4 * Copyright (C) 2000-2010 Axis Communications AB
8 #include <linux/interrupt.h>
9 #include <linux/module.h>
10 #include <linux/wait.h>
11 #include <asm/uaccess.h>
12 #include <arch/system.h>
14 extern int find_fixup_code(struct pt_regs *);
15 extern void die_if_kernel(const char *, struct pt_regs *, long);
16 extern void show_registers(struct pt_regs *regs);
18 /* debug of low-level TLB reload */
27 /* debug of higher-level faults */
30 /* current active page directory */
32 DEFINE_PER_CPU(pgd_t *, current_pgd);
33 unsigned long cris_signal_return_page;
36 * This routine handles page faults. It determines the address,
37 * and the problem, and then passes it off to one of the appropriate
40 * Notice that the address we're given is aligned to the page the fault
41 * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
45 * bit 0 == 0 means no page found, 1 means protection fault
46 * bit 1 == 0 means read, 1 means write
48 * If this routine detects a bad access, it returns 1, otherwise it
53 do_page_fault(unsigned long address, struct pt_regs *regs,
54 int protection, int writeaccess)
56 struct task_struct *tsk;
58 struct vm_area_struct * vma;
61 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
64 "Page fault for %lX on %X at %lX, prot %d write %d\n",
65 address, smp_processor_id(), instruction_pointer(regs),
66 protection, writeaccess));
71 * We fault-in kernel-space virtual memory on-demand. The
72 * 'reference' page table is init_mm.pgd.
74 * NOTE! We MUST NOT take any locks for this case. We may
75 * be in an interrupt or a critical region, and should
76 * only copy the information from the master page table,
79 * NOTE2: This is done so that, when updating the vmalloc
80 * mappings we don't have to walk all processes pgdirs and
81 * add the high mappings all at once. Instead we do it as they
82 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
83 * bit set so sometimes the TLB can use a lingering entry.
85 * This verifies that the fault happens in kernel space
86 * and that the fault was not a protection error (error_code & 1).
89 if (address >= VMALLOC_START &&
94 /* When stack execution is not allowed we store the signal
95 * trampolines in the reserved cris_signal_return_page.
96 * Handle this in the exact same way as vmalloc (we know
97 * that the mapping is there and is valid so no need to
98 * call handle_mm_fault).
100 if (cris_signal_return_page &&
101 address == cris_signal_return_page &&
102 !protection && user_mode(regs))
105 /* we can and should enable interrupts at this point */
109 info.si_code = SEGV_MAPERR;
112 * If we're in an interrupt or "atomic" operation or have no
113 * user context, we must not take the fault.
116 if (in_atomic() || !mm)
120 flags |= FAULT_FLAG_USER;
122 down_read(&mm->mmap_sem);
123 vma = find_vma(mm, address);
126 if (vma->vm_start <= address)
128 if (!(vma->vm_flags & VM_GROWSDOWN))
130 if (user_mode(regs)) {
132 * accessing the stack below usp is always a bug.
133 * we get page-aligned addresses so we can only check
134 * if we're within a page from usp, but that might be
135 * enough to catch brutal errors at least.
137 if (address + PAGE_SIZE < rdusp())
140 if (expand_stack(vma, address))
144 * Ok, we have a good vm_area for this memory access, so
149 info.si_code = SEGV_ACCERR;
151 /* first do some preliminary protection checks */
153 if (writeaccess == 2){
154 if (!(vma->vm_flags & VM_EXEC))
156 } else if (writeaccess == 1) {
157 if (!(vma->vm_flags & VM_WRITE))
159 flags |= FAULT_FLAG_WRITE;
161 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
166 * If for any reason at all we couldn't handle the fault,
167 * make sure we exit gracefully rather than endlessly redo
171 fault = handle_mm_fault(mm, vma, address, flags);
173 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
176 if (unlikely(fault & VM_FAULT_ERROR)) {
177 if (fault & VM_FAULT_OOM)
179 else if (fault & VM_FAULT_SIGSEGV)
181 else if (fault & VM_FAULT_SIGBUS)
186 if (flags & FAULT_FLAG_ALLOW_RETRY) {
187 if (fault & VM_FAULT_MAJOR)
191 if (fault & VM_FAULT_RETRY) {
192 flags &= ~FAULT_FLAG_ALLOW_RETRY;
193 flags |= FAULT_FLAG_TRIED;
196 * No need to up_read(&mm->mmap_sem) as we would
197 * have already released it in __lock_page_or_retry
205 up_read(&mm->mmap_sem);
209 * Something tried to access memory that isn't in our memory map..
