2 * Page fault handler for SH with an MMU.
4 * Copyright (C) 1999 Niibe Yutaka
5 * Copyright (C) 2003 - 2012 Paul Mundt
7 * Based on linux/arch/i386/mm/fault.c:
8 * Copyright (C) 1995 Linus Torvalds
10 * This file is subject to the terms and conditions of the GNU General Public
11 * License. See the file "COPYING" in the main directory of this archive
14 #include <linux/kernel.h>
16 #include <linux/hardirq.h>
17 #include <linux/kprobes.h>
18 #include <linux/perf_event.h>
19 #include <linux/kdebug.h>
20 #include <asm/io_trapped.h>
21 #include <asm/mmu_context.h>
22 #include <asm/tlbflush.h>
23 #include <asm/traps.h>
25 static inline int notify_page_fault(struct pt_regs *regs, int trap)
29 if (kprobes_built_in() && !user_mode(regs)) {
31 if (kprobe_running() && kprobe_fault_handler(regs, trap))
40 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
41 struct task_struct *tsk)
45 info.si_signo = si_signo;
47 info.si_code = si_code;
48 info.si_addr = (void __user *)address;
50 force_sig_info(si_signo, &info, tsk);
54 * This is useful to dump out the page tables associated with
57 static void show_pte(struct mm_struct *mm, unsigned long addr)
70 printk(KERN_ALERT "pgd = %p\n", pgd);
71 pgd += pgd_index(addr);
72 printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
73 (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
88 pud = pud_offset(pgd, addr);
89 if (PTRS_PER_PUD != 1)
90 printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
101 pmd = pmd_offset(pud, addr);
102 if (PTRS_PER_PMD != 1)
103 printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
114 /* We must not map this if we have highmem enabled */
115 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
118 pte = pte_offset_kernel(pmd, addr);
119 printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
126 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
128 unsigned index = pgd_index(address);
134 pgd_k = init_mm.pgd + index;
136 if (!pgd_present(*pgd_k))
139 pud = pud_offset(pgd, address);
140 pud_k = pud_offset(pgd_k, address);
141 if (!pud_present(*pud_k))
144 if (!pud_present(*pud))
145 set_pud(pud, *pud_k);
147 pmd = pmd_offset(pud, address);
148 pmd_k = pmd_offset(pud_k, address);
149 if (!pmd_present(*pmd_k))
152 if (!pmd_present(*pmd))
153 set_pmd(pmd, *pmd_k);
156 * The page tables are fully synchronised so there must
157 * be another reason for the fault. Return NULL here to
158 * signal that we have not taken care of the fault.
160 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
167 #ifdef CONFIG_SH_STORE_QUEUES
168 #define __FAULT_ADDR_LIMIT P3_ADDR_MAX
170 #define __FAULT_ADDR_LIMIT VMALLOC_END
174 * Handle a fault on the vmalloc or module mapping area
176 static noinline int vmalloc_fault(unsigned long address)
182 /* Make sure we are in vmalloc/module/P3 area: */
183 if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
187 * Synchronize this task's top level page-table
188 * with the 'reference' page table.
190 * Do _not_ use "current" here. We might be inside
191 * an interrupt in the middle of a task switch..
194 pmd_k = vmalloc_sync_one(pgd_k, address);
198 pte_k = pte_offset_kernel(pmd_k, address);
199 if (!pte_present(*pte_k))
206 show_fault_oops(struct pt_regs *regs, unsigned long address)
208 if (!oops_may_print())
211 printk(KERN_ALERT "BUG: unable to handle kernel ");
212 if (address < PAGE_SIZE)
213 printk(KERN_CONT "NULL pointer dereference");
215 printk(KERN_CONT "paging request");
217 printk(KERN_CONT " at %08lx\n", address);
218 printk(KERN_ALERT "PC:");
219 printk_address(regs->pc, 1);
221 show_pte(NULL, address);
225 no_context(struct pt_regs *regs, unsigned long error_code,
226 unsigned long address)
228 /* Are we prepared to handle this kernel fault? */
229 if (fixup_exception(regs))
232 if (handle_trapped_io(regs, address))
236 * Oops. The kernel tried to access some bad page. We'll have to
237 * terminate things with extreme prejudice.
241 show_fault_oops(regs, address);
243 die("Oops", regs, error_code);
249 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
250 unsigned long address, int si_code)
252 struct task_struct *tsk = current;
254 /* User mode accesses just cause a SIGSEGV */
255 if (user_mode(regs)) {
257 * It's possible to have interrupts off here:
261 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
266 no_context(regs, error_code, address);
270 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
271 unsigned long address)
273 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
277 __bad_area(struct pt_regs *regs, unsigned long error_code,
278 unsigned long address, int si_code)
280 struct mm_struct *mm = current->mm;
283 * Something tried to access memory that isn't in our memory map..
284 * Fix it, but check if it's kernel or user first..
