xref: /linux/arch/openrisc/mm/fault.c (revision d198b34f3855eee2571dda03eea75a09c7c31480)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * OpenRISC fault.c
4  *
5  * Linux architectural port borrowing liberally from similar works of
6  * others.  All original copyrights apply as per the original source
7  * declaration.
8  *
9  * Modifications for the OpenRISC architecture:
10  * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
11  * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
12  */
13 
14 #include <linux/mm.h>
15 #include <linux/interrupt.h>
16 #include <linux/extable.h>
17 #include <linux/sched/signal.h>
18 
19 #include <linux/uaccess.h>
20 #include <asm/siginfo.h>
21 #include <asm/signal.h>
22 
23 #define NUM_TLB_ENTRIES 64
24 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
25 
26 unsigned long pte_misses;	/* updated by do_page_fault() */
27 unsigned long pte_errors;	/* updated by do_page_fault() */
28 
29 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
30  *            - also look into include/asm-or32/mmu_context.h
31  */
32 volatile pgd_t *current_pgd[NR_CPUS];
33 
34 extern void die(char *, struct pt_regs *, long);
35 
36 /*
37  * This routine handles page faults.  It determines the address,
38  * and the problem, and then passes it off to one of the appropriate
39  * routines.
40  *
41  * If this routine detects a bad access, it returns 1, otherwise it
42  * returns 0.
43  */
44 
45 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
46 			      unsigned long vector, int write_acc)
47 {
48 	struct task_struct *tsk;
49 	struct mm_struct *mm;
50 	struct vm_area_struct *vma;
51 	int si_code;
52 	vm_fault_t fault;
53 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
54 
55 	tsk = current;
56 
57 	/*
58 	 * We fault-in kernel-space virtual memory on-demand. The
59 	 * 'reference' page table is init_mm.pgd.
60 	 *
61 	 * NOTE! We MUST NOT take any locks for this case. We may
62 	 * be in an interrupt or a critical region, and should
63 	 * only copy the information from the master page table,
64 	 * nothing more.
65 	 *
66 	 * NOTE2: This is done so that, when updating the vmalloc
67 	 * mappings we don't have to walk all processes pgdirs and
68 	 * add the high mappings all at once. Instead we do it as they
69 	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
70 	 * bit set so sometimes the TLB can use a lingering entry.
71 	 *
72 	 * This verifies that the fault happens in kernel space
73 	 * and that the fault was not a protection error.
74 	 */
75 
76 	if (address >= VMALLOC_START &&
77 	    (vector != 0x300 && vector != 0x400) &&
78 	    !user_mode(regs))
79 		goto vmalloc_fault;
80 
81 	/* If exceptions were enabled, we can reenable them here */
82 	if (user_mode(regs)) {
83 		/* Exception was in userspace: reenable interrupts */
84 		local_irq_enable();
85 		flags |= FAULT_FLAG_USER;
86 	} else {
87 		/* If exception was in a syscall, then IRQ's may have
88 		 * been enabled or disabled.  If they were enabled,
89 		 * reenable them.
90 		 */
91 		if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
92 			local_irq_enable();
93 	}
94 
95 	mm = tsk->mm;
96 	si_code = SEGV_MAPERR;
97 
98 	/*
99 	 * If we're in an interrupt or have no user
100 	 * context, we must not take the fault..
101 	 */
102 
103 	if (in_interrupt() || !mm)
104 		goto no_context;
105 
106 retry:
107 	down_read(&mm->mmap_sem);
108 	vma = find_vma(mm, address);
109 
110 	if (!vma)
111 		goto bad_area;
112 
113 	if (vma->vm_start <= address)
114 		goto good_area;
115 
116 	if (!(vma->vm_flags & VM_GROWSDOWN))
117 		goto bad_area;
118 
119 	if (user_mode(regs)) {
120 		/*
121 		 * accessing the stack below usp is always a bug.
122 		 * we get page-aligned addresses so we can only check
123 		 * if we're within a page from usp, but that might be
124 		 * enough to catch brutal errors at least.
125 		 */
126 		if (address + PAGE_SIZE < regs->sp)
127 			goto bad_area;
128 	}
129 	if (expand_stack(vma, address))
130 		goto bad_area;
131 
132 	/*
133 	 * Ok, we have a good vm_area for this memory access, so
134 	 * we can handle it..
135 	 */
136 
137 good_area:
138 	si_code = SEGV_ACCERR;
139 
140 	/* first do some preliminary protection checks */
141 
142 	if (write_acc) {
143 		if (!(vma->vm_flags & VM_WRITE))
144 			goto bad_area;
145 		flags |= FAULT_FLAG_WRITE;
146 	} else {
147 		/* not present */
148 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
149 			goto bad_area;
150 	}
151 
152 	/* are we trying to execute nonexecutable area */
153 	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
154 		goto bad_area;
155 
156 	/*
157 	 * If for any reason at all we couldn't handle the fault,
158 	 * make sure we exit gracefully rather than endlessly redo
159 	 * the fault.
