1/*
2 * linux/arch/unicore32/mm/fault.c
3 *
4 * Code specific to PKUnity SoC and UniCore ISA
5 *
6 * Copyright (C) 2001-2010 GUAN Xue-tao
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 */
12#include <linux/module.h>
13#include <linux/signal.h>
14#include <linux/mm.h>
15#include <linux/hardirq.h>
16#include <linux/init.h>
17#include <linux/kprobes.h>
18#include <linux/uaccess.h>
19#include <linux/page-flags.h>
20#include <linux/sched.h>
21#include <linux/io.h>
22
23#include <asm/pgtable.h>
24#include <asm/tlbflush.h>
25
26/*
27 * Fault status register encodings.  We steal bit 31 for our own purposes.
28 */
29#define FSR_LNX_PF		(1 << 31)
30
31static inline int fsr_fs(unsigned int fsr)
32{
33	/* xyabcde will be abcde+xy */
34	return (fsr & 31) + ((fsr & (3 << 5)) >> 5);
35}
36
37/*
38 * This is useful to dump out the page tables associated with
39 * 'addr' in mm 'mm'.
40 */
41void show_pte(struct mm_struct *mm, unsigned long addr)
42{
43	pgd_t *pgd;
44
45	if (!mm)
46		mm = &init_mm;
47
48	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
49	pgd = pgd_offset(mm, addr);
50	printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
51
52	do {
53		pmd_t *pmd;
54		pte_t *pte;
55
56		if (pgd_none(*pgd))
57			break;
58
59		if (pgd_bad(*pgd)) {
60			printk("(bad)");
61			break;
62		}
63
64		pmd = pmd_offset((pud_t *) pgd, addr);
65		if (PTRS_PER_PMD != 1)
66			printk(", *pmd=%08lx", pmd_val(*pmd));
67
68		if (pmd_none(*pmd))
69			break;
70
71		if (pmd_bad(*pmd)) {
72			printk("(bad)");
73			break;
74		}
75
76		/* We must not map this if we have highmem enabled */
77		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
78			break;
79
80		pte = pte_offset_map(pmd, addr);
81		printk(", *pte=%08lx", pte_val(*pte));
82		pte_unmap(pte);
83	} while (0);
84
85	printk("\n");
86}
87
88/*
89 * Oops.  The kernel tried to access some page that wasn't present.
90 */
91static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
92		unsigned int fsr, struct pt_regs *regs)
93{
94	/*
95	 * Are we prepared to handle this kernel fault?
96	 */
97	if (fixup_exception(regs))
98		return;
99
100	/*
101	 * No handler, we'll have to terminate things with extreme prejudice.
102	 */
103	bust_spinlocks(1);
104	printk(KERN_ALERT
105	       "Unable to handle kernel %s at virtual address %08lx\n",
106	       (addr < PAGE_SIZE) ? "NULL pointer dereference" :
107	       "paging request", addr);
108
109	show_pte(mm, addr);
110	die("Oops", regs, fsr);
111	bust_spinlocks(0);
112	do_exit(SIGKILL);
113}
114
115/*
116 * Something tried to access memory that isn't in our memory map..
117 * User mode accesses just cause a SIGSEGV
118 */
119static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
120		unsigned int fsr, unsigned int sig, int code,
121		struct pt_regs *regs)
122{
123	struct siginfo si;
124
125	tsk->thread.address = addr;
126	tsk->thread.error_code = fsr;
127	tsk->thread.trap_no = 14;
128	si.si_signo = sig;
129	si.si_errno = 0;
130	si.si_code = code;
131	si.si_addr = (void __user *)addr;
132	force_sig_info(sig, &si, tsk);
133}
134
135void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
136{
137	struct task_struct *tsk = current;
138	struct mm_struct *mm = tsk->active_mm;
139
140	/*
141	 * If we are in kernel mode at this point, we
142	 * have no context to handle this fault with.
