1/* MN10300 MMU Fault handler
2 *
3 * Copyright (C) 2007 Matsushita Electric Industrial Co., Ltd.
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Modified by David Howells (dhowells@redhat.com)
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public Licence
9 * as published by the Free Software Foundation; either version
10 * 2 of the Licence, or (at your option) any later version.
11 */
12
13#include <linux/signal.h>
14#include <linux/sched.h>
15#include <linux/kernel.h>
16#include <linux/errno.h>
17#include <linux/string.h>
18#include <linux/types.h>
19#include <linux/ptrace.h>
20#include <linux/mman.h>
21#include <linux/mm.h>
22#include <linux/smp.h>
23#include <linux/interrupt.h>
24#include <linux/init.h>
25#include <linux/vt_kern.h>		/* For unblank_screen() */
26
27#include <asm/uaccess.h>
28#include <asm/pgalloc.h>
29#include <asm/hardirq.h>
30#include <asm/cpu-regs.h>
31#include <asm/debugger.h>
32#include <asm/gdb-stub.h>
33
34/*
35 * Unlock any spinlocks which will prevent us from getting the
36 * message out
37 */
38void bust_spinlocks(int yes)
39{
40	if (yes) {
41		oops_in_progress = 1;
42	} else {
43		int loglevel_save = console_loglevel;
44#ifdef CONFIG_VT
45		unblank_screen();
46#endif
47		oops_in_progress = 0;
48		/*
49		 * OK, the message is on the console.  Now we call printk()
50		 * without oops_in_progress set so that printk will give klogd
51		 * a poke.  Hold onto your hats...
52		 */
53		console_loglevel = 15;	/* NMI oopser may have shut the console
54					 * up */
55		printk(" ");
56		console_loglevel = loglevel_save;
57	}
58}
59
60void do_BUG(const char *file, int line)
61{
62	bust_spinlocks(1);
63	printk(KERN_EMERG "------------[ cut here ]------------\n");
64	printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
65}
66
67#if 0
68static void print_pagetable_entries(pgd_t *pgdir, unsigned long address)
69{
70	pgd_t *pgd;
71	pmd_t *pmd;
72	pte_t *pte;
73
74	pgd = pgdir + __pgd_offset(address);
75	printk(KERN_DEBUG "pgd entry %p: %016Lx\n",
76	       pgd, (long long) pgd_val(*pgd));
77
78	if (!pgd_present(*pgd)) {
79		printk(KERN_DEBUG "... pgd not present!\n");
80		return;
81	}
82	pmd = pmd_offset(pgd, address);
83	printk(KERN_DEBUG "pmd entry %p: %016Lx\n",
84	       pmd, (long long)pmd_val(*pmd));
85
86	if (!pmd_present(*pmd)) {
87		printk(KERN_DEBUG "... pmd not present!\n");
88		return;
89	}
90	pte = pte_offset(pmd, address);
91	printk(KERN_DEBUG "pte entry %p: %016Lx\n",
92	       pte, (long long) pte_val(*pte));
93
94	if (!pte_present(*pte))
95		printk(KERN_DEBUG "... pte not present!\n");
96}
97#endif
98
99/*
100 * This routine handles page faults.  It determines the address,
101 * and the problem, and then passes it off to one of the appropriate
102 * routines.
103 *
104 * fault_code:
105 * - LSW: either MMUFCR_IFC or MMUFCR_DFC as appropriate
106 * - MSW: 0 if data access, 1 if instruction access
107 * - bit 0: TLB miss flag
108 * - bit 1: initial write
109 * - bit 2: page invalid
110 * - bit 3: protection violation
111 * - bit 4: accessor (0=user 1=kernel)
112 * - bit 5: 0=read 1=write
113 * - bit 6-8: page protection spec
114 * - bit 9: illegal address
115 * - bit 16: 0=data 1=ins
116 *
117 */
118asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long fault_code,
119			      unsigned long address)
120{
121	struct vm_area_struct *vma;
122	struct task_struct *tsk;
123	struct mm_struct *mm;
124	unsigned long page;
125	siginfo_t info;
126	int fault;
127	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
128
129#ifdef CONFIG_GDBSTUB
130	/* handle GDB stub causing a fault */
131	if (gdbstub_busy) {
132		gdbstub_exception(regs, TBR & TBR_INT_CODE);
133		return;
134	}
135#endif
136
137#if 0
138	printk(KERN_DEBUG "--- do_page_fault(%p,%s:%04lx,%08lx)\n",
139	       regs,
140	       fault_code & 0x10000 ? "ins" : "data",
141	       fault_code & 0xffff, address);
142#endif
143
144	tsk = current;
145
146	/*
147	 * We fault-in kernel-space virtual memory on-demand. The
148	 * 'reference' page table is init_mm.pgd.
