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 <asm/io_trapped.h>
21#include <asm/mmu_context.h>
22#include <asm/tlbflush.h>
23#include <asm/traps.h>
24
25static inline int notify_page_fault(struct pt_regs *regs, int trap)
26{
27	int ret = 0;
28
29	if (kprobes_built_in() && !user_mode(regs)) {
30		preempt_disable();
31		if (kprobe_running() && kprobe_fault_handler(regs, trap))
32			ret = 1;
33		preempt_enable();
34	}
35
36	return ret;
37}
38
39static void
40force_sig_info_fault(int si_signo, int si_code, unsigned long address,
41		     struct task_struct *tsk)
42{
43	siginfo_t info;
44
45	info.si_signo	= si_signo;
46	info.si_errno	= 0;
47	info.si_code	= si_code;
48	info.si_addr	= (void __user *)address;
49
50	force_sig_info(si_signo, &info, tsk);
51}
52
53/*
54 * This is useful to dump out the page tables associated with
55 * 'addr' in mm 'mm'.
56 */
57static void show_pte(struct mm_struct *mm, unsigned long addr)
58{
59	pgd_t *pgd;
60
61	if (mm) {
62		pgd = mm->pgd;
63	} else {
64		pgd = get_TTB();
65
66		if (unlikely(!pgd))
67			pgd = swapper_pg_dir;
68	}
69
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));
74
75	do {
76		pud_t *pud;
77		pmd_t *pmd;
78		pte_t *pte;
79
80		if (pgd_none(*pgd))
81			break;
82
83		if (pgd_bad(*pgd)) {
84			printk("(bad)");
85			break;
86		}
87
88		pud = pud_offset(pgd, addr);
89		if (PTRS_PER_PUD != 1)
90			printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
91			       (u64)pud_val(*pud));
92
93		if (pud_none(*pud))
94			break;
95
96		if (pud_bad(*pud)) {
97			printk("(bad)");
98			break;
99		}
100
101		pmd = pmd_offset(pud, addr);
102		if (PTRS_PER_PMD != 1)
103			printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
104			       (u64)pmd_val(*pmd));
105
106		if (pmd_none(*pmd))
107			break;
108
109		if (pmd_bad(*pmd)) {
110			printk("(bad)");
111			break;
112		}
113
114		/* We must not map this if we have highmem enabled */
115		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
116			break;
117
118		pte = pte_offset_kernel(pmd, addr);
119		printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
120		       (u64)pte_val(*pte));
121	} while (0);
122
123	printk("\n");
124}
125
126static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
127{
128	unsigned index = pgd_index(address);
129	pgd_t *pgd_k;
130	pud_t *pud, *pud_k;
131	pmd_t *pmd, *pmd_k;
132
133	pgd += index;
134	pgd_k = init_mm.pgd + index;
135
136	if (!pgd_present(*pgd_k))
137		return NULL;
138
139	pud = pud_offset(pgd, address);
140	pud_k = pud_offset(pgd_k, address);
141	if (!pud_present(*pud_k))
142		return NULL;
143
144	if (!pud_present(*pud))
145	    set_pud(pud, *pud_k);
146
147	pmd = pmd_offset(pud, address);
148	pmd_k = pmd_offset(pud_k, address);
149	if (!pmd_present(*pmd_k))
150		return NULL;
151
152	if (!pmd_present(*pmd))
153		set_pmd(pmd, *pmd_k);
154	else {
155		/*
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.
159		 */
160		BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
161		return NULL;
162	}
163
164	return pmd_k;
165}
166
167#ifdef CONFIG_SH_STORE_QUEUES
168#define __FAULT_ADDR_LIMIT	P3_ADDR_MAX
169#else
170#define __FAULT_ADDR_LIMIT	VMALLOC_END
171#endif
172
173/*
174 * Handle a fault on the vmalloc or module mapping area
175 */
176static noinline int vmalloc_fault(unsigned long address)
177{
178	pgd_t *pgd_k;
179	pmd_t *pmd_k;
180	pte_t *pte_k;
181
182	/* Make sure we are in vmalloc/module/P3 area: */
183	if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
184		return -1;
185
186	/*
187	 * Synchronize this task's top level page-table
188	 * with the 'reference' page table.
189	 *
190	 * Do _not_ use "current" here. We might be inside
191	 * an interrupt in the middle of a task switch..
