2 * MMU fault handling support.
4 * Copyright (C) 1998-2002 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
7 #include <linux/sched.h>
8 #include <linux/kernel.h>
10 #include <linux/interrupt.h>
11 #include <linux/kprobes.h>
12 #include <linux/kdebug.h>
13 #include <linux/prefetch.h>
15 #include <asm/pgtable.h>
16 #include <asm/processor.h>
17 #include <asm/uaccess.h>
19 extern int die(char *, struct pt_regs *, long);
22 static inline int notify_page_fault(struct pt_regs *regs, int trap)
26 if (!user_mode(regs)) {
27 /* kprobe_running() needs smp_processor_id() */
29 if (kprobe_running() && kprobe_fault_handler(regs, trap))
37 static inline int notify_page_fault(struct pt_regs *regs, int trap)
44 * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
45 * (inside region 5, on ia64) and that page is present.
48 mapped_kernel_page_is_present (unsigned long address)
55 pgd = pgd_offset_k(address);
56 if (pgd_none(*pgd) || pgd_bad(*pgd))
59 pud = pud_offset(pgd, address);
60 if (pud_none(*pud) || pud_bad(*pud))
63 pmd = pmd_offset(pud, address);
64 if (pmd_none(*pmd) || pmd_bad(*pmd))
67 ptep = pte_offset_kernel(pmd, address);
72 return pte_present(pte);
75 # define VM_READ_BIT 0
76 # define VM_WRITE_BIT 1
77 # define VM_EXEC_BIT 2
80 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
82 int signal = SIGSEGV, code = SEGV_MAPERR;
83 struct vm_area_struct *vma, *prev_vma;
84 struct mm_struct *mm = current->mm;
88 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
90 mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
91 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
93 /* mmap_sem is performance critical.... */
94 prefetchw(&mm->mmap_sem);
97 * If we're in an interrupt or have no user context, we must not take the fault..
99 if (in_atomic() || !mm)
102 #ifdef CONFIG_VIRTUAL_MEM_MAP
104 * If fault is in region 5 and we are in the kernel, we may already
105 * have the mmap_sem (pfn_valid macro is called during mmap). There
106 * is no vma for region 5 addr's anyway, so skip getting the semaphore
107 * and go directly to the exception handling code.
110 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
115 * This is to handle the kprobes on user space access instructions
117 if (notify_page_fault(regs, TRAP_BRKPT))
121 flags |= FAULT_FLAG_USER;
123 flags |= FAULT_FLAG_WRITE;
125 down_read(&mm->mmap_sem);
127 vma = find_vma_prev(mm, address, &prev_vma);
128 if (!vma && !prev_vma )
132 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
134 * May find no vma, but could be that the last vm area is the
135 * register backing store that needs to expand upwards, in
136 * this case vma will be null, but prev_vma will ne non-null
138 if (( !vma && prev_vma ) || (address < vma->vm_start) )
139 goto check_expansion;
144 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
146 # if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
147 || (1 << VM_EXEC_BIT) != VM_EXEC)
148 # error File is out of sync with <linux/mm.h>. Please update.
151 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
154 if ((vma->vm_flags & mask) != mask)
158 * If for any reason at all we couldn't handle the fault, make
159 * sure we exit gracefully rather than endlessly redo the
162 fault = handle_mm_fault(mm, vma, address, flags);
164 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
167 if (unlikely(fault & VM_FAULT_ERROR)) {
169 * We ran out of memory, or some other thing happened
170 * to us that made us unable to handle the page fault
173 if (fault & VM_FAULT_OOM) {
175 } else if (fault & VM_FAULT_SIGBUS) {
182 if (flags & FAULT_FLAG_ALLOW_RETRY) {
183 if (fault & VM_FAULT_MAJOR)
187 if (fault & VM_FAULT_RETRY) {
188 flags &= ~FAULT_FLAG_ALLOW_RETRY;
189 flags |= FAULT_FLAG_TRIED;
191 /* No need to up_read(&mm->mmap_sem) as we would
192 * have already released it in __lock_page_or_retry
200 up_read(&mm->mmap_sem);
204 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
207 if (!(vma->vm_flags & VM_GROWSDOWN))
209 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
210 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
212 if (expand_stack(vma, address))
216 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
217 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
220 * Since the register backing store is accessed sequentially,
221 * we disallow growing it by more than a page at a time.
223 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
225 if (expand_upwards(vma, address))
231 up_read(&mm->mmap_sem);
232 #ifdef CONFIG_VIRTUAL_MEM_MAP
235 if ((isr & IA64_ISR_SP)
236 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
239 * This fault was due to a speculative load or lfetch.fault, set the "ed"
240 * bit in the psr to ensure forward progress. (Target register will get a
241 * NaT for ld.s, lfetch will be canceled.)
243 ia64_psr(regs)->ed = 1;
246 if (user_mode(regs)) {
247 si.si_signo = signal;
250 si.si_addr = (void __user *) address;
252 si.si_flags = __ISR_VALID;
253 force_sig_info(signal, &si, current);
258 if ((isr & IA64_ISR_SP)
259 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
262 * This fault was due to a speculative load or lfetch.fault, set the "ed"
263 * bit in the psr to ensure forward progress. (Target register will get a
264 * NaT for ld.s, lfetch will be canceled.)
266 ia64_psr(regs)->ed = 1;
271 * Since we have no vma's for region 5, we might get here even if the address is
272 * valid, due to the VHPT walker inserting a non present translation that becomes
273 * stale. If that happens, the non present fault handler already purged the stale
274 * translation, which fixed the problem. So, we check to see if the translation is
275 * valid, and return if it is.
277 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
280 if (ia64_done_with_exception(regs))
284 * Oops. The kernel tried to access some bad page. We'll have to terminate things
285 * with extreme prejudice.
289 if (address < PAGE_SIZE)
290 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
292 printk(KERN_ALERT "Unable to handle kernel paging request at "
293 "virtual address %016lx\n", address);
294 if (die("Oops", regs, isr))
302 up_read(&mm->mmap_sem);
303 if (!user_mode(regs))
305 pagefault_out_of_memory();