xref: /openbmc/linux/arch/openrisc/mm/fault.c (revision 3eb0f5193b497083391aa05d35210d5645211eef)
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/extable.h>
21 #include <linux/sched/signal.h>
22 
23 #include <linux/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[NR_CPUS];
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 	clear_siginfo(&info);
60 	tsk = current;
61 
62 	/*
63 	 * We fault-in kernel-space virtual memory on-demand. The
64 	 * 'reference' page table is init_mm.pgd.
65 	 *
66 	 * NOTE! We MUST NOT take any locks for this case. We may
67 	 * be in an interrupt or a critical region, and should
68 	 * only copy the information from the master page table,
69 	 * nothing more.
70 	 *
71 	 * NOTE2: This is done so that, when updating the vmalloc
72 	 * mappings we don't have to walk all processes pgdirs and
73 	 * add the high mappings all at once. Instead we do it as they
74 	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
75 	 * bit set so sometimes the TLB can use a lingering entry.
76 	 *
77 	 * This verifies that the fault happens in kernel space
78 	 * and that the fault was not a protection error.
79 	 */
80 
81 	if (address >= VMALLOC_START &&
82 	    (vector != 0x300 && vector != 0x400) &&
83 	    !user_mode(regs))
84 		goto vmalloc_fault;
85 
86 	/* If exceptions were enabled, we can reenable them here */
87 	if (user_mode(regs)) {
88 		/* Exception was in userspace: reenable interrupts */
89 		local_irq_enable();
90 		flags |= FAULT_FLAG_USER;
91 	} else {
92 		/* If exception was in a syscall, then IRQ's may have
93 		 * been enabled or disabled.  If they were enabled,
94 		 * reenable them.
95 		 */
96 		if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
97 			local_irq_enable();
98 	}
99 
100 	mm = tsk->mm;
101 	info.si_code = SEGV_MAPERR;
102 
103 	/*
104 	 * If we're in an interrupt or have no user
105 	 * context, we must not take the fault..
106 	 */
107 
108 	if (in_interrupt() || !mm)
109 		goto no_context;
110 
111 retry:
112 	down_read(&mm->mmap_sem);
113 	vma = find_vma(mm, address);
114 
115 	if (!vma)
116 		goto bad_area;
117 
118 	if (vma->vm_start <= address)
119 		goto good_area;
120 
121 	if (!(vma->vm_flags & VM_GROWSDOWN))
122 		goto bad_area;
123 
124 	if (user_mode(regs)) {
125 		/*
126 		 * accessing the stack below usp is always a bug.
127 		 * we get page-aligned addresses so we can only check
128 		 * if we're within a page from usp, but that might be
129 		 * enough to catch brutal errors at least.
130 		 */
131 		if (address + PAGE_SIZE < regs->sp)
132 			goto bad_area;
133 	}
134 	if (expand_stack(vma, address))
135 		goto bad_area;
136 
137 	/*
138 	 * Ok, we have a good vm_area for this memory access, so
139 	 * we can handle it..
140 	 */
141 
142 good_area:
143 	info.si_code = SEGV_ACCERR;
144 
145 	/* first do some preliminary protection checks */
146 
147 	if (write_acc) {
148 		if (!(vma->vm_flags & VM_WRITE))
149 			goto bad_area;
150 		flags |= FAULT_FLAG_WRITE;
151 	} else {
152 		/* not present */
153 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
154 			goto bad_area;
155 	}
156 
157 	/* are we trying to execute nonexecutable area */
158 	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
159 		goto bad_area;
160 
161 	/*
162 	 * If for any reason at all we couldn't handle the fault,
163 	 * make sure we exit gracefully rather than endlessly redo
164 	 * the fault.
165 	 */
166 
167 	fault = handle_mm_fault(vma, address, flags);
168 
169 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
170 		return;
171 
172 	if (unlikely(fault & VM_FAULT_ERROR)) {
173 		if (fault & VM_FAULT_OOM)
174 			goto out_of_memory;
175 		else if (fault & VM_FAULT_SIGSEGV)
176 			goto bad_area;
177 		else if (fault & VM_FAULT_SIGBUS)
178 			goto do_sigbus;
179 		BUG();
180 	}
181 
182 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
183 		/*RGD modeled on Cris */
184 		if (fault & VM_FAULT_MAJOR)
185 			tsk->maj_flt++;
186 		else
187 			tsk->min_flt++;
188 		if (fault & VM_FAULT_RETRY) {
189 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
190 			flags |= FAULT_FLAG_TRIED;
191 
192 			 /* No need to up_read(&mm->mmap_sem) as we would
193 			 * have already released it in __lock_page_or_retry
194 			 * in mm/filemap.c.
195 			 */
196 
197 			goto retry;
198 		}
199 	}
200 
201 	up_read(&mm->mmap_sem);
202 	return;
203 
204 	/*
205 	 * Something tried to access memory that isn't in our memory map..
206 	 * Fix it, but check if it's kernel or user first..
