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>
13 extern int find_fixup_code(struct pt_regs *);
14 extern void die_if_kernel(const char *, struct pt_regs *, long);
15 extern void show_registers(struct pt_regs *regs);
17 /* debug of low-level TLB reload */
26 /* debug of higher-level faults */
29 /* current active page directory */
31 DEFINE_PER_CPU(pgd_t *, current_pgd);
32 unsigned long cris_signal_return_page;
35 * This routine handles page faults. It determines the address,
36 * and the problem, and then passes it off to one of the appropriate
39 * Notice that the address we're given is aligned to the page the fault
40 * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
44 * bit 0 == 0 means no page found, 1 means protection fault
45 * bit 1 == 0 means read, 1 means write
47 * If this routine detects a bad access, it returns 1, otherwise it
52 do_page_fault(unsigned long address, struct pt_regs *regs,
53 int protection, int writeaccess)
55 struct task_struct *tsk;
57 struct vm_area_struct * vma;
62 "Page fault for %lX on %X at %lX, prot %d write %d\n",
63 address, smp_processor_id(), instruction_pointer(regs),
64 protection, writeaccess));
69 * We fault-in kernel-space virtual memory on-demand. The
70 * 'reference' page table is init_mm.pgd.
72 * NOTE! We MUST NOT take any locks for this case. We may
73 * be in an interrupt or a critical region, and should
74 * only copy the information from the master page table,
77 * NOTE2: This is done so that, when updating the vmalloc
78 * mappings we don't have to walk all processes pgdirs and
79 * add the high mappings all at once. Instead we do it as they
80 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
81 * bit set so sometimes the TLB can use a lingering entry.
83 * This verifies that the fault happens in kernel space
84 * and that the fault was not a protection error (error_code & 1).
87 if (address >= VMALLOC_START &&
92 /* When stack execution is not allowed we store the signal
93 * trampolines in the reserved cris_signal_return_page.
94 * Handle this in the exact same way as vmalloc (we know
95 * that the mapping is there and is valid so no need to
96 * call handle_mm_fault).
98 if (cris_signal_return_page &&
99 address == cris_signal_return_page &&
100 !protection && user_mode(regs))
103 /* we can and should enable interrupts at this point */
107 info.si_code = SEGV_MAPERR;
110 * If we're in an interrupt or "atomic" operation or have no
111 * user context, we must not take the fault.
114 if (in_atomic() || !mm)
117 down_read(&mm->mmap_sem);
118 vma = find_vma(mm, address);
121 if (vma->vm_start <= address)
123 if (!(vma->vm_flags & VM_GROWSDOWN))
125 if (user_mode(regs)) {
127 * accessing the stack below usp is always a bug.
128 * we get page-aligned addresses so we can only check
129 * if we're within a page from usp, but that might be
130 * enough to catch brutal errors at least.
132 if (address + PAGE_SIZE < rdusp())
135 if (expand_stack(vma, address))
139 * Ok, we have a good vm_area for this memory access, so
144 info.si_code = SEGV_ACCERR;
146 /* first do some preliminary protection checks */
148 if (writeaccess == 2){
149 if (!(vma->vm_flags & VM_EXEC))
151 } else if (writeaccess == 1) {
152 if (!(vma->vm_flags & VM_WRITE))
155 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
160 * If for any reason at all we couldn't handle the fault,
161 * make sure we exit gracefully rather than endlessly redo
165 fault = handle_mm_fault(mm, vma, address, (writeaccess & 1) ? FAULT_FLAG_WRITE : 0);
166 if (unlikely(fault & VM_FAULT_ERROR)) {
167 if (fault & VM_FAULT_OOM)
169 else if (fault & VM_FAULT_SIGSEGV)
171 else if (fault & VM_FAULT_SIGBUS)
175 if (fault & VM_FAULT_MAJOR)
180 up_read(&mm->mmap_sem);
184 * Something tried to access memory that isn't in our memory map..
185 * Fix it, but check if it's kernel or user first..