210 * Fix it, but check if it's kernel or user first..
214 up_read(&mm->mmap_sem);
216 bad_area_nosemaphore:
217 DPG(show_registers(regs));
219 /* User mode accesses just cause a SIGSEGV */
221 if (user_mode(regs)) {
222 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
223 "address %08lx at pc %08lx\n",
225 address, instruction_pointer(regs));
227 /* With DPG on, we've already dumped registers above. */
229 show_registers(regs);
231 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
232 DECLARE_WAIT_QUEUE_HEAD(wq);
233 wait_event_interruptible(wq, 0 == 1);
235 info.si_signo = SIGSEGV;
237 /* info.si_code has been set above */
238 info.si_addr = (void *)address;
239 force_sig_info(SIGSEGV, &info, tsk);
246 /* Are we prepared to handle this kernel fault?
248 * (The kernel has valid exception-points in the source
249 * when it accesses user-memory. When it fails in one
250 * of those points, we find it in a table and do a jump
251 * to some fixup code that loads an appropriate error
255 if (find_fixup_code(regs))
259 * Oops. The kernel tried to access some bad page. We'll have to
260 * terminate things with extreme prejudice.
263 if (!oops_in_progress) {
264 oops_in_progress = 1;
265 if ((unsigned long) (address) < PAGE_SIZE)
266 printk(KERN_ALERT "Unable to handle kernel NULL "
267 "pointer dereference");
269 printk(KERN_ALERT "Unable to handle kernel access"
270 " at virtual address %08lx\n", address);
272 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
273 oops_in_progress = 0;
279 * We ran out of memory, or some other thing happened to us that made
280 * us unable to handle the page fault gracefully.
284 up_read(&mm->mmap_sem);
285 if (!user_mode(regs))
287 pagefault_out_of_memory();
291 up_read(&mm->mmap_sem);
294 * Send a sigbus, regardless of whether we were in kernel
297 info.si_signo = SIGBUS;
299 info.si_code = BUS_ADRERR;
300 info.si_addr = (void *)address;
301 force_sig_info(SIGBUS, &info, tsk);
303 /* Kernel mode? Handle exceptions or die */
304 if (!user_mode(regs))
311 * Synchronize this task's top level page-table
312 * with the 'reference' page table.
314 * Use current_pgd instead of tsk->active_mm->pgd
315 * since the latter might be unavailable if this
316 * code is executed in a misfortunately run irq
317 * (like inside schedule() between switch_mm and
321 int offset = pgd_index(address);
327 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
328 pgd_k = init_mm.pgd + offset;
330 /* Since we're two-level, we don't need to do both
331 * set_pgd and set_pmd (they do the same thing). If
332 * we go three-level at some point, do the right thing
333 * with pgd_present and set_pgd here.
335 * Also, since the vmalloc area is global, we don't
336 * need to copy individual PTE's, it is enough to
337 * copy the pgd pointer into the pte page of the
338 * root task. If that is there, we'll find our pte if
342 pud = pud_offset(pgd, address);
343 pud_k = pud_offset(pgd_k, address);
344 if (!pud_present(*pud_k))
347 pmd = pmd_offset(pud, address);
348 pmd_k = pmd_offset(pud_k, address);
350 if (!pmd_present(*pmd_k))
351 goto bad_area_nosemaphore;
353 set_pmd(pmd, *pmd_k);
355 /* Make sure the actual PTE exists as well to
356 * catch kernel vmalloc-area accesses to non-mapped
357 * addresses. If we don't do this, this will just
358 * silently loop forever.
361 pte_k = pte_offset_kernel(pmd_k, address);
362 if (!pte_present(*pte_k))
369 /* Find fixup code. */
371 find_fixup_code(struct pt_regs *regs)
373 const struct exception_table_entry *fixup;
374 /* in case of delay slot fault (v32) */
375 unsigned long ip = (instruction_pointer(regs) & ~0x1);
377 fixup = search_exception_tables(ip);
379 /* Adjust the instruction pointer in the stackframe. */
380 instruction_pointer(regs) = fixup->fixup;