143	 */
144	if (user_mode(regs))
145		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
146	else
147		__do_kernel_fault(mm, addr, fsr, regs);
148}
149
150#define VM_FAULT_BADMAP		0x010000
151#define VM_FAULT_BADACCESS	0x020000
152
153/*
154 * Check that the permissions on the VMA allow for the fault which occurred.
155 * If we encountered a write fault, we must have write permission, otherwise
156 * we allow any permission.
157 */
158static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
159{
160	unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
161
162	if (!(fsr ^ 0x12))	/* write? */
163		mask = VM_WRITE;
164	if (fsr & FSR_LNX_PF)
165		mask = VM_EXEC;
166
167	return vma->vm_flags & mask ? false : true;
168}
169
170static int __do_pf(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
171		unsigned int flags, struct task_struct *tsk)
172{
173	struct vm_area_struct *vma;
174	int fault;
175
176	vma = find_vma(mm, addr);
177	fault = VM_FAULT_BADMAP;
178	if (unlikely(!vma))
179		goto out;
180	if (unlikely(vma->vm_start > addr))
181		goto check_stack;
182
183	/*
184	 * Ok, we have a good vm_area for this
185	 * memory access, so we can handle it.
186	 */
187good_area:
188	if (access_error(fsr, vma)) {
189		fault = VM_FAULT_BADACCESS;
190		goto out;
191	}
192
193	/*
194	 * If for any reason at all we couldn't handle the fault, make
195	 * sure we exit gracefully rather than endlessly redo the fault.
196	 */
197	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
198	return fault;
199
200check_stack:
201	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
202		goto good_area;
203out:
204	return fault;
205}
206
207static int do_pf(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
208{
209	struct task_struct *tsk;
210	struct mm_struct *mm;
211	int fault, sig, code;
212	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
213
214	tsk = current;
215	mm = tsk->mm;
216
217	/*
218	 * If we're in an interrupt or have no user
219	 * context, we must not take the fault..
220	 */
221	if (in_atomic() || !mm)
222		goto no_context;
223
224	if (user_mode(regs))
225		flags |= FAULT_FLAG_USER;
226	if (!(fsr ^ 0x12))
227		flags |= FAULT_FLAG_WRITE;
228
229	/*
230	 * As per x86, we may deadlock here.  However, since the kernel only
231	 * validly references user space from well defined areas of the code,
232	 * we can bug out early if this is from code which shouldn't.
233	 */
234	if (!down_read_trylock(&mm->mmap_sem)) {
235		if (!user_mode(regs)
236		    && !search_exception_tables(regs->UCreg_pc))
237			goto no_context;
238retry:
239		down_read(&mm->mmap_sem);
240	} else {
241		/*
242		 * The above down_read_trylock() might have succeeded in
243		 * which case, we'll have missed the might_sleep() from
244		 * down_read()
245		 */
246		might_sleep();
247#ifdef CONFIG_DEBUG_VM
248		if (!user_mode(regs) &&
249		    !search_exception_tables(regs->UCreg_pc))
250			goto no_context;
251#endif
252	}
253
254	fault = __do_pf(mm, addr, fsr, flags, tsk);
255
256	/* If we need to retry but a fatal signal is pending, handle the
257	 * signal first. We do not need to release the mmap_sem because
258	 * it would already be released in __lock_page_or_retry in
259	 * mm/filemap.c. */
260	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
261		return 0;
262
263	if (!(fault & VM_FAULT_ERROR) && (flags & FAULT_FLAG_ALLOW_RETRY)) {
264		if (fault & VM_FAULT_MAJOR)
265			tsk->maj_flt++;
266		else
267			tsk->min_flt++;
268		if (fault & VM_FAULT_RETRY) {
269			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
270			* of starvation. */
271			flags &= ~FAULT_FLAG_ALLOW_RETRY;
272			goto retry;
273		}
274	}
275
276	up_read(&mm->mmap_sem);
277
278	/*
279	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
280	 */
281	if (likely(!(fault &
282	       (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
283		return 0;
284
285	/*
286	 * If we are in kernel mode at this point, we
287	 * have no context to handle this fault with.
