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>
14 #include <asm/pgtable.h>
15 #include <asm/processor.h>
16 #include <asm/system.h>
17 #include <asm/uaccess.h>
19 extern void 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() && kprobes_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);
76 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
78 int signal = SIGSEGV, code = SEGV_MAPERR;
79 struct vm_area_struct *vma, *prev_vma;
80 struct mm_struct *mm = current->mm;
85 /* mmap_sem is performance critical.... */
86 prefetchw(&mm->mmap_sem);
89 * If we're in an interrupt or have no user context, we must not take the fault..
91 if (in_atomic() || !mm)
94 #ifdef CONFIG_VIRTUAL_MEM_MAP
96 * If fault is in region 5 and we are in the kernel, we may already
97 * have the mmap_sem (pfn_valid macro is called during mmap). There
98 * is no vma for region 5 addr's anyway, so skip getting the semaphore
99 * and go directly to the exception handling code.
102 if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
107 * This is to handle the kprobes on user space access instructions
109 if (notify_page_fault(regs, TRAP_BRKPT))
112 down_read(&mm->mmap_sem);
114 vma = find_vma_prev(mm, address, &prev_vma);
115 if (!vma && !prev_vma )
119 * find_vma_prev() returns vma such that address < vma->vm_end or NULL
121 * May find no vma, but could be that the last vm area is the
122 * register backing store that needs to expand upwards, in
123 * this case vma will be null, but prev_vma will ne non-null
125 if (( !vma && prev_vma ) || (address < vma->vm_start) )
126 goto check_expansion;
131 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */
133 # define VM_READ_BIT 0
134 # define VM_WRITE_BIT 1
135 # define VM_EXEC_BIT 2
137 # if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
138 || (1 << VM_EXEC_BIT) != VM_EXEC)
139 # error File is out of sync with <linux/mm.h>. Please update.
142 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
145 mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
146 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
148 if ((vma->vm_flags & mask) != mask)
153 * If for any reason at all we couldn't handle the fault, make
154 * sure we exit gracefully rather than endlessly redo the
157 fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) != 0);
158 if (unlikely(fault & VM_FAULT_ERROR)) {
160 * We ran out of memory, or some other thing happened
161 * to us that made us unable to handle the page fault
164 if (fault & VM_FAULT_OOM) {
166 } else if (fault & VM_FAULT_SIGBUS) {
172 if (fault & VM_FAULT_MAJOR)
176 up_read(&mm->mmap_sem);
180 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
183 if (!(vma->vm_flags & VM_GROWSDOWN))
185 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
186 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
188 if (expand_stack(vma, address))
192 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
193 || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
196 * Since the register backing store is accessed sequentially,
197 * we disallow growing it by more than a page at a time.
199 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
201 if (expand_upwards(vma, address))
207 up_read(&mm->mmap_sem);
208 #ifdef CONFIG_VIRTUAL_MEM_MAP
211 if ((isr & IA64_ISR_SP)
212 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
215 * This fault was due to a speculative load or lfetch.fault, set the "ed"
216 * bit in the psr to ensure forward progress. (Target register will get a
217 * NaT for ld.s, lfetch will be canceled.)
219 ia64_psr(regs)->ed = 1;
222 if (user_mode(regs)) {
223 si.si_signo = signal;
226 si.si_addr = (void __user *) address;
228 si.si_flags = __ISR_VALID;
229 force_sig_info(signal, &si, current);
234 if ((isr & IA64_ISR_SP)
235 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
238 * This fault was due to a speculative load or lfetch.fault, set the "ed"
239 * bit in the psr to ensure forward progress. (Target register will get a
240 * NaT for ld.s, lfetch will be canceled.)
242 ia64_psr(regs)->ed = 1;
247 * Since we have no vma's for region 5, we might get here even if the address is
248 * valid, due to the VHPT walker inserting a non present translation that becomes
249 * stale. If that happens, the non present fault handler already purged the stale
250 * translation, which fixed the problem. So, we check to see if the translation is
251 * valid, and return if it is.
253 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
256 if (ia64_done_with_exception(regs))
260 * Oops. The kernel tried to access some bad page. We'll have to terminate things
261 * with extreme prejudice.
265 if (address < PAGE_SIZE)
266 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
268 printk(KERN_ALERT "Unable to handle kernel paging request at "
269 "virtual address %016lx\n", address);
270 die("Oops", regs, isr);
276 up_read(&mm->mmap_sem);
277 if (is_init(current)) {
279 down_read(&mm->mmap_sem);
282 printk(KERN_CRIT "VM: killing process %s\n", current->comm);