1/*
2 * Page fault handler for SH with an MMU.
3 *
4 *  Copyright (C) 1999  Niibe Yutaka
5 *  Copyright (C) 2003 - 2007  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 <asm/system.h>
19#include <asm/mmu_context.h>
20#include <asm/tlbflush.h>
21#include <asm/kgdb.h>
22
23/*
24 * This routine handles page faults.  It determines the address,
25 * and the problem, and then passes it off to one of the appropriate
26 * routines.
27 */
28asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
29					unsigned long writeaccess,
30					unsigned long address)
31{
32	struct task_struct *tsk;
33	struct mm_struct *mm;
34	struct vm_area_struct * vma;
35	unsigned long page;
36	int si_code;
37	siginfo_t info;
38
39	trace_hardirqs_on();
40	local_irq_enable();
41
42#ifdef CONFIG_SH_KGDB
43	if (kgdb_nofault && kgdb_bus_err_hook)
44		kgdb_bus_err_hook();
45#endif
46
47	tsk = current;
48	mm = tsk->mm;
49	si_code = SEGV_MAPERR;
50
51	if (unlikely(address >= TASK_SIZE)) {
52		/*
53		 * Synchronize this task's top level page-table
54		 * with the 'reference' page table.
55		 *
56		 * Do _not_ use "tsk" here. We might be inside
57		 * an interrupt in the middle of a task switch..
58		 */
59		int offset = pgd_index(address);
60		pgd_t *pgd, *pgd_k;
61		pud_t *pud, *pud_k;
62		pmd_t *pmd, *pmd_k;
63
64		pgd = get_TTB() + offset;
65		pgd_k = swapper_pg_dir + offset;
66
67		/* This will never happen with the folded page table. */
68		if (!pgd_present(*pgd)) {
69			if (!pgd_present(*pgd_k))
70				goto bad_area_nosemaphore;
71			set_pgd(pgd, *pgd_k);
72			return;
73		}
74
75		pud = pud_offset(pgd, address);
76		pud_k = pud_offset(pgd_k, address);
77		if (pud_present(*pud) || !pud_present(*pud_k))
78			goto bad_area_nosemaphore;
79		set_pud(pud, *pud_k);
80
81		pmd = pmd_offset(pud, address);
82		pmd_k = pmd_offset(pud_k, address);
83		if (pmd_present(*pmd) || !pmd_present(*pmd_k))
84			goto bad_area_nosemaphore;
85		set_pmd(pmd, *pmd_k);
86
87		return;
88	}
89
90	/*
91	 * If we're in an interrupt or have no user
92	 * context, we must not take the fault..
93	 */
94	if (in_atomic() || !mm)
95		goto no_context;
96
97	down_read(&mm->mmap_sem);
98
99	vma = find_vma(mm, address);
100	if (!vma)
101		goto bad_area;
102	if (vma->vm_start <= address)
103		goto good_area;
104	if (!(vma->vm_flags & VM_GROWSDOWN))
105		goto bad_area;
106	if (expand_stack(vma, address))
107		goto bad_area;
108/*
109 * Ok, we have a good vm_area for this memory access, so
110 * we can handle it..
111 */
112good_area:
113	si_code = SEGV_ACCERR;
114	if (writeaccess) {
115		if (!(vma->vm_flags & VM_WRITE))
116			goto bad_area;
117	} else {
118		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
119			goto bad_area;
120	}
121
122	/*
123	 * If for any reason at all we couldn't handle the fault,
124	 * make sure we exit gracefully rather than endlessly redo
125	 * the fault.
126	 */
127survive:
128	switch (handle_mm_fault(mm, vma, address, writeaccess)) {
129		case VM_FAULT_MINOR:
130			tsk->min_flt++;
131			break;
132		case VM_FAULT_MAJOR:
133			tsk->maj_flt++;
134			break;
135		case VM_FAULT_SIGBUS:
136			goto do_sigbus;
137		case VM_FAULT_OOM:
138			goto out_of_memory;
139		default:
140			BUG();
141	}
142
143	up_read(&mm->mmap_sem);
144	return;
145
146/*
147 * Something tried to access memory that isn't in our memory map..
148 * Fix it, but check if it's kernel or user first..
149 */
150bad_area:
151	up_read(&mm->mmap_sem);
152
153bad_area_nosemaphore:
154	if (user_mode(regs)) {
155		info.si_signo = SIGSEGV;
156		info.si_errno = 0;
157		info.si_code = si_code;
158		info.si_addr = (void *) address;
159		force_sig_info(SIGSEGV, &info, tsk);
160		return;
161	}
162
163no_context:
164	/* Are we prepared to handle this kernel fault?  */
165	if (fixup_exception(regs))
166		return;
167
168/*
169 * Oops. The kernel tried to access some bad page. We'll have to
170 * terminate things with extreme prejudice.