288	 */
289	if (!user_mode(regs))
290		goto no_context;
291
292	if (fault & VM_FAULT_OOM) {
293		/*
294		 * We ran out of memory, call the OOM killer, and return to
295		 * userspace (which will retry the fault, or kill us if we
296		 * got oom-killed)
297		 */
298		pagefault_out_of_memory();
299		return 0;
300	}
301
302	if (fault & VM_FAULT_SIGBUS) {
303		/*
304		 * We had some memory, but were unable to
305		 * successfully fix up this page fault.
306		 */
307		sig = SIGBUS;
308		code = BUS_ADRERR;
309	} else {
310		/*
311		 * Something tried to access memory that
312		 * isn't in our memory map..
313		 */
314		sig = SIGSEGV;
315		code = fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR;
316	}
317
318	__do_user_fault(tsk, addr, fsr, sig, code, regs);
319	return 0;
320
321no_context:
322	__do_kernel_fault(mm, addr, fsr, regs);
323	return 0;
324}
325
326/*
327 * First Level Translation Fault Handler
328 *
329 * We enter here because the first level page table doesn't contain
330 * a valid entry for the address.
331 *
332 * If the address is in kernel space (>= TASK_SIZE), then we are
333 * probably faulting in the vmalloc() area.
334 *
335 * If the init_task's first level page tables contains the relevant
336 * entry, we copy the it to this task.  If not, we send the process
337 * a signal, fixup the exception, or oops the kernel.
338 *
339 * NOTE! We MUST NOT take any locks for this case. We may be in an
340 * interrupt or a critical region, and should only copy the information
341 * from the master page table, nothing more.
342 */
343static int do_ifault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
344{
345	unsigned int index;
346	pgd_t *pgd, *pgd_k;
347	pmd_t *pmd, *pmd_k;
348
349	if (addr < TASK_SIZE)
350		return do_pf(addr, fsr, regs);
351
352	if (user_mode(regs))
353		goto bad_area;
354
355	index = pgd_index(addr);
356
357	pgd = cpu_get_pgd() + index;
358	pgd_k = init_mm.pgd + index;
359
360	if (pgd_none(*pgd_k))
361		goto bad_area;
362
363	pmd_k = pmd_offset((pud_t *) pgd_k, addr);
364	pmd = pmd_offset((pud_t *) pgd, addr);
365
366	if (pmd_none(*pmd_k))
367		goto bad_area;
368
369	set_pmd(pmd, *pmd_k);
370	flush_pmd_entry(pmd);
371	return 0;
372
373bad_area:
374	do_bad_area(addr, fsr, regs);
375	return 0;
376}
377
378/*
379 * This abort handler always returns "fault".
380 */
381static int do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
382{
383	return 1;
384}
385
386static int do_good(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
387{
388	unsigned int res1, res2;
389
390	printk("dabt exception but no error!\n");
391
392	__asm__ __volatile__(
393			"mff %0,f0\n"
394			"mff %1,f1\n"
395			: "=r"(res1), "=r"(res2)
396			:
397			: "memory");
398
399	printk(KERN_EMERG "r0 :%08x  r1 :%08x\n", res1, res2);
400	panic("shut up\n");
401	return 0;
402}
403
404static struct fsr_info {
405	int (*fn) (unsigned long addr, unsigned int fsr, struct pt_regs *regs);
406	int sig;
407	int code;
408	const char *name;
409} fsr_info[] = {
410	/*
411	 * The following are the standard Unicore-I and UniCore-II aborts.
