xref: /openbmc/linux/arch/openrisc/mm/fault.c (revision b737eecd)
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/siginfo.h>
22 #include <asm/signal.h>
23 
24 #define NUM_TLB_ENTRIES 64
25 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
26 
27 unsigned long pte_misses;	/* updated by do_page_fault() */
28 unsigned long pte_errors;	/* updated by do_page_fault() */
29 
30 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
31  *            - also look into include/asm-or32/mmu_context.h
32  */
33 volatile pgd_t *current_pgd[NR_CPUS];
34 
35 extern void die(char *, struct pt_regs *, long);
36 
37 /*
38  * This routine handles page faults.  It determines the address,
39  * and the problem, and then passes it off to one of the appropriate
40  * routines.
41  *
42  * If this routine detects a bad access, it returns 1, otherwise it
43  * returns 0.
44  */
45 
46 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
47 			      unsigned long vector, int write_acc)
48 {
49 	struct task_struct *tsk;
50 	struct mm_struct *mm;
51 	struct vm_area_struct *vma;
52 	int si_code;
53 	vm_fault_t fault;
54 	unsigned int flags = FAULT_FLAG_DEFAULT;
55 
56 	tsk = current;
57 
58 	/*
59 	 * We fault-in kernel-space virtual memory on-demand. The
60 	 * 'reference' page table is init_mm.pgd.
61 	 *
62 	 * NOTE! We MUST NOT take any locks for this case. We may
63 	 * be in an interrupt or a critical region, and should
64 	 * only copy the information from the master page table,
65 	 * nothing more.
66 	 *
67 	 * NOTE2: This is done so that, when updating the vmalloc
68 	 * mappings we don't have to walk all processes pgdirs and
69 	 * add the high mappings all at once. Instead we do it as they
70 	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
71 	 * bit set so sometimes the TLB can use a lingering entry.
72 	 *
73 	 * This verifies that the fault happens in kernel space
74 	 * and that the fault was not a protection error.
75 	 */
76 
77 	if (address >= VMALLOC_START &&
78 	    (vector != 0x300 && vector != 0x400) &&
79 	    !user_mode(regs))
80 		goto vmalloc_fault;
81 
82 	/* If exceptions were enabled, we can reenable them here */
83 	if (user_mode(regs)) {
84 		/* Exception was in userspace: reenable interrupts */
85 		local_irq_enable();
86 		flags |= FAULT_FLAG_USER;
87 	} else {
88 		/* If exception was in a syscall, then IRQ's may have
89 		 * been enabled or disabled.  If they were enabled,
90 		 * reenable them.
91 		 */
92 		if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
93 			local_irq_enable();
94 	}
95 
96 	mm = tsk->mm;
97 	si_code = SEGV_MAPERR;
98 
99 	/*
100 	 * If we're in an interrupt or have no user
101 	 * context, we must not take the fault..
102 	 */
103 
104 	if (in_interrupt() || !mm)
105 		goto no_context;
106 
107 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
108 
109 retry:
110 	mmap_read_lock(mm);
111 	vma = find_vma(mm, address);
112 
113 	if (!vma)
114 		goto bad_area;
115 
116 	if (vma->vm_start <= address)
117 		goto good_area;
118 
119 	if (!(vma->vm_flags & VM_GROWSDOWN))
120 		goto bad_area;
121 
122 	if (user_mode(regs)) {
123 		/*
124 		 * accessing the stack below usp is always a bug.
125 		 * we get page-aligned addresses so we can only check
126 		 * if we're within a page from usp, but that might be
127 		 * enough to catch brutal errors at least.
128 		 */
129 		if (address + PAGE_SIZE < regs->sp)
130 			goto bad_area;
131 	}
132 	if (expand_stack(vma, address))
133 		goto bad_area;
134 
135 	/*
136 	 * Ok, we have a good vm_area for this memory access, so
137 	 * we can handle it..
138 	 */
139 
140 good_area:
141 	si_code = SEGV_ACCERR;
142 
143 	/* first do some preliminary protection checks */
144 
145 	if (write_acc) {
146 		if (!(vma->vm_flags & VM_WRITE))
147 			goto bad_area;
148 		flags |= FAULT_FLAG_WRITE;
149 	} else {
150 		/* not present */
151 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
152 			goto bad_area;
153 	}
154 
155 	/* are we trying to execute nonexecutable area */
156 	if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
157 		goto bad_area;
158 
159 	/*
160 	 * If for any reason at all we couldn't handle the fault,
161 	 * make sure we exit gracefully rather than endlessly redo
162 	 * the fault.
