xref: /openbmc/linux/arch/openrisc/mm/fault.c (revision 827634ad)
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 
30 unsigned long pte_misses;	/* updated by do_page_fault() */
31 unsigned 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  */
36 volatile pgd_t *current_pgd;
37 
38 extern 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 
49 asmlinkage 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 
110 retry:
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 
141 good_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_SIGSEGV)
175 			goto bad_area;
176 		else if (fault & VM_FAULT_SIGBUS)
177 			goto do_sigbus;
178 		BUG();
179 	}
180 
181 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
182 		/*RGD modeled on Cris */
183 		if (fault & VM_FAULT_MAJOR)
184 			tsk->maj_flt++;
185 		else
186 			tsk->min_flt++;
187 		if (fault & VM_FAULT_RETRY) {
188 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
189 			flags |= FAULT_FLAG_TRIED;
190 
191 			 /* No need to up_read(&mm->mmap_sem) as we would
192 			 * have already released it in __lock_page_or_retry
193 			 * in mm/filemap.c.
194 			 */
195 
196 			goto retry;
197 		}
198 	}
199 
200 	up_read(&mm->mmap_sem);
201 	return;
202 
203 	/*
204 	 * Something tried to access memory that isn't in our memory map..
205 	 * Fix it, but check if it's kernel or user first..
206 	 */
207 
208 bad_area:
209 	up_read(&mm->mmap_sem);
210 
211 bad_area_nosemaphore:
212 
213 	/* User mode accesses just cause a SIGSEGV */
214 
215 	if (user_mode(regs)) {
216 		info.si_signo = SIGSEGV;
217 		info.si_errno = 0;
218 		/* info.si_code has been set above */
219 		info.si_addr = (void *)address;
220 		force_sig_info(SIGSEGV, &info, tsk);
221 		return;
222 	}
223 
224 no_context:
225 
226 	/* Are we prepared to handle this kernel fault?
227 	 *
228 	 * (The kernel has valid exception-points in the source
229 	 *  when it acesses user-memory. When it fails in one
230 	 *  of those points, we find it in a table and do a jump
231 	 *  to some fixup code that loads an appropriate error
232 	 *  code)
233 	 */
234 
235 	{
236 		const struct exception_table_entry *entry;
237 
238 		__asm__ __volatile__("l.nop 42");
239 
240 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
241 			/* Adjust the instruction pointer in the stackframe */
242 			regs->pc = entry->fixup;
243 			return;
244 		}
245 	}
246 
247 	/*
248 	 * Oops. The kernel tried to access some bad page. We'll have to
249 	 * terminate things with extreme prejudice.
250 	 */
251 
252 	if ((unsigned long)(address) < PAGE_SIZE)
253 		printk(KERN_ALERT
254 		       "Unable to handle kernel NULL pointer dereference");
255 	else
256 		printk(KERN_ALERT "Unable to handle kernel access");
257 	printk(" at virtual address 0x%08lx\n", address);
258 
259 	die("Oops", regs, write_acc);
260 
261 	do_exit(SIGKILL);
262 
263 	/*
264 	 * We ran out of memory, or some other thing happened to us that made
265 	 * us unable to handle the page fault gracefully.
266 	 */
267 
268 out_of_memory:
269 	__asm__ __volatile__("l.nop 42");
270 	__asm__ __volatile__("l.nop 1");
271 
272 	up_read(&mm->mmap_sem);
273 	if (!user_mode(regs))
274 		goto no_context;
275 	pagefault_out_of_memory();
276 	return;
277 
278 do_sigbus:
279 	up_read(&mm->mmap_sem);
280 
281 	/*
282 	 * Send a sigbus, regardless of whether we were in kernel
283 	 * or user mode.
284 	 */
285 	info.si_signo = SIGBUS;
286 	info.si_errno = 0;
287 	info.si_code = BUS_ADRERR;
288 	info.si_addr = (void *)address;
289 	force_sig_info(SIGBUS, &info, tsk);
290 
291 	/* Kernel mode? Handle exceptions or die */
292 	if (!user_mode(regs))
293 		goto no_context;
294 	return;
295 
296 vmalloc_fault:
297 	{
298 		/*
299 		 * Synchronize this task's top level page-table
300 		 * with the 'reference' page table.
301 		 *
302 		 * Use current_pgd instead of tsk->active_mm->pgd
303 		 * since the latter might be unavailable if this
304 		 * code is executed in a misfortunately run irq
305 		 * (like inside schedule() between switch_mm and
306 		 *  switch_to...).
307 		 */
308 
309 		int offset = pgd_index(address);
310 		pgd_t *pgd, *pgd_k;
311 		pud_t *pud, *pud_k;
312 		pmd_t *pmd, *pmd_k;
313 		pte_t *pte_k;
314 
315 /*
316 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
317 			 "since current_pgd assign a proper value somewhere\n"
318 			 "anyhow we don't need this at the moment\n");
319 
320 		phx_mmu("vmalloc_fault");
321 */
322 		pgd = (pgd_t *)current_pgd + offset;
323 		pgd_k = init_mm.pgd + offset;
324 
325 		/* Since we're two-level, we don't need to do both
326 		 * set_pgd and set_pmd (they do the same thing). If
327 		 * we go three-level at some point, do the right thing
328 		 * with pgd_present and set_pgd here.
329 		 *
330 		 * Also, since the vmalloc area is global, we don't
331 		 * need to copy individual PTE's, it is enough to
332 		 * copy the pgd pointer into the pte page of the
333 		 * root task. If that is there, we'll find our pte if
334 		 * it exists.
335 		 */
336 
337 		pud = pud_offset(pgd, address);
338 		pud_k = pud_offset(pgd_k, address);
339 		if (!pud_present(*pud_k))
340 			goto no_context;
341 
342 		pmd = pmd_offset(pud, address);
343 		pmd_k = pmd_offset(pud_k, address);
344 
345 		if (!pmd_present(*pmd_k))
346 			goto bad_area_nosemaphore;
347 
348 		set_pmd(pmd, *pmd_k);
349 
350 		/* Make sure the actual PTE exists as well to
351 		 * catch kernel vmalloc-area accesses to non-mapped
352 		 * addresses. If we don't do this, this will just
353 		 * silently loop forever.
354 		 */
355 
356 		pte_k = pte_offset_kernel(pmd_k, address);
357 		if (!pte_present(*pte_k))
358 			goto no_context;
359 
360 		return;
361 	}
362 }
363