1/*
2 * Page fault handler for SH with an MMU.
3 *
4 *  Copyright (C) 1999  Niibe Yutaka
5 *  Copyright (C) 2003 - 2012  Paul Mundt
6 *
7 *  Based on linux/arch/i386/mm/fault.c:
8 *   Copyright (C) 1995  Linus Torvalds
9 *
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
12 * for more details.
13 */
14#include <linux/kernel.h>
15#include <linux/mm.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>
25
26static inline int notify_page_fault(struct pt_regs *regs, int trap)
27{
28	int ret = 0;
29
30	if (kprobes_built_in() && !user_mode(regs)) {
31		preempt_disable();
32		if (kprobe_running() && kprobe_fault_handler(regs, trap))
33			ret = 1;
34		preempt_enable();
35	}
36
37	return ret;
38}
39
40static void
41force_sig_info_fault(int si_signo, int si_code, unsigned long address,
42		     struct task_struct *tsk)
43{
44	siginfo_t info;
45
46	info.si_signo	= si_signo;
47	info.si_errno	= 0;
48	info.si_code	= si_code;
49	info.si_addr	= (void __user *)address;
50
51	force_sig_info(si_signo, &info, tsk);
52}
53
54/*
55 * This is useful to dump out the page tables associated with
56 * 'addr' in mm 'mm'.
57 */
58static void show_pte(struct mm_struct *mm, unsigned long addr)
59{
60	pgd_t *pgd;
61
62	if (mm) {
63		pgd = mm->pgd;
64	} else {
65		pgd = get_TTB();
66
67		if (unlikely(!pgd))
68			pgd = swapper_pg_dir;
69	}
70
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));
75
76	do {
77		pud_t *pud;
78		pmd_t *pmd;
79		pte_t *pte;
80
81		if (pgd_none(*pgd))
82			break;
83
84		if (pgd_bad(*pgd)) {
85			printk("(bad)");
86			break;
87		}
88
89		pud = pud_offset(pgd, addr);
90		if (PTRS_PER_PUD != 1)
91			printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
92			       (u64)pud_val(*pud));
93
94		if (pud_none(*pud))
95			break;
96
97		if (pud_bad(*pud)) {
98			printk("(bad)");
99			break;
100		}
101
102		pmd = pmd_offset(pud, addr);
103		if (PTRS_PER_PMD != 1)
104			printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
105			       (u64)pmd_val(*pmd));
106
107		if (pmd_none(*pmd))
108			break;
109
110		if (pmd_bad(*pmd)) {
111			printk("(bad)");
112			break;
113		}
114
115		/* We must not map this if we have highmem enabled */
116		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
117			break;
118
119		pte = pte_offset_kernel(pmd, addr);
120		printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
121		       (u64)pte_val(*pte));
122	} while (0);
123
124	printk("\n");
125}
126
127static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
128{
129	unsigned index = pgd_index(address);
130	pgd_t *pgd_k;
131	pud_t *pud, *pud_k;
132	pmd_t *pmd, *pmd_k;
133
134	pgd += index;
135	pgd_k = init_mm.pgd + index;
136
137	if (!pgd_present(*pgd_k))
138		return NULL;
139
140	pud = pud_offset(pgd, address);
141	pud_k = pud_offset(pgd_k, address);
142	if (!pud_present(*pud_k))
143		return NULL;
144
145	if (!pud_present(*pud))
146	    set_pud(pud, *pud_k);
147
148	pmd = pmd_offset(pud, address);
149	pmd_k = pmd_offset(pud_k, address);
150	if (!pmd_present(*pmd_k))
151		return NULL;
152
153	if (!pmd_present(*pmd))
154		set_pmd(pmd, *pmd_k);
155	else {
156		/*
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.
160		 */
161		BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
162		return NULL;
163	}
164
165	return pmd_k;
166}
167
168#ifdef CONFIG_SH_STORE_QUEUES
169#define __FAULT_ADDR_LIMIT	P3_ADDR_MAX
170#else
171#define __FAULT_ADDR_LIMIT	VMALLOC_END
172#endif
173
174/*
175 * Handle a fault on the vmalloc or module mapping area
176 */
177static noinline int vmalloc_fault(unsigned long address)
178{
179	pgd_t *pgd_k;
180	pmd_t *pmd_k;
181	pte_t *pte_k;
182
183	/* Make sure we are in vmalloc/module/P3 area: */
184	if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
185		return -1;
186
187	/*
188	 * Synchronize this task's top level page-table
189	 * with the 'reference' page table.
