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 <linux/uaccess.h>
21 #include <asm/io_trapped.h>
22 #include <asm/mmu_context.h>
23 #include <asm/tlbflush.h>
24 #include <asm/traps.h>
26 static inline int notify_page_fault(struct pt_regs
*regs
, int trap
)
30 if (kprobes_built_in() && !user_mode(regs
)) {
32 if (kprobe_running() && kprobe_fault_handler(regs
, trap
))
41 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
42 struct task_struct
*tsk
)
46 info
.si_signo
= si_signo
;
48 info
.si_code
= si_code
;
49 info
.si_addr
= (void __user
*)address
;
51 force_sig_info(si_signo
, &info
, tsk
);
55 * This is useful to dump out the page tables associated with
58 static void show_pte(struct mm_struct
*mm
, unsigned long addr
)
71 printk(KERN_ALERT
"pgd = %p\n", pgd
);
72 pgd
+= pgd_index(addr
);
73 printk(KERN_ALERT
"[%08lx] *pgd=%0*Lx", addr
,
74 (u32
)(sizeof(*pgd
) * 2), (u64
)pgd_val(*pgd
));
89 pud
= pud_offset(pgd
, addr
);
90 if (PTRS_PER_PUD
!= 1)
91 printk(", *pud=%0*Lx", (u32
)(sizeof(*pud
) * 2),
102 pmd
= pmd_offset(pud
, addr
);
103 if (PTRS_PER_PMD
!= 1)
104 printk(", *pmd=%0*Lx", (u32
)(sizeof(*pmd
) * 2),
115 /* We must not map this if we have highmem enabled */
116 if (PageHighMem(pfn_to_page(pmd_val(*pmd
) >> PAGE_SHIFT
)))
119 pte
= pte_offset_kernel(pmd
, addr
);
120 printk(", *pte=%0*Lx", (u32
)(sizeof(*pte
) * 2),
127 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
129 unsigned index
= pgd_index(address
);
135 pgd_k
= init_mm
.pgd
+ index
;
137 if (!pgd_present(*pgd_k
))
140 pud
= pud_offset(pgd
, address
);
141 pud_k
= pud_offset(pgd_k
, address
);
142 if (!pud_present(*pud_k
))
145 if (!pud_present(*pud
))
146 set_pud(pud
, *pud_k
);
148 pmd
= pmd_offset(pud
, address
);
149 pmd_k
= pmd_offset(pud_k
, address
);
150 if (!pmd_present(*pmd_k
))
153 if (!pmd_present(*pmd
))
154 set_pmd(pmd
, *pmd_k
);
157 * The page tables are fully synchronised so there must
158 * be another reason for the fault. Return NULL here to
159 * signal that we have not taken care of the fault.
161 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
168 #ifdef CONFIG_SH_STORE_QUEUES
169 #define __FAULT_ADDR_LIMIT P3_ADDR_MAX
171 #define __FAULT_ADDR_LIMIT VMALLOC_END
175 * Handle a fault on the vmalloc or module mapping area
177 static noinline
int vmalloc_fault(unsigned long address
)
183 /* Make sure we are in vmalloc/module/P3 area: */
184 if (!(address
>= VMALLOC_START
&& address
< __FAULT_ADDR_LIMIT
))
188 * Synchronize this task's top level page-table
189 * with the 'reference' page table.
191 * Do _not_ use "current" here. We might be inside
192 * an interrupt in the middle of a task switch..
195 pmd_k
= vmalloc_sync_one(pgd_k
, address
);
199 pte_k
= pte_offset_kernel(pmd_k
, address
);
200 if (!pte_present(*pte_k
))
207 show_fault_oops(struct pt_regs
*regs
, unsigned long address
)
209 if (!oops_may_print())
212 printk(KERN_ALERT
"BUG: unable to handle kernel ");
213 if (address
< PAGE_SIZE
)
214 printk(KERN_CONT
"NULL pointer dereference");
216 printk(KERN_CONT
"paging request");
218 printk(KERN_CONT
" at %08lx\n", address
);
219 printk(KERN_ALERT
"PC:");
220 printk_address(regs
->pc
, 1);
222 show_pte(NULL
, address
);
226 no_context(struct pt_regs
*regs
, unsigned long error_code
,
227 unsigned long address
)
229 /* Are we prepared to handle this kernel fault? */
230 if (fixup_exception(regs
))
233 if (handle_trapped_io(regs
, address
))
237 * Oops. The kernel tried to access some bad page. We'll have to
238 * terminate things with extreme prejudice.
242 show_fault_oops(regs
, address
);
244 die("Oops", regs
, error_code
);
250 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
251 unsigned long address
, int si_code
)
253 struct task_struct
*tsk
= current
;
255 /* User mode accesses just cause a SIGSEGV */
256 if (user_mode(regs
)) {
258 * It's possible to have interrupts off here:
262 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
267 no_context(regs
, error_code
, address
);
271 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
272 unsigned long address
)
274 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
278 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
279 unsigned long address
, int si_code
)
281 struct mm_struct
*mm
= current
->mm
;
284 * Something tried to access memory that isn't in our memory map..
285 * Fix it, but check if it's kernel or user first..