171 *
172 */
173	if (address < PAGE_SIZE)
174		printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
175	else
176		printk(KERN_ALERT "Unable to handle kernel paging request");
177	printk(" at virtual address %08lx\n", address);
178	printk(KERN_ALERT "pc = %08lx\n", regs->pc);
179	page = (unsigned long)get_TTB();
180	if (page) {
181		page = ((unsigned long *) page)[address >> PGDIR_SHIFT];
182		printk(KERN_ALERT "*pde = %08lx\n", page);
183		if (page & _PAGE_PRESENT) {
184			page &= PAGE_MASK;
185			address &= 0x003ff000;
186			page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
187			printk(KERN_ALERT "*pte = %08lx\n", page);
188		}
189	}
190	die("Oops", regs, writeaccess);
191	do_exit(SIGKILL);
192
193/*
194 * We ran out of memory, or some other thing happened to us that made
195 * us unable to handle the page fault gracefully.
196 */
197out_of_memory:
198	up_read(&mm->mmap_sem);
199	if (is_init(current)) {
200		yield();
201		down_read(&mm->mmap_sem);
202		goto survive;
203	}
204	printk("VM: killing process %s\n", tsk->comm);
205	if (user_mode(regs))
206		do_exit(SIGKILL);
207	goto no_context;
208
209do_sigbus:
210	up_read(&mm->mmap_sem);
211
212	/*
213	 * Send a sigbus, regardless of whether we were in kernel
214	 * or user mode.
215	 */
216	info.si_signo = SIGBUS;
217	info.si_errno = 0;
218	info.si_code = BUS_ADRERR;
219	info.si_addr = (void *)address;
220	force_sig_info(SIGBUS, &info, tsk);
221
222	/* Kernel mode? Handle exceptions or die */
223	if (!user_mode(regs))
224		goto no_context;
225}
226
227#ifdef CONFIG_SH_STORE_QUEUES
228/*
229 * This is a special case for the SH-4 store queues, as pages for this
230 * space still need to be faulted in before it's possible to flush the
231 * store queue cache for writeout to the remapped region.
232 */
233#define P3_ADDR_MAX		(P4SEG_STORE_QUE + 0x04000000)
234#else
235#define P3_ADDR_MAX		P4SEG
236#endif
237
238/*
239 * Called with interrupts disabled.
240 */
241asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
242					 unsigned long writeaccess,
243					 unsigned long address)
244{
245	pgd_t *pgd;
246	pud_t *pud;
247	pmd_t *pmd;
248	pte_t *pte;
249	pte_t entry;
250	struct mm_struct *mm = current->mm;
251	spinlock_t *ptl = NULL;
252	int ret = 1;
253
254#ifdef CONFIG_SH_KGDB
255	if (kgdb_nofault && kgdb_bus_err_hook)
256		kgdb_bus_err_hook();
257#endif
258
259	/*
260	 * We don't take page faults for P1, P2, and parts of P4, these
261	 * are always mapped, whether it be due to legacy behaviour in
262	 * 29-bit mode, or due to PMB configuration in 32-bit mode.
263	 */
264	if (address >= P3SEG && address < P3_ADDR_MAX) {
265		pgd = pgd_offset_k(address);
266		mm = NULL;
267	} else {
268		if (unlikely(address >= TASK_SIZE || !mm))
269			return 1;
270
271		pgd = pgd_offset(mm, address);
272	}
273
274	pud = pud_offset(pgd, address);
275	if (pud_none_or_clear_bad(pud))
276		return 1;
277	pmd = pmd_offset(pud, address);
278	if (pmd_none_or_clear_bad(pmd))
279		return 1;
280
281	if (mm)
282		pte = pte_offset_map_lock(mm, pmd, address, &ptl);
283	else
284		pte = pte_offset_kernel(pmd, address);
285
286	entry = *pte;
287	if (unlikely(pte_none(entry) || pte_not_present(entry)))
288		goto unlock;
289	if (unlikely(writeaccess && !pte_write(entry)))
290		goto unlock;
291
292	if (writeaccess)
293		entry = pte_mkdirty(entry);
294	entry = pte_mkyoung(entry);
295
296#ifdef CONFIG_CPU_SH4
297	/*
298	 * ITLB is not affected by "ldtlb" instruction.
299	 * So, we need to flush the entry by ourselves.
300	 */
301	local_flush_tlb_one(get_asid(), address & PAGE_MASK);
302#endif
303
304	set_pte(pte, entry);
305	update_mmu_cache(NULL, address, entry);
306	ret = 0;
307unlock:
308	if (mm)
309		pte_unmap_unlock(pte, ptl);
310	return ret;
311}
312