xref: /openbmc/linux/arch/openrisc/mm/fault.c (revision 2d99a7ec)
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 #include <linux/perf_event.h>
19 
20 #include <linux/uaccess.h>
21 #include <asm/mmu_context.h>
22 #include <asm/siginfo.h>
23 #include <asm/signal.h>
24 
25 #define NUM_TLB_ENTRIES 64
26 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
27 
28 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
29  *            - also look into include/asm/mmu_context.h
30  */
31 volatile pgd_t *current_pgd[NR_CPUS];
32 
33 extern void __noreturn die(char *, struct pt_regs *, long);
34 
35 /*
36  * This routine handles page faults.  It determines the address,
37  * and the problem, and then passes it off to one of the appropriate
38  * routines.
39  *
40  * If this routine detects a bad access, it returns 1, otherwise it
41  * returns 0.
42  */
43 
44 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
45 			      unsigned long vector, int write_acc)
46 {
47 	struct task_struct *tsk;
48 	struct mm_struct *mm;
49 	struct vm_area_struct *vma;
50 	int si_code;
51 	vm_fault_t fault;
52 	unsigned int flags = FAULT_FLAG_DEFAULT;
53 
54 	tsk = current;
55 
56 	/*
57 	 * We fault-in kernel-space virtual memory on-demand. The
58 	 * 'reference' page table is init_mm.pgd.
59 	 *
60 	 * NOTE! We MUST NOT take any locks for this case. We may
61 	 * be in an interrupt or a critical region, and should
62 	 * only copy the information from the master page table,
63 	 * nothing more.
64 	 *
65 	 * NOTE2: This is done so that, when updating the vmalloc
66 	 * mappings we don't have to walk all processes pgdirs and
67 	 * add the high mappings all at once. Instead we do it as they
68 	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
69 	 * bit set so sometimes the TLB can use a lingering entry.
70 	 *
71 	 * This verifies that the fault happens in kernel space
72 	 * and that the fault was not a protection error.
73 	 */
74 
75 	if (address >= VMALLOC_START &&
76 	    (vector != 0x300 && vector != 0x400) &&
77 	    !user_mode(regs))
78 		goto vmalloc_fault;
79 
80 	/* If exceptions were enabled, we can reenable them here */
81 	if (user_mode(regs)) {
82 		/* Exception was in userspace: reenable interrupts */
83 		local_irq_enable();
84 		flags |= FAULT_FLAG_USER;
85 	} else {
86 		/* If exception was in a syscall, then IRQ's may have
87 		 * been enabled or disabled.  If they were enabled,
88 		 * reenable them.
89 		 */
90 		if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
91 			local_irq_enable();
92 	}
93 
94 	mm = tsk->mm;
95 	si_code = SEGV_MAPERR;
96 
97 	/*
98 	 * If we're in an interrupt or have no user
99 	 * context, we must not take the fault..
100 	 */
101 
102 	if (in_interrupt() || !mm)
103 		goto no_context;
104 
105 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
106 
107 retry:
108 	mmap_read_lock(mm);
109 	vma = find_vma(mm, address);
110 
111 	if (!vma)
112 		goto bad_area;
113 
114 	if (vma->vm_start <= address)
115 		goto good_area;
116 
117 	if (!(vma->vm_flags & VM_GROWSDOWN))
118 		goto bad_area;
119 
120 	if (user_mode(regs)) {
121 		/*
122 		 * accessing the stack below usp is always a bug.
123 		 * we get page-aligned addresses so we can only check
124 		 * if we're within a page from usp, but that might be
125 		 * enough to catch brutal errors at least.
126 		 */
127 		if (address + PAGE_SIZE < regs->sp)
128 			goto bad_area;
129 	}
130 	if (expand_stack(vma, address))
131 		goto bad_area;
132 
133 	/*
134 	 * Ok, we have a good vm_area for this memory access, so
135 	 * we can handle it..
136 	 */
137 
138 good_area:
139 	si_code = SEGV_ACCERR;
140 
141 	/* first do some preliminary protection checks */
142 
143 	if (write_acc) {
144 		if (!(vma->vm_flags & VM_WRITE))
145 			goto bad_area;
146 		flags |= FAULT_FLAG_WRITE;
147 	} else {
148 		/* not present */
149 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
150 			goto bad_area;
151 	}
152 
153 	/* are we trying to execute nonexecutable area */
154 	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
155 		goto bad_area;
156 
157 	/*
158 	 * If for any reason at all we couldn't handle the fault,
159 	 * make sure we exit gracefully rather than endlessly redo
160 	 * the fault.
