xref: /openbmc/linux/arch/openrisc/mm/fault.c (revision 2a598d0b)
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 	vma = expand_stack(mm, address);
131 	if (!vma)
132 		goto bad_area_nosemaphore;
133 
134 	/*
135 	 * Ok, we have a good vm_area for this memory access, so
136 	 * we can handle it..
137 	 */
138 
139 good_area:
140 	si_code = SEGV_ACCERR;
141 
142 	/* first do some preliminary protection checks */
143 
144 	if (write_acc) {
145 		if (!(vma->vm_flags & VM_WRITE))
146 			goto bad_area;
147 		flags |= FAULT_FLAG_WRITE;
148 	} else {
149 		/* not present */
150 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
151 			goto bad_area;
152 	}
153 
154 	/* are we trying to execute nonexecutable area */
155 	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
156 		goto bad_area;
157 
158 	/*
159 	 * If for any reason at all we couldn't handle the fault,
160 	 * make sure we exit gracefully rather than endlessly redo
161 	 * the fault.
162 	 */
163 
164 	fault = handle_mm_fault(vma, address, flags, regs);
165 
166 	if (fault_signal_pending(fault, regs)) {
167 		if (!user_mode(regs))
168 			goto no_context;
169 		return;
170 	}
171 
172 	/* The fault is fully completed (including releasing mmap lock) */
173 	if (fault & VM_FAULT_COMPLETED)
174 		return;
175 
176 	if (unlikely(fault & VM_FAULT_ERROR)) {
177 		if (fault & VM_FAULT_OOM)
178 			goto out_of_memory;
179 		else if (fault & VM_FAULT_SIGSEGV)
180 			goto bad_area;
181 		else if (fault & VM_FAULT_SIGBUS)
182 			goto do_sigbus;
183 		BUG();
184 	}
185 
186 	/*RGD modeled on Cris */
187 	if (fault & VM_FAULT_RETRY) {
188 		flags |= FAULT_FLAG_TRIED;
189 
190 		/* No need to mmap_read_unlock(mm) as we would
191 		 * have already released it in __lock_page_or_retry
192 		 * in mm/filemap.c.
193 		 */
194 
195 		goto retry;
196 	}
197 
198 	mmap_read_unlock(mm);
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 
206 bad_area:
207 	mmap_read_unlock(mm);
208 
209 bad_area_nosemaphore:
210 
211 	/* User mode accesses just cause a SIGSEGV */
212 
213 	if (user_mode(regs)) {
214 		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
215 		return;
216 	}
217 
218 no_context:
219 
220 	/* Are we prepared to handle this kernel fault?
221 	 *
222 	 * (The kernel has valid exception-points in the source
223 	 *  when it acesses user-memory. When it fails in one
224 	 *  of those points, we find it in a table and do a jump
225 	 *  to some fixup code that loads an appropriate error
226 	 *  code)
227 	 */
228 
229 	{
230 		const struct exception_table_entry *entry;
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 	/*
254 	 * We ran out of memory, or some other thing happened to us that made
255 	 * us unable to handle the page fault gracefully.
256 	 */
257 
258 out_of_memory:
259 	mmap_read_unlock(mm);
260 	if (!user_mode(regs))
261 		goto no_context;
262 	pagefault_out_of_memory();
263 	return;
264 
265 do_sigbus:
266 	mmap_read_unlock(mm);
267 
268 	/*
269 	 * Send a sigbus, regardless of whether we were in kernel
270 	 * or user mode.
271 	 */
272 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
273 
274 	/* Kernel mode? Handle exceptions or die */
275 	if (!user_mode(regs))
276 		goto no_context;
277 	return;
278 
279 vmalloc_fault:
280 	{
281 		/*
282 		 * Synchronize this task's top level page-table
283 		 * with the 'reference' page table.
284 		 *
285 		 * Use current_pgd instead of tsk->active_mm->pgd
286 		 * since the latter might be unavailable if this
287 		 * code is executed in a misfortunately run irq
288 		 * (like inside schedule() between switch_mm and
289 		 *  switch_to...).
290 		 */
291 
292 		int offset = pgd_index(address);
293 		pgd_t *pgd, *pgd_k;
294 		p4d_t *p4d, *p4d_k;
295 		pud_t *pud, *pud_k;
296 		pmd_t *pmd, *pmd_k;
297 		pte_t *pte_k;
298 
299 /*
300 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
301 			 "since current_pgd assign a proper value somewhere\n"
302 			 "anyhow we don't need this at the moment\n");
303 
304 		phx_mmu("vmalloc_fault");
305 */
306 		pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
307 		pgd_k = init_mm.pgd + offset;
308 
309 		/* Since we're two-level, we don't need to do both
310 		 * set_pgd and set_pmd (they do the same thing). If
311 		 * we go three-level at some point, do the right thing
312 		 * with pgd_present and set_pgd here.
313 		 *
314 		 * Also, since the vmalloc area is global, we don't
315 		 * need to copy individual PTE's, it is enough to
316 		 * copy the pgd pointer into the pte page of the
317 		 * root task. If that is there, we'll find our pte if
318 		 * it exists.
319 		 */
320 
321 		p4d = p4d_offset(pgd, address);
322 		p4d_k = p4d_offset(pgd_k, address);
323 		if (!p4d_present(*p4d_k))
324 			goto no_context;
325 
326 		pud = pud_offset(p4d, address);
327 		pud_k = pud_offset(p4d_k, address);
328 		if (!pud_present(*pud_k))
329 			goto no_context;
330 
331 		pmd = pmd_offset(pud, address);
332 		pmd_k = pmd_offset(pud_k, address);
333 
334 		if (!pmd_present(*pmd_k))
335 			goto bad_area_nosemaphore;
336 
337 		set_pmd(pmd, *pmd_k);
338 
339 		/* Make sure the actual PTE exists as well to
340 		 * catch kernel vmalloc-area accesses to non-mapped
341 		 * addresses. If we don't do this, this will just
342 		 * silently loop forever.
343 		 */
344 
345 		pte_k = pte_offset_kernel(pmd_k, address);
346 		if (!pte_present(*pte_k))
347 			goto no_context;
348 
349 		return;
350 	}
351 }
352