1/*
2 * OpenRISC fault.c
3 *
4 * Linux architectural port borrowing liberally from similar works of
5 * others.  All original copyrights apply as per the original source
6 * declaration.
7 *
8 * Modifications for the OpenRISC architecture:
9 * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
10 * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
11 *
12 *      This program is free software; you can redistribute it and/or
13 *      modify it under the terms of the GNU General Public License
14 *      as published by the Free Software Foundation; either version
15 *      2 of the License, or (at your option) any later version.
16 */
17
18#include <linux/mm.h>
19#include <linux/interrupt.h>
20#include <linux/module.h>
21#include <linux/sched.h>
22
23#include <asm/uaccess.h>
24#include <asm/siginfo.h>
25#include <asm/signal.h>
26
27#define NUM_TLB_ENTRIES 64
28#define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
29
30unsigned long pte_misses;	/* updated by do_page_fault() */
31unsigned long pte_errors;	/* updated by do_page_fault() */
32
33/* __PHX__ :: - check the vmalloc_fault in do_page_fault()
34 *            - also look into include/asm-or32/mmu_context.h
35 */
36volatile pgd_t *current_pgd;
37
38extern void die(char *, struct pt_regs *, long);
39
40/*
41 * This routine handles page faults.  It determines the address,
42 * and the problem, and then passes it off to one of the appropriate
43 * routines.
44 *
45 * If this routine detects a bad access, it returns 1, otherwise it
46 * returns 0.
47 */
48
49asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
50			      unsigned long vector, int write_acc)
51{
52	struct task_struct *tsk;
53	struct mm_struct *mm;
54	struct vm_area_struct *vma;
55	siginfo_t info;
56	int fault;
57	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
58
59	tsk = current;
60
61	/*
62	 * We fault-in kernel-space virtual memory on-demand. The
63	 * 'reference' page table is init_mm.pgd.
64	 *
65	 * NOTE! We MUST NOT take any locks for this case. We may
66	 * be in an interrupt or a critical region, and should
67	 * only copy the information from the master page table,
68	 * nothing more.
69	 *
70	 * NOTE2: This is done so that, when updating the vmalloc
71	 * mappings we don't have to walk all processes pgdirs and
72	 * add the high mappings all at once. Instead we do it as they
73	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
74	 * bit set so sometimes the TLB can use a lingering entry.
75	 *
76	 * This verifies that the fault happens in kernel space
77	 * and that the fault was not a protection error.
78	 */
79
80	if (address >= VMALLOC_START &&
81	    (vector != 0x300 && vector != 0x400) &&
82	    !user_mode(regs))
83		goto vmalloc_fault;
84
85	/* If exceptions were enabled, we can reenable them here */
86	if (user_mode(regs)) {
87		/* Exception was in userspace: reenable interrupts */
88		local_irq_enable();
89		flags |= FAULT_FLAG_USER;
90	} else {
91		/* If exception was in a syscall, then IRQ's may have
92		 * been enabled or disabled.  If they were enabled,
93		 * reenable them.
94		 */
95		if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
96			local_irq_enable();
97	}
98
99	mm = tsk->mm;
100	info.si_code = SEGV_MAPERR;
101
102	/*
103	 * If we're in an interrupt or have no user
104	 * context, we must not take the fault..
105	 */
106
107	if (in_interrupt() || !mm)
108		goto no_context;
109
110retry:
111	down_read(&mm->mmap_sem);
112	vma = find_vma(mm, address);
113
114	if (!vma)
115		goto bad_area;
116
117	if (vma->vm_start <= address)
118		goto good_area;
119
120	if (!(vma->vm_flags & VM_GROWSDOWN))
121		goto bad_area;
122
123	if (user_mode(regs)) {
124		/*
125		 * accessing the stack below usp is always a bug.
126		 * we get page-aligned addresses so we can only check
127		 * if we're within a page from usp, but that might be
128		 * enough to catch brutal errors at least.
129		 */
130		if (address + PAGE_SIZE < regs->sp)
131			goto bad_area;
132	}
133	if (expand_stack(vma, address))
134		goto bad_area;
135
136	/*
137	 * Ok, we have a good vm_area for this memory access, so
138	 * we can handle it..
139	 */
140
141good_area:
142	info.si_code = SEGV_ACCERR;
143
144	/* first do some preliminary protection checks */
145
146	if (write_acc) {
147		if (!(vma->vm_flags & VM_WRITE))
148			goto bad_area;
149		flags |= FAULT_FLAG_WRITE;
150	} else {
151		/* not present */
152		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
153			goto bad_area;
154	}
155
156	/* are we trying to execute nonexecutable area */
157	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
158		goto bad_area;
159
160	/*
161	 * If for any reason at all we couldn't handle the fault,
162	 * make sure we exit gracefully rather than endlessly redo
163	 * the fault.
164	 */
165
166	fault = handle_mm_fault(mm, vma, address, flags);
167
168	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
169		return;
170
171	if (unlikely(fault & VM_FAULT_ERROR)) {
172		if (fault & VM_FAULT_OOM)
173			goto out_of_memory;
174		else if (fault & VM_FAULT_SIGBUS)
175			goto do_sigbus;
176		BUG();
177	}
178
179	if (flags & FAULT_FLAG_ALLOW_RETRY) {
180		/*RGD modeled on Cris */
181		if (fault & VM_FAULT_MAJOR)
182			tsk->maj_flt++;
183		else
184			tsk->min_flt++;
185		if (fault & VM_FAULT_RETRY) {
186			flags &= ~FAULT_FLAG_ALLOW_RETRY;
187			flags |= FAULT_FLAG_TRIED;
188
189			 /* No need to up_read(&mm->mmap_sem) as we would
190			 * have already released it in __lock_page_or_retry
191			 * in mm/filemap.c.