149	 *
150	 * NOTE! We MUST NOT take any locks for this case. We may
151	 * be in an interrupt or a critical region, and should
152	 * only copy the information from the master page table,
153	 * nothing more.
154	 *
155	 * This verifies that the fault happens in kernel space
156	 * and that the fault was a page not present (invalid) error
157	 */
158	if (address >= VMALLOC_START && address < VMALLOC_END &&
159	    (fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR &&
160	    (fault_code & MMUFCR_xFC_PGINVAL) == MMUFCR_xFC_PGINVAL
161	    )
162		goto vmalloc_fault;
163
164	mm = tsk->mm;
165	info.si_code = SEGV_MAPERR;
166
167	/*
168	 * If we're in an interrupt or have no user
169	 * context, we must not take the fault..
170	 */
171	if (in_atomic() || !mm)
172		goto no_context;
173
174	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR)
175		flags |= FAULT_FLAG_USER;
176retry:
177	down_read(&mm->mmap_sem);
178
179	vma = find_vma(mm, address);
180	if (!vma)
181		goto bad_area;
182	if (vma->vm_start <= address)
183		goto good_area;
184	if (!(vma->vm_flags & VM_GROWSDOWN))
185		goto bad_area;
186
187	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
188		/* accessing the stack below the stack pointer is always a
189		 * bug */
190		if ((address & PAGE_MASK) + 2 * PAGE_SIZE < regs->sp) {
191#if 0
192			printk(KERN_WARNING
193			       "[%d] ### Access below stack @%lx (sp=%lx)\n",
194			       current->pid, address, regs->sp);
195			printk(KERN_WARNING
196			       "vma [%08x - %08x]\n",
197			       vma->vm_start, vma->vm_end);
198			show_registers(regs);
199			printk(KERN_WARNING
200			       "[%d] ### Code: [%08lx]"
201			       " %02x %02x %02x %02x %02x %02x %02x %02x\n",
202			       current->pid,
203			       regs->pc,
204			       ((u8 *) regs->pc)[0],
205			       ((u8 *) regs->pc)[1],
206			       ((u8 *) regs->pc)[2],
207			       ((u8 *) regs->pc)[3],
208			       ((u8 *) regs->pc)[4],
209			       ((u8 *) regs->pc)[5],
210			       ((u8 *) regs->pc)[6],
211			       ((u8 *) regs->pc)[7]
212			       );
213#endif
214			goto bad_area;
215		}
216	}
217
218	if (expand_stack(vma, address))
219		goto bad_area;
220
221/*
222 * Ok, we have a good vm_area for this memory access, so
223 * we can handle it..
224 */
225good_area:
226	info.si_code = SEGV_ACCERR;
227	switch (fault_code & (MMUFCR_xFC_PGINVAL|MMUFCR_xFC_TYPE)) {
228	default:	/* 3: write, present */
229	case MMUFCR_xFC_TYPE_WRITE:
230#ifdef TEST_VERIFY_AREA
231		if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR)
232			printk(KERN_DEBUG "WP fault at %08lx\n", regs->pc);
233#endif
234		/* write to absent page */
235	case MMUFCR_xFC_PGINVAL | MMUFCR_xFC_TYPE_WRITE:
236		if (!(vma->vm_flags & VM_WRITE))
237			goto bad_area;
238		flags |= FAULT_FLAG_WRITE;
239		break;
240
241		/* read from protected page */
242	case MMUFCR_xFC_TYPE_READ:
243		goto bad_area;
244
245		/* read from absent page present */
246	case MMUFCR_xFC_PGINVAL | MMUFCR_xFC_TYPE_READ:
247		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
248			goto bad_area;
249		break;
250	}
251
252	/*
253	 * If for any reason at all we couldn't handle the fault,
254	 * make sure we exit gracefully rather than endlessly redo
255	 * the fault.
