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/sched/signal.h>
17 #include <linux/hardirq.h>
18 #include <linux/kprobes.h>
19 #include <linux/perf_event.h>
20 #include <linux/kdebug.h>
21 #include <linux/uaccess.h>
22 #include <asm/io_trapped.h>
23 #include <asm/mmu_context.h>
24 #include <asm/tlbflush.h>
25 #include <asm/traps.h>
28 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
)
30 force_sig_fault(si_signo
, si_code
, (void __user
*)address
);
34 * This is useful to dump out the page tables associated with
37 static void show_pte(struct mm_struct
*mm
, unsigned long addr
)
50 pr_alert("pgd = %p\n", pgd
);
51 pgd
+= pgd_index(addr
);
52 pr_alert("[%08lx] *pgd=%0*llx", addr
, (u32
)(sizeof(*pgd
) * 2),
69 p4d
= p4d_offset(pgd
, addr
);
70 if (PTRS_PER_P4D
!= 1)
71 pr_cont(", *p4d=%0*Lx", (u32
)(sizeof(*p4d
) * 2),
82 pud
= pud_offset(p4d
, addr
);
83 if (PTRS_PER_PUD
!= 1)
84 pr_cont(", *pud=%0*llx", (u32
)(sizeof(*pud
) * 2),
95 pmd
= pmd_offset(pud
, addr
);
96 if (PTRS_PER_PMD
!= 1)
97 pr_cont(", *pmd=%0*llx", (u32
)(sizeof(*pmd
) * 2),
108 /* We must not map this if we have highmem enabled */
109 if (PageHighMem(pfn_to_page(pmd_val(*pmd
) >> PAGE_SHIFT
)))
112 pte
= pte_offset_kernel(pmd
, addr
);
113 pr_cont(", *pte=%0*llx", (u32
)(sizeof(*pte
) * 2),
120 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
122 unsigned index
= pgd_index(address
);
129 pgd_k
= init_mm
.pgd
+ index
;
131 if (!pgd_present(*pgd_k
))
134 p4d
= p4d_offset(pgd
, address
);
135 p4d_k
= p4d_offset(pgd_k
, address
);
136 if (!p4d_present(*p4d_k
))
139 pud
= pud_offset(p4d
, address
);
140 pud_k
= pud_offset(p4d_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
"PC:");
212 pr_alert("BUG: unable to handle kernel %s at %08lx\n",
213 address
< PAGE_SIZE
? "NULL pointer dereference"
217 printk_address(regs
->pc
, 1, KERN_ALERT
);
219 show_pte(NULL
, address
);
223 no_context(struct pt_regs
*regs
, unsigned long error_code
,
224 unsigned long address
)
226 /* Are we prepared to handle this kernel fault? */
227 if (fixup_exception(regs
))
230 if (handle_trapped_io(regs
, address
))
234 * Oops. The kernel tried to access some bad page. We'll have to
235 * terminate things with extreme prejudice.
239 show_fault_oops(regs
, address
);
241 die("Oops", regs
, error_code
);
247 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
248 unsigned long address
, int si_code
)
250 /* User mode accesses just cause a SIGSEGV */
251 if (user_mode(regs
)) {
253 * It's possible to have interrupts off here:
257 force_sig_info_fault(SIGSEGV
, si_code
, address
);
262 no_context(regs
, error_code
, address
);
266 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
267 unsigned long address
)
269 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
273 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
274 unsigned long address
, int si_code
)
276 struct mm_struct
*mm
= current
->mm
;
279 * Something tried to access memory that isn't in our memory map..
280 * Fix it, but check if it's kernel or user first..
282 mmap_read_unlock(mm
);
284 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
288 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
290 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
294 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
295 unsigned long address
)
297 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
301 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
303 struct task_struct
*tsk
= current
;
304 struct mm_struct
*mm
= tsk
->mm
;
306 mmap_read_unlock(mm
);
308 /* Kernel mode? Handle exceptions or die: */
309 if (!user_mode(regs
))
310 no_context(regs
, error_code
, address
);
312 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
);
316 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
317 unsigned long address
, vm_fault_t fault
)
320 * Pagefault was interrupted by SIGKILL. We have no reason to
321 * continue pagefault.