192			 */
193
194			goto retry;
195		}
196	}
197
198	up_read(&mm->mmap_sem);
199	return;
200
201	/*
202	 * Something tried to access memory that isn't in our memory map..
203	 * Fix it, but check if it's kernel or user first..
204	 */
205
206bad_area:
207	up_read(&mm->mmap_sem);
208
209bad_area_nosemaphore:
210
211	/* User mode accesses just cause a SIGSEGV */
212
213	if (user_mode(regs)) {
214		info.si_signo = SIGSEGV;
215		info.si_errno = 0;
216		/* info.si_code has been set above */
217		info.si_addr = (void *)address;
218		force_sig_info(SIGSEGV, &info, tsk);
219		return;
220	}
221
222no_context:
223
224	/* Are we prepared to handle this kernel fault?
225	 *
226	 * (The kernel has valid exception-points in the source
227	 *  when it acesses user-memory. When it fails in one
228	 *  of those points, we find it in a table and do a jump
229	 *  to some fixup code that loads an appropriate error
230	 *  code)
231	 */
232
233	{
234		const struct exception_table_entry *entry;
235
236		__asm__ __volatile__("l.nop 42");
237
238		if ((entry = search_exception_tables(regs->pc)) != NULL) {
239			/* Adjust the instruction pointer in the stackframe */
240			regs->pc = entry->fixup;
241			return;
242		}
243	}
244
245	/*
246	 * Oops. The kernel tried to access some bad page. We'll have to
247	 * terminate things with extreme prejudice.
248	 */
249
250	if ((unsigned long)(address) < PAGE_SIZE)
251		printk(KERN_ALERT
252		       "Unable to handle kernel NULL pointer dereference");
253	else
254		printk(KERN_ALERT "Unable to handle kernel access");
255	printk(" at virtual address 0x%08lx\n", address);
256
257	die("Oops", regs, write_acc);
258
259	do_exit(SIGKILL);
260
261	/*
262	 * We ran out of memory, or some other thing happened to us that made
263	 * us unable to handle the page fault gracefully.
264	 */
265
266out_of_memory:
267	__asm__ __volatile__("l.nop 42");
268	__asm__ __volatile__("l.nop 1");
269
270	up_read(&mm->mmap_sem);
271	if (!user_mode(regs))
272		goto no_context;
273	pagefault_out_of_memory();
274	return;
275
276do_sigbus:
277	up_read(&mm->mmap_sem);
278
279	/*
280	 * Send a sigbus, regardless of whether we were in kernel
281	 * or user mode.
282	 */
283	info.si_signo = SIGBUS;
284	info.si_errno = 0;
285	info.si_code = BUS_ADRERR;
286	info.si_addr = (void *)address;
287	force_sig_info(SIGBUS, &info, tsk);
288
289	/* Kernel mode? Handle exceptions or die */
290	if (!user_mode(regs))
291		goto no_context;
292	return;
293
294vmalloc_fault:
295	{
296		/*
297		 * Synchronize this task's top level page-table
298		 * with the 'reference' page table.
299		 *
300		 * Use current_pgd instead of tsk->active_mm->pgd
301		 * since the latter might be unavailable if this
302		 * code is executed in a misfortunately run irq
303		 * (like inside schedule() between switch_mm and
304		 *  switch_to...).
305		 */
306
307		int offset = pgd_index(address);
308		pgd_t *pgd, *pgd_k;
309		pud_t *pud, *pud_k;
310		pmd_t *pmd, *pmd_k;
311		pte_t *pte_k;
312
313/*
314		phx_warn("do_page_fault(): vmalloc_fault will not work, "
315			 "since current_pgd assign a proper value somewhere\n"
316			 "anyhow we don't need this at the moment\n");
317
318		phx_mmu("vmalloc_fault");
319*/
320		pgd = (pgd_t *)current_pgd + offset;
321		pgd_k = init_mm.pgd + offset;
322
323		/* Since we're two-level, we don't need to do both
324		 * set_pgd and set_pmd (they do the same thing). If
325		 * we go three-level at some point, do the right thing
326		 * with pgd_present and set_pgd here.
327		 *
328		 * Also, since the vmalloc area is global, we don't
329		 * need to copy individual PTE's, it is enough to
330		 * copy the pgd pointer into the pte page of the
331		 * root task. If that is there, we'll find our pte if
332		 * it exists.
333		 */
334
335		pud = pud_offset(pgd, address);
336		pud_k = pud_offset(pgd_k, address);
337		if (!pud_present(*pud_k))
338			goto no_context;
339
340		pmd = pmd_offset(pud, address);
341		pmd_k = pmd_offset(pud_k, address);
342
343		if (!pmd_present(*pmd_k))
344			goto bad_area_nosemaphore;
345
346		set_pmd(pmd, *pmd_k);
347
348		/* Make sure the actual PTE exists as well to
349		 * catch kernel vmalloc-area accesses to non-mapped
350		 * addresses. If we don't do this, this will just
351		 * silently loop forever.
352		 */
353
354		pte_k = pte_offset_kernel(pmd_k, address);
355		if (!pte_present(*pte_k))
356			goto no_context;
357
358		return;
359	}
360}
361