190	 *
191	 * Do _not_ use "current" here. We might be inside
192	 * an interrupt in the middle of a task switch..
193	 */
194	pgd_k = get_TTB();
195	pmd_k = vmalloc_sync_one(pgd_k, address);
196	if (!pmd_k)
197		return -1;
198
199	pte_k = pte_offset_kernel(pmd_k, address);
200	if (!pte_present(*pte_k))
201		return -1;
202
203	return 0;
204}
205
206static void
207show_fault_oops(struct pt_regs *regs, unsigned long address)
208{
209	if (!oops_may_print())
210		return;
211
212	printk(KERN_ALERT "BUG: unable to handle kernel ");
213	if (address < PAGE_SIZE)
214		printk(KERN_CONT "NULL pointer dereference");
215	else
216		printk(KERN_CONT "paging request");
217
218	printk(KERN_CONT " at %08lx\n", address);
219	printk(KERN_ALERT "PC:");
220	printk_address(regs->pc, 1);
221
222	show_pte(NULL, address);
223}
224
225static noinline void
226no_context(struct pt_regs *regs, unsigned long error_code,
227	   unsigned long address)
228{
229	/* Are we prepared to handle this kernel fault?  */
230	if (fixup_exception(regs))
231		return;
232
233	if (handle_trapped_io(regs, address))
234		return;
235
236	/*
237	 * Oops. The kernel tried to access some bad page. We'll have to
238	 * terminate things with extreme prejudice.
239	 */
240	bust_spinlocks(1);
241
242	show_fault_oops(regs, address);
243
244	die("Oops", regs, error_code);
245	bust_spinlocks(0);
246	do_exit(SIGKILL);
247}
248
249static void
250__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
251		       unsigned long address, int si_code)
252{
253	struct task_struct *tsk = current;
254
255	/* User mode accesses just cause a SIGSEGV */
256	if (user_mode(regs)) {
257		/*
258		 * It's possible to have interrupts off here:
259		 */
260		local_irq_enable();
261
262		force_sig_info_fault(SIGSEGV, si_code, address, tsk);
263
264		return;
265	}
266
267	no_context(regs, error_code, address);
268}
269
270static noinline void
271bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
272		     unsigned long address)
273{
274	__bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
275}
276
277static void
278__bad_area(struct pt_regs *regs, unsigned long error_code,
279	   unsigned long address, int si_code)
280{
281	struct mm_struct *mm = current->mm;
282
283	/*
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..
286	 */
287	up_read(&mm->mmap_sem);
288
289	__bad_area_nosemaphore(regs, error_code, address, si_code);
290}
291
292static noinline void
293bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
294{
295	__bad_area(regs, error_code, address, SEGV_MAPERR);
296}
297
298static noinline void
299bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
300		      unsigned long address)
301{
302	__bad_area(regs, error_code, address, SEGV_ACCERR);
303}
304
305static void
306do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
307{
308	struct task_struct *tsk = current;
309	struct mm_struct *mm = tsk->mm;
310
311	up_read(&mm->mmap_sem);
312
313	/* Kernel mode? Handle exceptions or die: */
314	if (!user_mode(regs))
315		no_context(regs, error_code, address);
316
317	force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
318}
319
320static noinline int
321mm_fault_error(struct pt_regs *regs, unsigned long error_code,
322	       unsigned long address, unsigned int fault)
323{
324	/*
325	 * Pagefault was interrupted by SIGKILL. We have no reason to
326	 * continue pagefault.
