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);
16 /* debug of low-level TLB reload */
25 /* debug of higher-level faults */
28 /* current active page directory */
30 DEFINE_PER_CPU(pgd_t *, current_pgd);
31 unsigned long cris_signal_return_page;
34 * This routine handles page faults. It determines the address,
35 * and the problem, and then passes it off to one of the appropriate
38 * Notice that the address we're given is aligned to the page the fault
39 * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
43 * bit 0 == 0 means no page found, 1 means protection fault
44 * bit 1 == 0 means read, 1 means write
46 * If this routine detects a bad access, it returns 1, otherwise it
51 do_page_fault(unsigned long address, struct pt_regs *regs,
52 int protection, int writeaccess)
54 struct task_struct *tsk;
56 struct vm_area_struct * vma;
61 "Page fault for %lX on %X at %lX, prot %d write %d\n",
62 address, smp_processor_id(), instruction_pointer(regs),
63 protection, writeaccess));
68 * We fault-in kernel-space virtual memory on-demand. The
69 * 'reference' page table is init_mm.pgd.
71 * NOTE! We MUST NOT take any locks for this case. We may
72 * be in an interrupt or a critical region, and should
73 * only copy the information from the master page table,
76 * NOTE2: This is done so that, when updating the vmalloc
77 * mappings we don't have to walk all processes pgdirs and
78 * add the high mappings all at once. Instead we do it as they
79 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
80 * bit set so sometimes the TLB can use a lingering entry.
82 * This verifies that the fault happens in kernel space
83 * and that the fault was not a protection error (error_code & 1).
86 if (address >= VMALLOC_START &&
91 /* When stack execution is not allowed we store the signal
92 * trampolines in the reserved cris_signal_return_page.
93 * Handle this in the exact same way as vmalloc (we know
94 * that the mapping is there and is valid so no need to
95 * call handle_mm_fault).
97 if (cris_signal_return_page &&
98 address == cris_signal_return_page &&
99 !protection && user_mode(regs))
102 /* we can and should enable interrupts at this point */
106 info.si_code = SEGV_MAPERR;
109 * If we're in an interrupt or "atomic" operation or have no
110 * user context, we must not take the fault.
113 if (in_atomic() || !mm)
116 down_read(&mm->mmap_sem);
117 vma = find_vma(mm, address);
120 if (vma->vm_start <= address)
122 if (!(vma->vm_flags & VM_GROWSDOWN))
124 if (user_mode(regs)) {
126 * accessing the stack below usp is always a bug.
127 * we get page-aligned addresses so we can only check
128 * if we're within a page from usp, but that might be
129 * enough to catch brutal errors at least.
131 if (address + PAGE_SIZE < rdusp())
134 if (expand_stack(vma, address))
138 * Ok, we have a good vm_area for this memory access, so
143 info.si_code = SEGV_ACCERR;
145 /* first do some preliminary protection checks */
147 if (writeaccess == 2){
148 if (!(vma->vm_flags & VM_EXEC))
150 } else if (writeaccess == 1) {
151 if (!(vma->vm_flags & VM_WRITE))
154 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
159 * If for any reason at all we couldn't handle the fault,
160 * make sure we exit gracefully rather than endlessly redo
164 fault = handle_mm_fault(mm, vma, address, (writeaccess & 1) ? FAULT_FLAG_WRITE : 0);
165 if (unlikely(fault & VM_FAULT_ERROR)) {
166 if (fault & VM_FAULT_OOM)
168 else if (fault & VM_FAULT_SIGBUS)
172 if (fault & VM_FAULT_MAJOR)
177 up_read(&mm->mmap_sem);
181 * Something tried to access memory that isn't in our memory map..
182 * Fix it, but check if it's kernel or user first..
186 up_read(&mm->mmap_sem);
188 bad_area_nosemaphore:
189 DPG(show_registers(regs));
191 /* User mode accesses just cause a SIGSEGV */
193 if (user_mode(regs)) {
194 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
195 "address %08lx at pc %08lx\n",
197 address, instruction_pointer(regs));
198 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
199 DECLARE_WAIT_QUEUE_HEAD(wq);
200 wait_event_interruptible(wq, 0 == 1);
202 info.si_signo = SIGSEGV;
204 /* info.si_code has been set above */
205 info.si_addr = (void *)address;
206 force_sig_info(SIGSEGV, &info, tsk);
213 /* Are we prepared to handle this kernel fault?
215 * (The kernel has valid exception-points in the source
216 * when it accesses user-memory. When it fails in one
217 * of those points, we find it in a table and do a jump
218 * to some fixup code that loads an appropriate error
222 if (find_fixup_code(regs))
226 * Oops. The kernel tried to access some bad page. We'll have to
227 * terminate things with extreme prejudice.
230 if (!oops_in_progress) {
231 oops_in_progress = 1;
232 if ((unsigned long) (address) < PAGE_SIZE)
233 printk(KERN_ALERT "Unable to handle kernel NULL "
234 "pointer dereference");
236 printk(KERN_ALERT "Unable to handle kernel access"
237 " at virtual address %08lx\n", address);
239 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
240 oops_in_progress = 0;
246 * We ran out of memory, or some other thing happened to us that made
247 * us unable to handle the page fault gracefully.
251 up_read(&mm->mmap_sem);
252 if (!user_mode(regs))
254 pagefault_out_of_memory();
258 up_read(&mm->mmap_sem);
261 * Send a sigbus, regardless of whether we were in kernel
264 info.si_signo = SIGBUS;
266 info.si_code = BUS_ADRERR;
267 info.si_addr = (void *)address;
268 force_sig_info(SIGBUS, &info, tsk);
270 /* Kernel mode? Handle exceptions or die */
271 if (!user_mode(regs))
278 * Synchronize this task's top level page-table
279 * with the 'reference' page table.
281 * Use current_pgd instead of tsk->active_mm->pgd
282 * since the latter might be unavailable if this
283 * code is executed in a misfortunately run irq
284 * (like inside schedule() between switch_mm and
288 int offset = pgd_index(address);
294 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
295 pgd_k = init_mm.pgd + offset;
297 /* Since we're two-level, we don't need to do both
298 * set_pgd and set_pmd (they do the same thing). If
299 * we go three-level at some point, do the right thing
300 * with pgd_present and set_pgd here.
302 * Also, since the vmalloc area is global, we don't
303 * need to copy individual PTE's, it is enough to
304 * copy the pgd pointer into the pte page of the
305 * root task. If that is there, we'll find our pte if
309 pud = pud_offset(pgd, address);
310 pud_k = pud_offset(pgd_k, address);
311 if (!pud_present(*pud_k))
314 pmd = pmd_offset(pud, address);
315 pmd_k = pmd_offset(pud_k, address);
317 if (!pmd_present(*pmd_k))
318 goto bad_area_nosemaphore;
320 set_pmd(pmd, *pmd_k);
322 /* Make sure the actual PTE exists as well to
323 * catch kernel vmalloc-area accesses to non-mapped
324 * addresses. If we don't do this, this will just
325 * silently loop forever.
328 pte_k = pte_offset_kernel(pmd_k, address);
329 if (!pte_present(*pte_k))
336 /* Find fixup code. */
338 find_fixup_code(struct pt_regs *regs)
340 const struct exception_table_entry *fixup;
341 /* in case of delay slot fault (v32) */
342 unsigned long ip = (instruction_pointer(regs) & ~0x1);
344 fixup = search_exception_tables(ip);
346 /* Adjust the instruction pointer in the stackframe. */
347 instruction_pointer(regs) = fixup->fixup;