160 	 */
161 
162 	fault = handle_mm_fault(vma, address, flags);
163 
164 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
165 		return;
166 
167 	if (unlikely(fault & VM_FAULT_ERROR)) {
168 		if (fault & VM_FAULT_OOM)
169 			goto out_of_memory;
170 		else if (fault & VM_FAULT_SIGSEGV)
171 			goto bad_area;
172 		else if (fault & VM_FAULT_SIGBUS)
173 			goto do_sigbus;
174 		BUG();
175 	}
176 
177 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
178 		/*RGD modeled on Cris */
179 		if (fault & VM_FAULT_MAJOR)
180 			tsk->maj_flt++;
181 		else
182 			tsk->min_flt++;
183 		if (fault & VM_FAULT_RETRY) {
184 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
185 			flags |= FAULT_FLAG_TRIED;
186 
187 			 /* No need to up_read(&mm->mmap_sem) as we would
188 			 * have already released it in __lock_page_or_retry
189 			 * in mm/filemap.c.
190 			 */
191 
192 			goto retry;
193 		}
194 	}
195 
196 	up_read(&mm->mmap_sem);
197 	return;
198 
199 	/*
200 	 * Something tried to access memory that isn't in our memory map..
201 	 * Fix it, but check if it's kernel or user first..
202 	 */
203 
204 bad_area:
205 	up_read(&mm->mmap_sem);
206 
207 bad_area_nosemaphore:
208 
209 	/* User mode accesses just cause a SIGSEGV */
210 
211 	if (user_mode(regs)) {
212 		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
213 		return;
214 	}
215 
216 no_context:
217 
218 	/* Are we prepared to handle this kernel fault?
219 	 *
220 	 * (The kernel has valid exception-points in the source
221 	 *  when it acesses user-memory. When it fails in one
222 	 *  of those points, we find it in a table and do a jump
223 	 *  to some fixup code that loads an appropriate error
224 	 *  code)
225 	 */
226 
227 	{
228 		const struct exception_table_entry *entry;
229 
230 		__asm__ __volatile__("l.nop 42");
231 
232 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
233 			/* Adjust the instruction pointer in the stackframe */
234 			regs->pc = entry->fixup;
235 			return;
236 		}
237 	}
238 
239 	/*
240 	 * Oops. The kernel tried to access some bad page. We'll have to
241 	 * terminate things with extreme prejudice.
242 	 */
243 
244 	if ((unsigned long)(address) < PAGE_SIZE)
245 		printk(KERN_ALERT
246 		       "Unable to handle kernel NULL pointer dereference");
247 	else
248 		printk(KERN_ALERT "Unable to handle kernel access");
249 	printk(" at virtual address 0x%08lx\n", address);
250 
251 	die("Oops", regs, write_acc);
252 
253 	do_exit(SIGKILL);
254 
255 	/*
256 	 * We ran out of memory, or some other thing happened to us that made
257 	 * us unable to handle the page fault gracefully.
258 	 */
259 
260 out_of_memory:
261 	__asm__ __volatile__("l.nop 42");
262 	__asm__ __volatile__("l.nop 1");
263 
264 	up_read(&mm->mmap_sem);
265 	if (!user_mode(regs))
266 		goto no_context;
267 	pagefault_out_of_memory();
268 	return;
269 
270 do_sigbus:
271 	up_read(&mm->mmap_sem);
272 
273 	/*
274 	 * Send a sigbus, regardless of whether we were in kernel
275 	 * or user mode.
276 	 */
277 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
278 
279 	/* Kernel mode? Handle exceptions or die */
280 	if (!user_mode(regs))
281 		goto no_context;
282 	return;
283 
284 vmalloc_fault:
285 	{
286 		/*
287 		 * Synchronize this task's top level page-table
288 		 * with the 'reference' page table.
289 		 *
290 		 * Use current_pgd instead of tsk->active_mm->pgd
291 		 * since the latter might be unavailable if this
292 		 * code is executed in a misfortunately run irq
293 		 * (like inside schedule() between switch_mm and
294 		 *  switch_to...).
295 		 */
296 
297 		int offset = pgd_index(address);
298 		pgd_t *pgd, *pgd_k;
299 		pud_t *pud, *pud_k;
300 		pmd_t *pmd, *pmd_k;
301 		pte_t *pte_k;
302 
303 /*
304 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
305 			 "since current_pgd assign a proper value somewhere\n"
306 			 "anyhow we don't need this at the moment\n");
307 
308 		phx_mmu("vmalloc_fault");
309 */
310 		pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
311 		pgd_k = init_mm.pgd + offset;
312 
313 		/* Since we're two-level, we don't need to do both
314 		 * set_pgd and set_pmd (they do the same thing). If
315 		 * we go three-level at some point, do the right thing
316 		 * with pgd_present and set_pgd here.
317 		 *
318 		 * Also, since the vmalloc area is global, we don't
319 		 * need to copy individual PTE's, it is enough to
320 		 * copy the pgd pointer into the pte page of the
321 		 * root task. If that is there, we'll find our pte if
322 		 * it exists.
323 		 */
324 
325 		pud = pud_offset(pgd, address);
326 		pud_k = pud_offset(pgd_k, address);
327 		if (!pud_present(*pud_k))
328 			goto no_context;
329 
330 		pmd = pmd_offset(pud, address);
331 		pmd_k = pmd_offset(pud_k, address);
332 
333 		if (!pmd_present(*pmd_k))
334 			goto bad_area_nosemaphore;
335 
336 		set_pmd(pmd, *pmd_k);
337 
338 		/* Make sure the actual PTE exists as well to
339 		 * catch kernel vmalloc-area accesses to non-mapped
340 		 * addresses. If we don't do this, this will just
341 		 * silently loop forever.
342 		 */
343 
344 		pte_k = pte_offset_kernel(pmd_k, address);
345 		if (!pte_present(*pte_k))
346 			goto no_context;
347 
348 		return;
349 	}
350 }
351