327	 */
328	if (fatal_signal_pending(current)) {
329		if (!(fault & VM_FAULT_RETRY))
330			up_read(&current->mm->mmap_sem);
331		if (!user_mode(regs))
332			no_context(regs, error_code, address);
333		return 1;
334	}
335
336	if (!(fault & VM_FAULT_ERROR))
337		return 0;
338
339	if (fault & VM_FAULT_OOM) {
340		/* Kernel mode? Handle exceptions or die: */
341		if (!user_mode(regs)) {
342			up_read(&current->mm->mmap_sem);
343			no_context(regs, error_code, address);
344			return 1;
345		}
346		up_read(&current->mm->mmap_sem);
347
348		/*
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
351		 * oom-killed):
352		 */
353		pagefault_out_of_memory();
354	} else {
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);
359		else
360			BUG();
361	}
362
363	return 1;
364}
365
366static inline int access_error(int error_code, struct vm_area_struct *vma)
367{
368	if (error_code & FAULT_CODE_WRITE) {
369		/* write, present and write, not present: */
370		if (unlikely(!(vma->vm_flags & VM_WRITE)))
371			return 1;
372		return 0;
373	}
374
375	/* ITLB miss on NX page */
376	if (unlikely((error_code & FAULT_CODE_ITLB) &&
377		     !(vma->vm_flags & VM_EXEC)))
378		return 1;
379
380	/* read, not present: */
381	if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
382		return 1;
383
384	return 0;
385}
386
387static int fault_in_kernel_space(unsigned long address)
388{
389	return address >= TASK_SIZE;
390}
391
392/*
393 * This routine handles page faults.  It determines the address,
394 * and the problem, and then passes it off to one of the appropriate
395 * routines.
396 */
397asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
398					unsigned long error_code,
399					unsigned long address)
400{
401	unsigned long vec;
402	struct task_struct *tsk;
403	struct mm_struct *mm;
404	struct vm_area_struct * vma;
405	int fault;
406	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
407
408	tsk = current;
409	mm = tsk->mm;
410	vec = lookup_exception_vector();
411
412	/*
413	 * We fault-in kernel-space virtual memory on-demand. The
414	 * 'reference' page table is init_mm.pgd.
415	 *
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,
419	 * nothing more.
420	 */
421	if (unlikely(fault_in_kernel_space(address))) {
422		if (vmalloc_fault(address) >= 0)
423			return;
424		if (notify_page_fault(regs, vec))
425			return;
426
427		bad_area_nosemaphore(regs, error_code, address);
428		return;
429	}
430
431	if (unlikely(notify_page_fault(regs, vec)))
432		return;
433
434	/* Only enable interrupts if they were on before the fault */
435	if ((regs->sr & SR_IMASK) != SR_IMASK)
436		local_irq_enable();
437
438	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
439
440	/*
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:
443	 */
444	if (unlikely(faulthandler_disabled() || !mm)) {
445		bad_area_nosemaphore(regs, error_code, address);
446		return;
447	}
448
449retry:
450	down_read(&mm->mmap_sem);
451
452	vma = find_vma(mm, address);
453	if (unlikely(!vma)) {
454		bad_area(regs, error_code, address);
455		return;
456	}
457	if (likely(vma->vm_start <= address))
458		goto good_area;
459	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
460		bad_area(regs, error_code, address);
461		return;
462	}
463	if (unlikely(expand_stack(vma, address))) {
464		bad_area(regs, error_code, address);
465		return;
466	}
467
468	/*
469	 * Ok, we have a good vm_area for this memory access, so
470	 * we can handle it..
471	 */
472good_area:
473	if (unlikely(access_error(error_code, vma))) {
474		bad_area_access_error(regs, error_code, address);
475		return;
476	}
477
478	set_thread_fault_code(error_code);
479
480	if (user_mode(regs))
481		flags |= FAULT_FLAG_USER;
482	if (error_code & FAULT_CODE_WRITE)
483		flags |= FAULT_FLAG_WRITE;
484
485	/*
486	 * If for any reason at all we couldn't handle the fault,
487	 * make sure we exit gracefully rather than endlessly redo
488	 * the fault.
489	 */
490	fault = handle_mm_fault(mm, vma, address, flags);
491
492	if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
493		if (mm_fault_error(regs, error_code, address, fault))
494			return;
495
496	if (flags & FAULT_FLAG_ALLOW_RETRY) {
497		if (fault & VM_FAULT_MAJOR) {
498			tsk->maj_flt++;
499			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
500				      regs, address);
501		} else {
502			tsk->min_flt++;
503			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
504				      regs, address);
505		}
506		if (fault & VM_FAULT_RETRY) {
507			flags &= ~FAULT_FLAG_ALLOW_RETRY;
508			flags |= FAULT_FLAG_TRIED;
509
510			/*
511			 * No need to up_read(&mm->mmap_sem) as we would
512			 * have already released it in __lock_page_or_retry
513			 * in mm/filemap.c.
514			 */
515			goto retry;
516		}
517	}
518
519	up_read(&mm->mmap_sem);
520}
521