192	 */
193	pgd_k = get_TTB();
194	pmd_k = vmalloc_sync_one(pgd_k, address);
195	if (!pmd_k)
196		return -1;
197
198	pte_k = pte_offset_kernel(pmd_k, address);
199	if (!pte_present(*pte_k))
200		return -1;
201
202	return 0;
203}
204
205static void
206show_fault_oops(struct pt_regs *regs, unsigned long address)
207{
208	if (!oops_may_print())
209		return;
210
211	printk(KERN_ALERT "BUG: unable to handle kernel ");
212	if (address < PAGE_SIZE)
213		printk(KERN_CONT "NULL pointer dereference");
214	else
215		printk(KERN_CONT "paging request");
216
217	printk(KERN_CONT " at %08lx\n", address);
218	printk(KERN_ALERT "PC:");
219	printk_address(regs->pc, 1);
220
221	show_pte(NULL, address);
222}
223
224static noinline void
225no_context(struct pt_regs *regs, unsigned long error_code,
226	   unsigned long address)
227{
228	/* Are we prepared to handle this kernel fault?  */
229	if (fixup_exception(regs))
230		return;
231
232	if (handle_trapped_io(regs, address))
233		return;
234
235	/*
236	 * Oops. The kernel tried to access some bad page. We'll have to
237	 * terminate things with extreme prejudice.
238	 */
239	bust_spinlocks(1);
240
241	show_fault_oops(regs, address);
242
243	die("Oops", regs, error_code);
244	bust_spinlocks(0);
245	do_exit(SIGKILL);
246}
247
248static void
249__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
250		       unsigned long address, int si_code)
251{
252	struct task_struct *tsk = current;
253
254	/* User mode accesses just cause a SIGSEGV */
255	if (user_mode(regs)) {
256		/*
257		 * It's possible to have interrupts off here:
258		 */
259		local_irq_enable();
260
261		force_sig_info_fault(SIGSEGV, si_code, address, tsk);
262
263		return;
264	}
265
266	no_context(regs, error_code, address);
267}
268
269static noinline void
270bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
271		     unsigned long address)
272{
273	__bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
274}
275
276static void
277__bad_area(struct pt_regs *regs, unsigned long error_code,
278	   unsigned long address, int si_code)
279{
280	struct mm_struct *mm = current->mm;
281
282	/*
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..
285	 */
286	up_read(&mm->mmap_sem);
287
288	__bad_area_nosemaphore(regs, error_code, address, si_code);
289}
290
291static noinline void
292bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
293{
294	__bad_area(regs, error_code, address, SEGV_MAPERR);
295}
296
297static noinline void
298bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
299		      unsigned long address)
300{
301	__bad_area(regs, error_code, address, SEGV_ACCERR);
302}
303
304static void
305do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
306{
307	struct task_struct *tsk = current;
308	struct mm_struct *mm = tsk->mm;
309
310	up_read(&mm->mmap_sem);
311
312	/* Kernel mode? Handle exceptions or die: */
313	if (!user_mode(regs))
314		no_context(regs, error_code, address);
315
316	force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
317}
318
319static noinline int
320mm_fault_error(struct pt_regs *regs, unsigned long error_code,
321	       unsigned long address, unsigned int fault)
322{
323	/*
324	 * Pagefault was interrupted by SIGKILL. We have no reason to
325	 * continue pagefault.
326	 */
327	if (fatal_signal_pending(current)) {
328		if (!(fault & VM_FAULT_RETRY))
329			up_read(&current->mm->mmap_sem);
330		if (!user_mode(regs))
331			no_context(regs, error_code, address);
332		return 1;
333	}
334
335	if (!(fault & VM_FAULT_ERROR))
336		return 0;
337
338	if (fault & VM_FAULT_OOM) {
339		/* Kernel mode? Handle exceptions or die: */
340		if (!user_mode(regs)) {
341			up_read(&current->mm->mmap_sem);
342			no_context(regs, error_code, address);
343			return 1;
344		}
345		up_read(&current->mm->mmap_sem);
346
347		/*
348		 * We ran out of memory, call the OOM killer, and return the
349		 * userspace (which will retry the fault, or kill us if we got
350		 * oom-killed):
351		 */
352		pagefault_out_of_memory();
353	} else {
354		if (fault & VM_FAULT_SIGBUS)
355			do_sigbus(regs, error_code, address);
356		else if (fault & VM_FAULT_SIGSEGV)
357			bad_area(regs, error_code, address);
358		else
359			BUG();
360	}
361
362	return 1;
363}
364
365static inline int access_error(int error_code, struct vm_area_struct *vma)
366{
367	if (error_code & FAULT_CODE_WRITE) {
368		/* write, present and write, not present: */
369		if (unlikely(!(vma->vm_flags & VM_WRITE)))
370			return 1;
371		return 0;
372	}
373
374	/* ITLB miss on NX page */
375	if (unlikely((error_code & FAULT_CODE_ITLB) &&
376		     !(vma->vm_flags & VM_EXEC)))
377		return 1;
378
379	/* read, not present: */
380	if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
381		return 1;
382
383	return 0;
384}
385
386static int fault_in_kernel_space(unsigned long address)
387{
388	return address >= TASK_SIZE;
389}
390
391/*
392 * This routine handles page faults.  It determines the address,
393 * and the problem, and then passes it off to one of the appropriate
394 * routines.