189 up_read(&mm->mmap_sem);
191 bad_area_nosemaphore:
192 DPG(show_registers(regs));
194 /* User mode accesses just cause a SIGSEGV */
196 if (user_mode(regs)) {
197 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
198 "address %08lx at pc %08lx\n",
200 address, instruction_pointer(regs));
202 /* With DPG on, we've already dumped registers above. */
204 show_registers(regs);
206 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
207 DECLARE_WAIT_QUEUE_HEAD(wq);
208 wait_event_interruptible(wq, 0 == 1);
210 info.si_signo = SIGSEGV;
212 /* info.si_code has been set above */
213 info.si_addr = (void *)address;
214 force_sig_info(SIGSEGV, &info, tsk);
221 /* Are we prepared to handle this kernel fault?
223 * (The kernel has valid exception-points in the source
224 * when it accesses user-memory. When it fails in one
225 * of those points, we find it in a table and do a jump
226 * to some fixup code that loads an appropriate error
230 if (find_fixup_code(regs))
234 * Oops. The kernel tried to access some bad page. We'll have to
235 * terminate things with extreme prejudice.
238 if (!oops_in_progress) {
239 oops_in_progress = 1;
240 if ((unsigned long) (address) < PAGE_SIZE)
241 printk(KERN_ALERT "Unable to handle kernel NULL "
242 "pointer dereference");
244 printk(KERN_ALERT "Unable to handle kernel access"
245 " at virtual address %08lx\n", address);
247 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
248 oops_in_progress = 0;
254 * We ran out of memory, or some other thing happened to us that made
255 * us unable to handle the page fault gracefully.
259 up_read(&mm->mmap_sem);
260 if (!user_mode(regs))
262 pagefault_out_of_memory();
266 up_read(&mm->mmap_sem);
269 * Send a sigbus, regardless of whether we were in kernel
272 info.si_signo = SIGBUS;
274 info.si_code = BUS_ADRERR;
275 info.si_addr = (void *)address;
276 force_sig_info(SIGBUS, &info, tsk);
278 /* Kernel mode? Handle exceptions or die */
279 if (!user_mode(regs))
286 * Synchronize this task's top level page-table
287 * with the 'reference' page table.
289 * Use current_pgd instead of tsk->active_mm->pgd
290 * since the latter might be unavailable if this
291 * code is executed in a misfortunately run irq
292 * (like inside schedule() between switch_mm and
296 int offset = pgd_index(address);
302 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
303 pgd_k = init_mm.pgd + offset;
305 /* Since we're two-level, we don't need to do both
306 * set_pgd and set_pmd (they do the same thing). If
307 * we go three-level at some point, do the right thing
308 * with pgd_present and set_pgd here.
310 * Also, since the vmalloc area is global, we don't
311 * need to copy individual PTE's, it is enough to
312 * copy the pgd pointer into the pte page of the
313 * root task. If that is there, we'll find our pte if
317 pud = pud_offset(pgd, address);
318 pud_k = pud_offset(pgd_k, address);
319 if (!pud_present(*pud_k))
322 pmd = pmd_offset(pud, address);
323 pmd_k = pmd_offset(pud_k, address);
325 if (!pmd_present(*pmd_k))
326 goto bad_area_nosemaphore;
328 set_pmd(pmd, *pmd_k);
330 /* Make sure the actual PTE exists as well to
331 * catch kernel vmalloc-area accesses to non-mapped
332 * addresses. If we don't do this, this will just
333 * silently loop forever.
336 pte_k = pte_offset_kernel(pmd_k, address);
337 if (!pte_present(*pte_k))
344 /* Find fixup code. */
346 find_fixup_code(struct pt_regs *regs)
348 const struct exception_table_entry *fixup;
349 /* in case of delay slot fault (v32) */
350 unsigned long ip = (instruction_pointer(regs) & ~0x1);
352 fixup = search_exception_tables(ip);
354 /* Adjust the instruction pointer in the stackframe. */
355 instruction_pointer(regs) = fixup->fixup;