207 	 */
208 
209 bad_area:
210 	up_read(&mm->mmap_sem);
211 
212 bad_area_nosemaphore:
213 
214 	/* User mode accesses just cause a SIGSEGV */
215 
216 	if (user_mode(regs)) {
217 		info.si_signo = SIGSEGV;
218 		info.si_errno = 0;
219 		/* info.si_code has been set above */
220 		info.si_addr = (void *)address;
221 		force_sig_info(SIGSEGV, &info, tsk);
222 		return;
223 	}
224 
225 no_context:
226 
227 	/* Are we prepared to handle this kernel fault?
228 	 *
229 	 * (The kernel has valid exception-points in the source
230 	 *  when it acesses user-memory. When it fails in one
231 	 *  of those points, we find it in a table and do a jump
232 	 *  to some fixup code that loads an appropriate error
233 	 *  code)
234 	 */
235 
236 	{
237 		const struct exception_table_entry *entry;
238 
239 		__asm__ __volatile__("l.nop 42");
240 
241 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
242 			/* Adjust the instruction pointer in the stackframe */
243 			regs->pc = entry->fixup;
244 			return;
245 		}
246 	}
247 
248 	/*
249 	 * Oops. The kernel tried to access some bad page. We'll have to
250 	 * terminate things with extreme prejudice.
251 	 */
252 
253 	if ((unsigned long)(address) < PAGE_SIZE)
254 		printk(KERN_ALERT
255 		       "Unable to handle kernel NULL pointer dereference");
256 	else
257 		printk(KERN_ALERT "Unable to handle kernel access");
258 	printk(" at virtual address 0x%08lx\n", address);
259 
260 	die("Oops", regs, write_acc);
261 
262 	do_exit(SIGKILL);
263 
264 	/*
265 	 * We ran out of memory, or some other thing happened to us that made
266 	 * us unable to handle the page fault gracefully.
267 	 */
268 
269 out_of_memory:
270 	__asm__ __volatile__("l.nop 42");
271 	__asm__ __volatile__("l.nop 1");
272 
273 	up_read(&mm->mmap_sem);
274 	if (!user_mode(regs))
275 		goto no_context;
276 	pagefault_out_of_memory();
277 	return;
278 
279 do_sigbus:
280 	up_read(&mm->mmap_sem);
281 
282 	/*
283 	 * Send a sigbus, regardless of whether we were in kernel
284 	 * or user mode.
285 	 */
286 	info.si_signo = SIGBUS;
287 	info.si_errno = 0;
288 	info.si_code = BUS_ADRERR;
289 	info.si_addr = (void *)address;
290 	force_sig_info(SIGBUS, &info, tsk);
291 
292 	/* Kernel mode? Handle exceptions or die */
293 	if (!user_mode(regs))
294 		goto no_context;
295 	return;
296 
297 vmalloc_fault:
298 	{
299 		/*
300 		 * Synchronize this task's top level page-table
301 		 * with the 'reference' page table.
302 		 *
303 		 * Use current_pgd instead of tsk->active_mm->pgd
304 		 * since the latter might be unavailable if this
305 		 * code is executed in a misfortunately run irq
306 		 * (like inside schedule() between switch_mm and
307 		 *  switch_to...).
308 		 */
309 
310 		int offset = pgd_index(address);
311 		pgd_t *pgd, *pgd_k;
312 		pud_t *pud, *pud_k;
313 		pmd_t *pmd, *pmd_k;
314 		pte_t *pte_k;
315 
316 /*
317 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
318 			 "since current_pgd assign a proper value somewhere\n"
319 			 "anyhow we don't need this at the moment\n");
320 
321 		phx_mmu("vmalloc_fault");
322 */
323 		pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
324 		pgd_k = init_mm.pgd + offset;
325 
326 		/* Since we're two-level, we don't need to do both
327 		 * set_pgd and set_pmd (they do the same thing). If
328 		 * we go three-level at some point, do the right thing
329 		 * with pgd_present and set_pgd here.
330 		 *
331 		 * Also, since the vmalloc area is global, we don't
332 		 * need to copy individual PTE's, it is enough to
333 		 * copy the pgd pointer into the pte page of the
334 		 * root task. If that is there, we'll find our pte if
335 		 * it exists.
336 		 */
337 
338 		pud = pud_offset(pgd, address);
339 		pud_k = pud_offset(pgd_k, address);
340 		if (!pud_present(*pud_k))
341 			goto no_context;
342 
343 		pmd = pmd_offset(pud, address);
344 		pmd_k = pmd_offset(pud_k, address);
345 
346 		if (!pmd_present(*pmd_k))
347 			goto bad_area_nosemaphore;
348 
349 		set_pmd(pmd, *pmd_k);
350 
351 		/* Make sure the actual PTE exists as well to
352 		 * catch kernel vmalloc-area accesses to non-mapped
353 		 * addresses. If we don't do this, this will just
354 		 * silently loop forever.
355 		 */
356 
357 		pte_k = pte_offset_kernel(pmd_k, address);
358 		if (!pte_present(*pte_k))
359 			goto no_context;
360 
361 		return;
362 	}
363 }
364