1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (C) 2005-2017 Andes Technology Corporation
4 #include <linux/extable.h>
5 #include <linux/module.h>
6 #include <linux/signal.h>
7 #include <linux/ptrace.h>
9 #include <linux/init.h>
10 #include <linux/hardirq.h>
11 #include <linux/uaccess.h>
13 #include <asm/pgtable.h>
14 #include <asm/tlbflush.h>
16 extern void die(const char *str, struct pt_regs *regs, long err);
19 * This is useful to dump out the page tables associated with
22 void show_pte(struct mm_struct *mm, unsigned long addr)
28 pr_alert("pgd = %p\n", mm->pgd);
29 pgd = pgd_offset(mm, addr);
30 pr_alert("[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
43 pmd = pmd_offset(pgd, addr);
45 pr_alert(", *pmd=%08lx", pmd_val(*pmd));
56 if (IS_ENABLED(CONFIG_HIGHMEM))
59 /* We must not map this if we have highmem enabled */
60 pte = pte_offset_map(pmd, addr);
61 pr_alert(", *pte=%08lx", pte_val(*pte));
69 void do_page_fault(unsigned long entry, unsigned long addr,
70 unsigned int error_code, struct pt_regs *regs)
72 struct task_struct *tsk;
74 struct vm_area_struct *vma;
77 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
78 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
80 error_code = error_code & (ITYPE_mskINST | ITYPE_mskETYPE);
83 si_code = SEGV_MAPERR;
85 * We fault-in kernel-space virtual memory on-demand. The
86 * 'reference' page table is init_mm.pgd.
88 * NOTE! We MUST NOT take any locks for this case. We may
89 * be in an interrupt or a critical region, and should
90 * only copy the information from the master page table,
93 if (addr >= TASK_SIZE) {
95 goto bad_area_nosemaphore;
97 if (addr >= TASK_SIZE && addr < VMALLOC_END
98 && (entry == ENTRY_PTE_NOT_PRESENT))
104 /* Send a signal to the task for handling the unalignment access. */
105 if (entry == ENTRY_GENERAL_EXCPETION
106 && error_code == ETYPE_ALIGNMENT_CHECK) {
108 goto bad_area_nosemaphore;
114 * If we're in an interrupt or have no user
115 * context, we must not take the fault..
117 if (unlikely(faulthandler_disabled() || !mm))
121 * As per x86, we may deadlock here. However, since the kernel only
122 * validly references user space from well defined areas of the code,
123 * we can bug out early if this is from code which shouldn't.
125 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
126 if (!user_mode(regs) &&
127 !search_exception_tables(instruction_pointer(regs)))
130 down_read(&mm->mmap_sem);
133 * The above down_read_trylock() might have succeeded in which
134 * case, we'll have missed the might_sleep() from down_read().
137 if (IS_ENABLED(CONFIG_DEBUG_VM)) {
138 if (!user_mode(regs) &&
139 !search_exception_tables(instruction_pointer(regs)))
144 vma = find_vma(mm, addr);
149 if (vma->vm_start <= addr)
152 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
155 if (unlikely(expand_stack(vma, addr)))
159 * Ok, we have a good vm_area for this memory access, so
164 si_code = SEGV_ACCERR;
166 /* first do some preliminary protection checks */
167 if (entry == ENTRY_PTE_NOT_PRESENT) {
168 if (error_code & ITYPE_mskINST)
171 mask = VM_READ | VM_WRITE;
172 if (vma->vm_flags & VM_WRITE)
173 flags |= FAULT_FLAG_WRITE;
175 } else if (entry == ENTRY_TLB_MISC) {
176 switch (error_code & ITYPE_mskETYPE) {
182 flags |= FAULT_FLAG_WRITE;
189 flags |= FAULT_FLAG_WRITE;
198 if (!(vma->vm_flags & mask))
202 * If for any reason at all we couldn't handle the fault,
203 * make sure we exit gracefully rather than endlessly redo
207 fault = handle_mm_fault(vma, addr, flags);
210 * If we need to retry but a fatal signal is pending, handle the
211 * signal first. We do not need to release the mmap_sem because it
212 * would already be released in __lock_page_or_retry in mm/filemap.c.