286 up_read(&mm->mmap_sem);
288 __bad_area_nosemaphore(regs, error_code, address, si_code);
292 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
294 __bad_area(regs, error_code, address, SEGV_MAPERR);
298 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
299 unsigned long address)
301 __bad_area(regs, error_code, address, SEGV_ACCERR);
304 static void out_of_memory(void)
307 * We ran out of memory, call the OOM killer, and return the userspace
308 * (which will retry the fault, or kill us if we got oom-killed):
310 up_read(¤t->mm->mmap_sem);
312 pagefault_out_of_memory();
316 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
318 struct task_struct *tsk = current;
319 struct mm_struct *mm = tsk->mm;
321 up_read(&mm->mmap_sem);
323 /* Kernel mode? Handle exceptions or die: */
324 if (!user_mode(regs))
325 no_context(regs, error_code, address);
327 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
331 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
332 unsigned long address, unsigned int fault)
335 * Pagefault was interrupted by SIGKILL. We have no reason to
336 * continue pagefault.
338 if (fatal_signal_pending(current)) {
339 if (!(fault & VM_FAULT_RETRY))
340 up_read(¤t->mm->mmap_sem);
341 if (!user_mode(regs))
342 no_context(regs, error_code, address);
346 if (!(fault & VM_FAULT_ERROR))
349 if (fault & VM_FAULT_OOM) {
350 /* Kernel mode? Handle exceptions or die: */
351 if (!user_mode(regs)) {
352 up_read(¤t->mm->mmap_sem);
353 no_context(regs, error_code, address);
359 if (fault & VM_FAULT_SIGBUS)
360 do_sigbus(regs, error_code, address);
368 static inline int access_error(int error_code, struct vm_area_struct *vma)
370 if (error_code & FAULT_CODE_WRITE) {
371 /* write, present and write, not present: */
372 if (unlikely(!(vma->vm_flags & VM_WRITE)))
377 /* ITLB miss on NX page */
378 if (unlikely((error_code & FAULT_CODE_ITLB) &&
379 !(vma->vm_flags & VM_EXEC)))
382 /* read, not present: */
383 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
389 static int fault_in_kernel_space(unsigned long address)
391 return address >= TASK_SIZE;
395 * This routine handles page faults. It determines the address,
396 * and the problem, and then passes it off to one of the appropriate
399 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
400 unsigned long error_code,
401 unsigned long address)
404 struct task_struct *tsk;
405 struct mm_struct *mm;
406 struct vm_area_struct * vma;
408 int write = error_code & FAULT_CODE_WRITE;
409 unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
410 (write ? FAULT_FLAG_WRITE : 0));
414 vec = lookup_exception_vector();
417 * We fault-in kernel-space virtual memory on-demand. The
418 * 'reference' page table is init_mm.pgd.
420 * NOTE! We MUST NOT take any locks for this case. We may
421 * be in an interrupt or a critical region, and should
422 * only copy the information from the master page table,
425 if (unlikely(fault_in_kernel_space(address))) {
426 if (vmalloc_fault(address) >= 0)
428 if (notify_page_fault(regs, vec))
431 bad_area_nosemaphore(regs, error_code, address);
435 if (unlikely(notify_page_fault(regs, vec)))
438 /* Only enable interrupts if they were on before the fault */
439 if ((regs->sr & SR_IMASK) != SR_IMASK)
442 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
445 * If we're in an interrupt, have no user context or are running
446 * in an atomic region then we must not take the fault:
448 if (unlikely(in_atomic() || !mm)) {
449 bad_area_nosemaphore(regs, error_code, address);
454 down_read(&mm->mmap_sem);
456 vma = find_vma(mm, address);
457 if (unlikely(!vma)) {
458 bad_area(regs, error_code, address);
461 if (likely(vma->vm_start <= address))
463 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
464 bad_area(regs, error_code, address);
467 if (unlikely(expand_stack(vma, address))) {
468 bad_area(regs, error_code, address);
473 * Ok, we have a good vm_area for this memory access, so
477 if (unlikely(access_error(error_code, vma))) {
478 bad_area_access_error(regs, error_code, address);
482 set_thread_fault_code(error_code);
485 * If for any reason at all we couldn't handle the fault,
486 * make sure we exit gracefully rather than endlessly redo
489 fault = handle_mm_fault(mm, vma, address, flags);
491 if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
492 if (mm_fault_error(regs, error_code, address, fault))
495 if (flags & FAULT_FLAG_ALLOW_RETRY) {
496 if (fault & VM_FAULT_MAJOR) {
498 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
502 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
505 if (fault & VM_FAULT_RETRY) {
506 flags &= ~FAULT_FLAG_ALLOW_RETRY;
507 flags |= FAULT_FLAG_TRIED;
510 * No need to up_read(&mm->mmap_sem) as we would
511 * have already released it in __lock_page_or_retry
518 up_read(&mm->mmap_sem);