287 up_read(&mm
->mmap_sem
);
289 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
293 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
295 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
299 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
300 unsigned long address
)
302 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
306 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
308 struct task_struct
*tsk
= current
;
309 struct mm_struct
*mm
= tsk
->mm
;
311 up_read(&mm
->mmap_sem
);
313 /* Kernel mode? Handle exceptions or die: */
314 if (!user_mode(regs
))
315 no_context(regs
, error_code
, address
);
317 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
321 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
322 unsigned long address
, unsigned int fault
)
325 * Pagefault was interrupted by SIGKILL. We have no reason to
326 * continue pagefault.
328 if (fatal_signal_pending(current
)) {
329 if (!(fault
& VM_FAULT_RETRY
))
330 up_read(¤t
->mm
->mmap_sem
);
331 if (!user_mode(regs
))
332 no_context(regs
, error_code
, address
);
336 if (!(fault
& VM_FAULT_ERROR
))
339 if (fault
& VM_FAULT_OOM
) {
340 /* Kernel mode? Handle exceptions or die: */
341 if (!user_mode(regs
)) {
342 up_read(¤t
->mm
->mmap_sem
);
343 no_context(regs
, error_code
, address
);
346 up_read(¤t
->mm
->mmap_sem
);
349 * We ran out of memory, call the OOM killer, and return the
350 * userspace (which will retry the fault, or kill us if we got
353 pagefault_out_of_memory();
355 if (fault
& VM_FAULT_SIGBUS
)
356 do_sigbus(regs
, error_code
, address
);
357 else if (fault
& VM_FAULT_SIGSEGV
)
358 bad_area(regs
, error_code
, address
);
366 static inline int access_error(int error_code
, struct vm_area_struct
*vma
)
368 if (error_code
& FAULT_CODE_WRITE
) {
369 /* write, present and write, not present: */
370 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
375 /* ITLB miss on NX page */
376 if (unlikely((error_code
& FAULT_CODE_ITLB
) &&
377 !(vma
->vm_flags
& VM_EXEC
)))
380 /* read, not present: */
381 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
387 static int fault_in_kernel_space(unsigned long address
)
389 return address
>= TASK_SIZE
;
393 * This routine handles page faults. It determines the address,
394 * and the problem, and then passes it off to one of the appropriate
397 asmlinkage
void __kprobes
do_page_fault(struct pt_regs
*regs
,
398 unsigned long error_code
,
399 unsigned long address
)
402 struct task_struct
*tsk
;
403 struct mm_struct
*mm
;
404 struct vm_area_struct
* vma
;
406 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
410 vec
= lookup_exception_vector();
413 * We fault-in kernel-space virtual memory on-demand. The
414 * 'reference' page table is init_mm.pgd.
416 * NOTE! We MUST NOT take any locks for this case. We may
417 * be in an interrupt or a critical region, and should
418 * only copy the information from the master page table,
421 if (unlikely(fault_in_kernel_space(address
))) {
422 if (vmalloc_fault(address
) >= 0)
424 if (notify_page_fault(regs
, vec
))
427 bad_area_nosemaphore(regs
, error_code
, address
);
431 if (unlikely(notify_page_fault(regs
, vec
)))
434 /* Only enable interrupts if they were on before the fault */
435 if ((regs
->sr
& SR_IMASK
) != SR_IMASK
)
438 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
441 * If we're in an interrupt, have no user context or are running
442 * with pagefaults disabled then we must not take the fault:
444 if (unlikely(faulthandler_disabled() || !mm
)) {
445 bad_area_nosemaphore(regs
, error_code
, address
);
450 down_read(&mm
->mmap_sem
);
452 vma
= find_vma(mm
, address
);
453 if (unlikely(!vma
)) {
454 bad_area(regs
, error_code
, address
);
457 if (likely(vma
->vm_start
<= address
))
459 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
460 bad_area(regs
, error_code
, address
);
463 if (unlikely(expand_stack(vma
, address
))) {
464 bad_area(regs
, error_code
, address
);
469 * Ok, we have a good vm_area for this memory access, so
473 if (unlikely(access_error(error_code
, vma
))) {
474 bad_area_access_error(regs
, error_code
, address
);
478 set_thread_fault_code(error_code
);
481 flags
|= FAULT_FLAG_USER
;
482 if (error_code
& FAULT_CODE_WRITE
)
483 flags
|= FAULT_FLAG_WRITE
;
486 * If for any reason at all we couldn't handle the fault,
487 * make sure we exit gracefully rather than endlessly redo
490 fault
= handle_mm_fault(vma
, address
, flags
);
492 if (unlikely(fault
& (VM_FAULT_RETRY
| VM_FAULT_ERROR
)))
493 if (mm_fault_error(regs
, error_code
, address
, fault
))
496 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
497 if (fault
& VM_FAULT_MAJOR
) {
499 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1,
503 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1,
506 if (fault
& VM_FAULT_RETRY
) {
507 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
508 flags
|= FAULT_FLAG_TRIED
;
511 * No need to up_read(&mm->mmap_sem) as we would
512 * have already released it in __lock_page_or_retry
519 up_read(&mm
->mmap_sem
);