323 if (fault_signal_pending(fault
, regs
)) {
324 if (!user_mode(regs
))
325 no_context(regs
, error_code
, address
);
329 /* Release mmap_sem first if necessary */
330 if (!(fault
& VM_FAULT_RETRY
))
331 mmap_read_unlock(current
->mm
);
333 if (!(fault
& VM_FAULT_ERROR
))
336 if (fault
& VM_FAULT_OOM
) {
337 /* Kernel mode? Handle exceptions or die: */
338 if (!user_mode(regs
)) {
339 no_context(regs
, error_code
, address
);
344 * We ran out of memory, call the OOM killer, and return the
345 * userspace (which will retry the fault, or kill us if we got
348 pagefault_out_of_memory();
350 if (fault
& VM_FAULT_SIGBUS
)
351 do_sigbus(regs
, error_code
, address
);
352 else if (fault
& VM_FAULT_SIGSEGV
)
353 bad_area(regs
, error_code
, address
);
361 static inline int access_error(int error_code
, struct vm_area_struct
*vma
)
363 if (error_code
& FAULT_CODE_WRITE
) {
364 /* write, present and write, not present: */
365 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
370 /* ITLB miss on NX page */
371 if (unlikely((error_code
& FAULT_CODE_ITLB
) &&
372 !(vma
->vm_flags
& VM_EXEC
)))
375 /* read, not present: */
376 if (unlikely(!vma_is_accessible(vma
)))
382 static int fault_in_kernel_space(unsigned long address
)
384 return address
>= TASK_SIZE
;
388 * This routine handles page faults. It determines the address,
389 * and the problem, and then passes it off to one of the appropriate
392 asmlinkage
void __kprobes
do_page_fault(struct pt_regs
*regs
,
393 unsigned long error_code
,
394 unsigned long address
)
397 struct task_struct
*tsk
;
398 struct mm_struct
*mm
;
399 struct vm_area_struct
* vma
;
401 unsigned int flags
= FAULT_FLAG_DEFAULT
;
405 vec
= lookup_exception_vector();
408 * We fault-in kernel-space virtual memory on-demand. The
409 * 'reference' page table is init_mm.pgd.
411 * NOTE! We MUST NOT take any locks for this case. We may
412 * be in an interrupt or a critical region, and should
413 * only copy the information from the master page table,
416 if (unlikely(fault_in_kernel_space(address
))) {
417 if (vmalloc_fault(address
) >= 0)
419 if (kprobe_page_fault(regs
, vec
))
422 bad_area_nosemaphore(regs
, error_code
, address
);
426 if (unlikely(kprobe_page_fault(regs
, vec
)))
429 /* Only enable interrupts if they were on before the fault */
430 if ((regs
->sr
& SR_IMASK
) != SR_IMASK
)
433 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
436 * If we're in an interrupt, have no user context or are running
437 * with pagefaults disabled then we must not take the fault:
439 if (unlikely(faulthandler_disabled() || !mm
)) {
440 bad_area_nosemaphore(regs
, error_code
, address
);
447 vma
= find_vma(mm
, address
);
448 if (unlikely(!vma
)) {
449 bad_area(regs
, error_code
, address
);
452 if (likely(vma
->vm_start
<= address
))
454 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
455 bad_area(regs
, error_code
, address
);
458 if (unlikely(expand_stack(vma
, address
))) {
459 bad_area(regs
, error_code
, address
);
464 * Ok, we have a good vm_area for this memory access, so
468 if (unlikely(access_error(error_code
, vma
))) {
469 bad_area_access_error(regs
, error_code
, address
);
473 set_thread_fault_code(error_code
);
476 flags
|= FAULT_FLAG_USER
;
477 if (error_code
& FAULT_CODE_WRITE
)
478 flags
|= FAULT_FLAG_WRITE
;
481 * If for any reason at all we couldn't handle the fault,
482 * make sure we exit gracefully rather than endlessly redo
485 fault
= handle_mm_fault(vma
, address
, flags
);
487 if (unlikely(fault
& (VM_FAULT_RETRY
| VM_FAULT_ERROR
)))
488 if (mm_fault_error(regs
, error_code
, address
, fault
))
491 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
492 if (fault
& VM_FAULT_MAJOR
) {
494 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1,
498 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1,
501 if (fault
& VM_FAULT_RETRY
) {
502 flags
|= FAULT_FLAG_TRIED
;
505 * No need to mmap_read_unlock(mm) as we would
506 * have already released it in __lock_page_or_retry
513 mmap_read_unlock(mm
);