256	 */
257	fault = handle_mm_fault(mm, vma, address, flags);
258
259	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
260		return;
261
262	if (unlikely(fault & VM_FAULT_ERROR)) {
263		if (fault & VM_FAULT_OOM)
264			goto out_of_memory;
265		else if (fault & VM_FAULT_SIGSEGV)
266			goto bad_area;
267		else if (fault & VM_FAULT_SIGBUS)
268			goto do_sigbus;
269		BUG();
270	}
271	if (flags & FAULT_FLAG_ALLOW_RETRY) {
272		if (fault & VM_FAULT_MAJOR)
273			current->maj_flt++;
274		else
275			current->min_flt++;
276		if (fault & VM_FAULT_RETRY) {
277			flags &= ~FAULT_FLAG_ALLOW_RETRY;
278
279			 /* No need to up_read(&mm->mmap_sem) as we would
280			 * have already released it in __lock_page_or_retry
281			 * in mm/filemap.c.
282			 */
283
284			goto retry;
285		}
286	}
287
288	up_read(&mm->mmap_sem);
289	return;
290
291/*
292 * Something tried to access memory that isn't in our memory map..
293 * Fix it, but check if it's kernel or user first..
294 */
295bad_area:
296	up_read(&mm->mmap_sem);
297
298	/* User mode accesses just cause a SIGSEGV */
299	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
300		info.si_signo = SIGSEGV;
301		info.si_errno = 0;
302		/* info.si_code has been set above */
303		info.si_addr = (void *)address;
304		force_sig_info(SIGSEGV, &info, tsk);
305		return;
306	}
307
308no_context:
309	/* Are we prepared to handle this kernel fault?  */
310	if (fixup_exception(regs))
311		return;
312
313/*
314 * Oops. The kernel tried to access some bad page. We'll have to
315 * terminate things with extreme prejudice.
316 */
317
318	bust_spinlocks(1);
319
320	if (address < PAGE_SIZE)
321		printk(KERN_ALERT
322		       "Unable to handle kernel NULL pointer dereference");
323	else
324		printk(KERN_ALERT
325		       "Unable to handle kernel paging request");
326	printk(" at virtual address %08lx\n", address);
327	printk(" printing pc:\n");
328	printk(KERN_ALERT "%08lx\n", regs->pc);
329
330	debugger_intercept(fault_code & 0x00010000 ? EXCEP_IAERROR : EXCEP_DAERROR,
331			   SIGSEGV, SEGV_ACCERR, regs);
332
333	page = PTBR;
334	page = ((unsigned long *) __va(page))[address >> 22];
335	printk(KERN_ALERT "*pde = %08lx\n", page);
336	if (page & 1) {
337		page &= PAGE_MASK;
338		address &= 0x003ff000;
339		page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
340		printk(KERN_ALERT "*pte = %08lx\n", page);
341	}
342
343	die("Oops", regs, fault_code);
344	do_exit(SIGKILL);
345
346/*
347 * We ran out of memory, or some other thing happened to us that made
348 * us unable to handle the page fault gracefully.
349 */
350out_of_memory:
351	up_read(&mm->mmap_sem);
352	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
353		pagefault_out_of_memory();
354		return;
355	}
356	goto no_context;
357
358do_sigbus:
359	up_read(&mm->mmap_sem);
360
361	/*
362	 * Send a sigbus, regardless of whether we were in kernel
363	 * or user mode.
364	 */
365	info.si_signo = SIGBUS;
366	info.si_errno = 0;
367	info.si_code = BUS_ADRERR;
368	info.si_addr = (void *)address;
369	force_sig_info(SIGBUS, &info, tsk);
370
371	/* Kernel mode? Handle exceptions or die */
372	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR)
373		goto no_context;
374	return;
375
376vmalloc_fault:
377	{
378		/*
379		 * Synchronize this task's top level page-table
380		 * with the 'reference' page table.
381		 *
382		 * Do _not_ use "tsk" here. We might be inside
383		 * an interrupt in the middle of a task switch..
384		 */
385		int index = pgd_index(address);
386		pgd_t *pgd, *pgd_k;
387		pud_t *pud, *pud_k;
388		pmd_t *pmd, *pmd_k;
389		pte_t *pte_k;
390
391		pgd_k = init_mm.pgd + index;
392
393		if (!pgd_present(*pgd_k))
394			goto no_context;
395
396		pud_k = pud_offset(pgd_k, address);
397		if (!pud_present(*pud_k))
398			goto no_context;
399
400		pmd_k = pmd_offset(pud_k, address);
401		if (!pmd_present(*pmd_k))
402			goto no_context;
403
404		pgd = (pgd_t *) PTBR + index;
405		pud = pud_offset(pgd, address);
406		pmd = pmd_offset(pud, address);
407		set_pmd(pmd, *pmd_k);
408
409		pte_k = pte_offset_kernel(pmd_k, address);
410		if (!pte_present(*pte_k))
411			goto no_context;
412		return;
413	}
414}
415