395 */
396asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
397					unsigned long error_code,
398					unsigned long address)
399{
400	unsigned long vec;
401	struct task_struct *tsk;
402	struct mm_struct *mm;
403	struct vm_area_struct * vma;
404	int fault;
405	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
406
407	tsk = current;
408	mm = tsk->mm;
409	vec = lookup_exception_vector();
410
411	/*
412	 * We fault-in kernel-space virtual memory on-demand. The
413	 * 'reference' page table is init_mm.pgd.
414	 *
415	 * NOTE! We MUST NOT take any locks for this case. We may
416	 * be in an interrupt or a critical region, and should
417	 * only copy the information from the master page table,
418	 * nothing more.
419	 */
420	if (unlikely(fault_in_kernel_space(address))) {
421		if (vmalloc_fault(address) >= 0)
422			return;
423		if (notify_page_fault(regs, vec))
424			return;
425
426		bad_area_nosemaphore(regs, error_code, address);
427		return;
428	}
429
430	if (unlikely(notify_page_fault(regs, vec)))
431		return;
432
433	/* Only enable interrupts if they were on before the fault */
434	if ((regs->sr & SR_IMASK) != SR_IMASK)
435		local_irq_enable();
436
437	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
438
439	/*
440	 * If we're in an interrupt, have no user context or are running
441	 * in an atomic region then we must not take the fault:
442	 */
443	if (unlikely(in_atomic() || !mm)) {
444		bad_area_nosemaphore(regs, error_code, address);
445		return;
446	}
447
448retry:
449	down_read(&mm->mmap_sem);
450
451	vma = find_vma(mm, address);
452	if (unlikely(!vma)) {
453		bad_area(regs, error_code, address);
454		return;
455	}
456	if (likely(vma->vm_start <= address))
457		goto good_area;
458	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
459		bad_area(regs, error_code, address);
460		return;
461	}
462	if (unlikely(expand_stack(vma, address))) {
463		bad_area(regs, error_code, address);
464		return;
465	}
466
467	/*
468	 * Ok, we have a good vm_area for this memory access, so
469	 * we can handle it..
470	 */
471good_area:
472	if (unlikely(access_error(error_code, vma))) {
473		bad_area_access_error(regs, error_code, address);
474		return;
475	}
476
477	set_thread_fault_code(error_code);
478
479	if (user_mode(regs))
480		flags |= FAULT_FLAG_USER;
481	if (error_code & FAULT_CODE_WRITE)
482		flags |= FAULT_FLAG_WRITE;
483
484	/*
485	 * If for any reason at all we couldn't handle the fault,
486	 * make sure we exit gracefully rather than endlessly redo
487	 * the fault.
488	 */
489	fault = handle_mm_fault(mm, vma, address, flags);
490
491	if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
492		if (mm_fault_error(regs, error_code, address, fault))
493			return;
494
495	if (flags & FAULT_FLAG_ALLOW_RETRY) {
496		if (fault & VM_FAULT_MAJOR) {
497			tsk->maj_flt++;
498			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
499				      regs, address);
500		} else {
501			tsk->min_flt++;
502			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
503				      regs, address);
504		}
505		if (fault & VM_FAULT_RETRY) {
506			flags &= ~FAULT_FLAG_ALLOW_RETRY;
507			flags |= FAULT_FLAG_TRIED;
508
509			/*
510			 * No need to up_read(&mm->mmap_sem) as we would
511			 * have already released it in __lock_page_or_retry
512			 * in mm/filemap.c.
513			 */
514			goto retry;
515		}
516	}
517
518	up_read(&mm->mmap_sem);
519}
520