161 	 */
162 
163 	fault = handle_mm_fault(vma, address, flags, regs);
164 
165 	if (fault_signal_pending(fault, regs)) {
166 		if (!user_mode(regs))
167 			goto no_context;
168 		return;
169 	}
170 
171 	/* The fault is fully completed (including releasing mmap lock) */
172 	if (fault & VM_FAULT_COMPLETED)
173 		return;
174 
175 	if (unlikely(fault & VM_FAULT_ERROR)) {
176 		if (fault & VM_FAULT_OOM)
177 			goto out_of_memory;
178 		else if (fault & VM_FAULT_SIGSEGV)
179 			goto bad_area;
180 		else if (fault & VM_FAULT_SIGBUS)
181 			goto do_sigbus;
182 		BUG();
183 	}
184 
185 	/*RGD modeled on Cris */
186 	if (fault & VM_FAULT_RETRY) {
187 		flags |= FAULT_FLAG_TRIED;
188 
189 		/* No need to mmap_read_unlock(mm) 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 	mmap_read_unlock(mm);
198 	return;
199 
200 	/*
201 	 * Something tried to access memory that isn't in our memory map..
202 	 * Fix it, but check if it's kernel or user first..
203 	 */
204 
205 bad_area:
206 	mmap_read_unlock(mm);
207 
208 bad_area_nosemaphore:
209 
210 	/* User mode accesses just cause a SIGSEGV */
211 
212 	if (user_mode(regs)) {
213 		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
214 		return;
215 	}
216 
217 no_context:
218 
219 	/* Are we prepared to handle this kernel fault?
220 	 *
221 	 * (The kernel has valid exception-points in the source
222 	 *  when it acesses user-memory. When it fails in one
223 	 *  of those points, we find it in a table and do a jump
224 	 *  to some fixup code that loads an appropriate error
225 	 *  code)
226 	 */
227 
228 	{
229 		const struct exception_table_entry *entry;
230 
231 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
232 			/* Adjust the instruction pointer in the stackframe */
233 			regs->pc = entry->fixup;
234 			return;
235 		}
236 	}
237 
238 	/*
239 	 * Oops. The kernel tried to access some bad page. We'll have to
240 	 * terminate things with extreme prejudice.
241 	 */
242 
243 	if ((unsigned long)(address) < PAGE_SIZE)
244 		printk(KERN_ALERT
245 		       "Unable to handle kernel NULL pointer dereference");
246 	else
247 		printk(KERN_ALERT "Unable to handle kernel access");
248 	printk(" at virtual address 0x%08lx\n", address);
249 
250 	die("Oops", regs, write_acc);
251 
252 	/*
253 	 * We ran out of memory, or some other thing happened to us that made
254 	 * us unable to handle the page fault gracefully.
255 	 */
256 
257 out_of_memory:
258 	mmap_read_unlock(mm);
259 	if (!user_mode(regs))
260 		goto no_context;
261 	pagefault_out_of_memory();
262 	return;
263 
264 do_sigbus:
265 	mmap_read_unlock(mm);
266 
267 	/*
268 	 * Send a sigbus, regardless of whether we were in kernel
269 	 * or user mode.
270 	 */
271 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
272 
273 	/* Kernel mode? Handle exceptions or die */
274 	if (!user_mode(regs))
275 		goto no_context;
276 	return;
277 
278 vmalloc_fault:
279 	{
280 		/*
281 		 * Synchronize this task's top level page-table
282 		 * with the 'reference' page table.
283 		 *
284 		 * Use current_pgd instead of tsk->active_mm->pgd
285 		 * since the latter might be unavailable if this
286 		 * code is executed in a misfortunately run irq
287 		 * (like inside schedule() between switch_mm and
288 		 *  switch_to...).
289 		 */
290 
291 		int offset = pgd_index(address);
292 		pgd_t *pgd, *pgd_k;
293 		p4d_t *p4d, *p4d_k;
294 		pud_t *pud, *pud_k;
295 		pmd_t *pmd, *pmd_k;
296 		pte_t *pte_k;
297 
298 /*
299 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
300 			 "since current_pgd assign a proper value somewhere\n"
301 			 "anyhow we don't need this at the moment\n");
302 
303 		phx_mmu("vmalloc_fault");
304 */
305 		pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
306 		pgd_k = init_mm.pgd + offset;
307 
308 		/* Since we're two-level, we don't need to do both
309 		 * set_pgd and set_pmd (they do the same thing). If
310 		 * we go three-level at some point, do the right thing
311 		 * with pgd_present and set_pgd here.
312 		 *
313 		 * Also, since the vmalloc area is global, we don't
314 		 * need to copy individual PTE's, it is enough to
315 		 * copy the pgd pointer into the pte page of the
316 		 * root task. If that is there, we'll find our pte if
317 		 * it exists.
318 		 */
319 
320 		p4d = p4d_offset(pgd, address);
321 		p4d_k = p4d_offset(pgd_k, address);
322 		if (!p4d_present(*p4d_k))
323 			goto no_context;
324 
325 		pud = pud_offset(p4d, address);
326 		pud_k = pud_offset(p4d_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