214 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
215 if (!user_mode(regs))
220 if (unlikely(fault & VM_FAULT_ERROR)) {
221 if (fault & VM_FAULT_OOM)
223 else if (fault & VM_FAULT_SIGBUS)
230 * Major/minor page fault accounting is only done on the initial
231 * attempt. If we go through a retry, it is extremely likely that the
232 * page will be found in page cache at that point.
234 if (flags & FAULT_FLAG_ALLOW_RETRY) {
235 if (fault & VM_FAULT_MAJOR)
239 if (fault & VM_FAULT_RETRY) {
240 flags &= ~FAULT_FLAG_ALLOW_RETRY;
241 flags |= FAULT_FLAG_TRIED;
243 /* No need to up_read(&mm->mmap_sem) as we would
244 * have already released it in __lock_page_or_retry
251 up_read(&mm->mmap_sem);
255 * Something tried to access memory that isn't in our memory map..
256 * Fix it, but check if it's kernel or user first..
259 up_read(&mm->mmap_sem);
261 bad_area_nosemaphore:
263 /* User mode accesses just cause a SIGSEGV */
265 if (user_mode(regs)) {
266 tsk->thread.address = addr;
267 tsk->thread.error_code = error_code;
268 tsk->thread.trap_no = entry;
269 force_sig_fault(SIGSEGV, si_code, (void __user *)addr, tsk);
275 /* Are we prepared to handle this kernel fault?
277 * (The kernel has valid exception-points in the source
278 * when it acesses user-memory. When it fails in one
279 * of those points, we find it in a table and do a jump
280 * to some fixup code that loads an appropriate error
285 const struct exception_table_entry *entry;
288 search_exception_tables(instruction_pointer(regs))) !=
290 /* Adjust the instruction pointer in the stackframe */
291 instruction_pointer(regs) = entry->fixup;
297 * Oops. The kernel tried to access some bad page. We'll have to
298 * terminate things with extreme prejudice.
302 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
303 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
304 "paging request", addr);
307 die("Oops", regs, error_code);
314 * We ran out of memory, or some other thing happened to us that made
315 * us unable to handle the page fault gracefully.
319 up_read(&mm->mmap_sem);
320 if (!user_mode(regs))
322 pagefault_out_of_memory();
326 up_read(&mm->mmap_sem);
328 /* Kernel mode? Handle exceptions or die */
329 if (!user_mode(regs))
335 tsk->thread.address = addr;
336 tsk->thread.error_code = error_code;
337 tsk->thread.trap_no = entry;
338 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr, tsk);
345 * Synchronize this task's top level page-table
346 * with the 'reference' page table.
348 * Use current_pgd instead of tsk->active_mm->pgd
349 * since the latter might be unavailable if this
350 * code is executed in a misfortunately run irq
351 * (like inside schedule() between switch_mm and
355 unsigned int index = pgd_index(addr);
361 pgd = (pgd_t *) __va(__nds32__mfsr(NDS32_SR_L1_PPTB)) + index;
362 pgd_k = init_mm.pgd + index;
364 if (!pgd_present(*pgd_k))
367 pud = pud_offset(pgd, addr);
368 pud_k = pud_offset(pgd_k, addr);
369 if (!pud_present(*pud_k))
372 pmd = pmd_offset(pud, addr);
373 pmd_k = pmd_offset(pud_k, addr);
374 if (!pmd_present(*pmd_k))
377 if (!pmd_present(*pmd))
378 set_pmd(pmd, *pmd_k);
380 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
383 * Since the vmalloc area is global, we don't
384 * need to copy individual PTE's, it is enough to
385 * copy the pgd pointer into the pte page of the
386 * root task. If that is there, we'll find our pte if
390 /* Make sure the actual PTE exists as well to
391 * catch kernel vmalloc-area accesses to non-mapped
392 * addres. If we don't do this, this will just
393 * silently loop forever.
396 pte_k = pte_offset_kernel(pmd_k, addr);
397 if (!pte_present(*pte_k))