * we can handle it..
136 */
137
138 good_area:
139 si_code = SEGV_ACCERR;
140
141 /* first do some preliminary protection checks */
142
143 if (write_acc) {
144 if (!(vma->vm_flags & VM_WRITE))
145 goto bad_area;
146 flags |= FAULT_FLAG_WRITE;
147 } else {
148 /* not present */
149 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
150 goto bad_area;
151 }
152
153 /* are we trying to execute nonexecutable area */
154 if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
155 goto bad_area;
156
157 /*
158 * If for any reason at all we couldn't handle the fault,
159 * make sure we exit gracefully rather than endlessly redo
160 * the fault.
161 */
162
163 fault = handle_mm_fault(vma, address, flags, regs);
164
165 if (fault_signal_pending(fault, regs))
166 return;
167
168 if (unlikely(fault & VM_FAULT_ERROR)) {
169 if (fault & VM_FAULT_OOM)
170 goto out_of_memory;
171 else if (fault & VM_FAULT_SIGSEGV)
172 goto bad_area;
173 else if (fault & VM_FAULT_SIGBUS)
174 goto do_sigbus;
175 BUG();
176 }
177
178 /*RGD modeled on Cris */
179 if (fault & VM_FAULT_RETRY) {
180 flags |= FAULT_FLAG_TRIED;
181
182 /* No need to mmap_read_unlock(mm) as we would
183 * have already released it in __lock_page_or_retry
184 * in mm/filemap.c.
185 */
186
187 goto retry;
188 }
189
190 mmap_read_unlock(mm);
191 return;
192
193 /*
194 * Something tried to access memory that isn't in our memory map..
195 * Fix it, but check if it's kernel or user first..
196 */
197
198 bad_area:
199 mmap_read_unlock(mm);
200
201 bad_area_nosemaphore:
202
203 /* User mode accesses just cause a SIGSEGV */
204
205 if (user_mode(regs)) {
206 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
207 return;
208 }
209
210 no_context:
211
212 /* Are we prepared to handle this kernel fault?
213 *
214 * (The kernel has valid exception-points in the source
215 * when it acesses user-memory. When it fails in one
216 * of those points, we find it in a table and do a jump
217 * to some fixup code that loads an appropriate error
218 * code)
219 */
220
221 {
222 const struct exception_table_entry *entry;
223
224 if ((entry = search_exception_tables(regs->pc)) != NULL) {
225 /* Adjust the instruction pointer in the stackframe */
226 regs->pc = entry->fixup;
227 return;
228 }
229 }
230
231 /*
232 * Oops. The kernel tried to access some bad page. We'll have to
233 * terminate things with extreme prejudice.
234 */
235
236 if ((unsigned long)(address) < PAGE_SIZE)
237 printk(KERN_ALERT
238 "Unable to handle kernel NULL pointer dereference");
239 else
240 printk(KERN_ALERT "Unable to handle kernel access");
241 printk(" at virtual address 0x%08lx\n", address);
242
243 die("Oops", regs, write_acc);
244
245 /*
246 * We ran out of memory, or some other thing happened to us that made
247 * us unable to handle the page fault gracefully.
248 */
249
250 out_of_memory:
251 mmap_read_unlock(mm);
252 if (!user_mode(regs))
253 goto no_context;
254 pagefault_out_of_memory();
255 return;
256
257 do_sigbus:
258 mmap_read_unlock(mm);
259
260 /*
261 * Send a sigbus, regardless of whether we were in kernel
262 * or user mode.
263 */
264 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
265
266 /* Kernel mode? Handle exceptions or die */
267 if (!user_mode(regs))
268 goto no_context;
269 return;
270
271 vmalloc_fault:
272 {
273 /*
274 * Synchronize this task's top level page-table
275 * with the 'reference' page table.
276 *
277 * Use current_pgd instead of tsk->active_mm->pgd
278 * since the latter might be unavailable if this
279 * code is executed in a misfortunately run irq
280 * (like inside schedule() between switch_mm and
281 * switch_to...).
282 */
283
284 int offset = pgd_index(address);
285 pgd_t *pgd, *pgd_k;
286 p4d_t *p4d, *p4d_k;
287 pud_t *pud, *pud_k;
288 pmd_t *pmd, *pmd_k;
289 pte_t *pte_k;
290
291 /*
292 phx_warn("do_page_fault(): vmalloc_fault will not work, "
293 "since current_pgd assign a proper value somewhere\n"
294 "anyhow we don't need this at the moment\n");
295
296 phx_mmu("vmalloc_fault");
297 */
298 pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
299 pgd_k = init_mm.pgd + offset;
300
301 /* Since we're two-level, we don't need to do both
302 * set_pgd and set_pmd (they do the same thing). If
303 * we go three-level at some point, do the right thing
304 * with pgd_present and set_pgd here.
305 *
306 * Also, since the vmalloc area is global, we don't
307 * need to copy individual PTE's, it is enough to
308 * copy the pgd pointer into the pte page of the
309 * root task. If that is there, we'll find our pte if
310 * it exists.
311 */
312
313 p4d = p4d_offset(pgd, address);
314 p4d_k = p4d_offset(pgd_k, address);
315 if (!p4d_present(*p4d_k))
316 goto no_context;
317
318 pud = pud_offset(p4d, address);
319 pud_k = pud_offset(p4d_k, address);
320 if (!pud_present(*pud_k))
321 goto no_context;
322
323 pmd = pmd_offset(pud, address);
324 pmd_k = pmd_offset(pud_k, address);
325
326 if (!pmd_present(*pmd_k))
327 goto bad_area_nosemaphore;
328
329 set_pmd(pmd, *pmd_k);
330
331 /* Make sure the actual PTE exists as well to
332 * catch kernel vmalloc-area accesses to non-mapped
333 * addresses. If we don't do this, this will just
334 * silently loop forever.
335 */
336
337 pte_k = pte_offset_kernel(pmd_k, address);
338 if (!pte_present(*pte_k))
339 goto no_context;
340
341 return;
342 }
343 }
344