412	 */
413	{ do_good,	SIGBUS,  0,		"no error"		},
414	{ do_bad,	SIGBUS,  BUS_ADRALN,	"alignment exception"	},
415	{ do_bad,	SIGBUS,  BUS_OBJERR,	"external exception"	},
416	{ do_bad,	SIGBUS,  0,		"burst operation"	},
417	{ do_bad,	SIGBUS,  0,		"unknown 00100"		},
418	{ do_ifault,	SIGSEGV, SEGV_MAPERR,	"2nd level pt non-exist"},
419	{ do_bad,	SIGBUS,  0,		"2nd lvl large pt non-exist" },
420	{ do_bad,	SIGBUS,  0,		"invalid pte"		},
421	{ do_pf,	SIGSEGV, SEGV_MAPERR,	"page miss"		},
422	{ do_bad,	SIGBUS,  0,		"middle page miss"	},
423	{ do_bad,	SIGBUS,	 0,		"large page miss"	},
424	{ do_pf,	SIGSEGV, SEGV_MAPERR,	"super page (section) miss" },
425	{ do_bad,	SIGBUS,  0,		"unknown 01100"		},
426	{ do_bad,	SIGBUS,  0,		"unknown 01101"		},
427	{ do_bad,	SIGBUS,  0,		"unknown 01110"		},
428	{ do_bad,	SIGBUS,  0,		"unknown 01111"		},
429	{ do_bad,	SIGBUS,  0,		"addr: up 3G or IO"	},
430	{ do_pf,	SIGSEGV, SEGV_ACCERR,	"read unreadable addr"	},
431	{ do_pf,	SIGSEGV, SEGV_ACCERR,	"write unwriteable addr"},
432	{ do_pf,	SIGSEGV, SEGV_ACCERR,	"exec unexecutable addr"},
433	{ do_bad,	SIGBUS,  0,		"unknown 10100"		},
434	{ do_bad,	SIGBUS,  0,		"unknown 10101"		},
435	{ do_bad,	SIGBUS,  0,		"unknown 10110"		},
436	{ do_bad,	SIGBUS,  0,		"unknown 10111"		},
437	{ do_bad,	SIGBUS,  0,		"unknown 11000"		},
438	{ do_bad,	SIGBUS,  0,		"unknown 11001"		},
439	{ do_bad,	SIGBUS,  0,		"unknown 11010"		},
440	{ do_bad,	SIGBUS,  0,		"unknown 11011"		},
441	{ do_bad,	SIGBUS,  0,		"unknown 11100"		},
442	{ do_bad,	SIGBUS,  0,		"unknown 11101"		},
443	{ do_bad,	SIGBUS,  0,		"unknown 11110"		},
444	{ do_bad,	SIGBUS,  0,		"unknown 11111"		}
445};
446
447void __init hook_fault_code(int nr,
448		int (*fn) (unsigned long, unsigned int, struct pt_regs *),
449		int sig, int code, const char *name)
450{
451	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
452		BUG();
453
454	fsr_info[nr].fn   = fn;
455	fsr_info[nr].sig  = sig;
456	fsr_info[nr].code = code;
457	fsr_info[nr].name = name;
458}
459
460/*
461 * Dispatch a data abort to the relevant handler.
462 */
463asmlinkage void do_DataAbort(unsigned long addr, unsigned int fsr,
464			struct pt_regs *regs)
465{
466	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
467	struct siginfo info;
468
469	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
470		return;
471
472	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
473	       inf->name, fsr, addr);
474
475	info.si_signo = inf->sig;
476	info.si_errno = 0;
477	info.si_code = inf->code;
478	info.si_addr = (void __user *)addr;
479	uc32_notify_die("", regs, &info, fsr, 0);
480}
481
482asmlinkage void do_PrefetchAbort(unsigned long addr,
483			unsigned int ifsr, struct pt_regs *regs)
484{
485	const struct fsr_info *inf = fsr_info + fsr_fs(ifsr);
486	struct siginfo info;
487
488	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
489		return;
490
491	printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
492	       inf->name, ifsr, addr);
493
494	info.si_signo = inf->sig;
495	info.si_errno = 0;
496	info.si_code = inf->code;
497	info.si_addr = (void __user *)addr;
498	uc32_notify_die("", regs, &info, ifsr, 0);
499}
500