163 	 */
164 
165 	fault = handle_mm_fault(vma, address, flags, regs);
166 
167 	if (fault_signal_pending(fault, regs))
168 		return;
169 
170 	if (unlikely(fault & VM_FAULT_ERROR)) {
171 		if (fault & VM_FAULT_OOM)
172 			goto out_of_memory;
173 		else if (fault & VM_FAULT_SIGSEGV)
174 			goto bad_area;
175 		else if (fault & VM_FAULT_SIGBUS)
176 			goto do_sigbus;
177 		BUG();
178 	}
179 
180 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
181 		/*RGD modeled on Cris */
182 		if (fault & VM_FAULT_RETRY) {
183 			flags |= FAULT_FLAG_TRIED;
184 
185 			 /* No need to mmap_read_unlock(mm) as we would
186 			 * have already released it in __lock_page_or_retry
187 			 * in mm/filemap.c.
188 			 */
189 
190 			goto retry;
191 		}
192 	}
193 
194 	mmap_read_unlock(mm);
195 	return;
196 
197 	/*
198 	 * Something tried to access memory that isn't in our memory map..
199 	 * Fix it, but check if it's kernel or user first..
200 	 */
201 
202 bad_area:
203 	mmap_read_unlock(mm);
204 
205 bad_area_nosemaphore:
206 
207 	/* User mode accesses just cause a SIGSEGV */
208 
209 	if (user_mode(regs)) {
210 		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
211 		return;
212 	}
213 
214 no_context:
215 
216 	/* Are we prepared to handle this kernel fault?
217 	 *
218 	 * (The kernel has valid exception-points in the source
219 	 *  when it acesses user-memory. When it fails in one
220 	 *  of those points, we find it in a table and do a jump
221 	 *  to some fixup code that loads an appropriate error
222 	 *  code)
223 	 */
224 
225 	{
226 		const struct exception_table_entry *entry;
227 
228 		__asm__ __volatile__("l.nop 42");
229 
230 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
231 			/* Adjust the instruction pointer in the stackframe */
232 			regs->pc = entry->fixup;
233 			return;
234 		}
235 	}
236 
237 	/*
238 	 * Oops. The kernel tried to access some bad page. We'll have to
239 	 * terminate things with extreme prejudice.
240 	 */
241 
242 	if ((unsigned long)(address) < PAGE_SIZE)
243 		printk(KERN_ALERT
244 		       "Unable to handle kernel NULL pointer dereference");
245 	else
246 		printk(KERN_ALERT "Unable to handle kernel access");
247 	printk(" at virtual address 0x%08lx\n", address);
248 
249 	die("Oops", regs, write_acc);
250 
251 	do_exit(SIGKILL);
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 	__asm__ __volatile__("l.nop 42");
260 	__asm__ __volatile__("l.nop 1");
261 
262 	mmap_read_unlock(mm);
263 	if (!user_mode(regs))
264 		goto no_context;
265 	pagefault_out_of_memory();
266 	return;
267 
268 do_sigbus:
269 	mmap_read_unlock(mm);
270 
271 	/*
272 	 * Send a sigbus, regardless of whether we were in kernel
273 	 * or user mode.
274 	 */
275 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
276 
277 	/* Kernel mode? Handle exceptions or die */
278 	if (!user_mode(regs))
279 		goto no_context;
280 	return;
281 
282 vmalloc_fault:
283 	{
284 		/*
285 		 * Synchronize this task's top level page-table
286 		 * with the 'reference' page table.
287 		 *
288 		 * Use current_pgd instead of tsk->active_mm->pgd
289 		 * since the latter might be unavailable if this
290 		 * code is executed in a misfortunately run irq
291 		 * (like inside schedule() between switch_mm and
292 		 *  switch_to...).
293 		 */
294 
295 		int offset = pgd_index(address);
296 		pgd_t *pgd, *pgd_k;
297 		p4d_t *p4d, *p4d_k;
298 		pud_t *pud, *pud_k;
299 		pmd_t *pmd, *pmd_k;
300 		pte_t *pte_k;
301 
302 /*
303 		phx_warn("do_page_fault(): vmalloc_fault will not work, "
304 			 "since current_pgd assign a proper value somewhere\n"
305 			 "anyhow we don't need this at the moment\n");
306 
307 		phx_mmu("vmalloc_fault");
308 */
309 		pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset;
310 		pgd_k = init_mm.pgd + offset;
311 
312 		/* Since we're two-level, we don't need to do both
313 		 * set_pgd and set_pmd (they do the same thing). If
314 		 * we go three-level at some point, do the right thing
315 		 * with pgd_present and set_pgd here.
316 		 *
317 		 * Also, since the vmalloc area is global, we don't
318 		 * need to copy individual PTE's, it is enough to
319 		 * copy the pgd pointer into the pte page of the
320 		 * root task. If that is there, we'll find our pte if
321 		 * it exists.
322 		 */
323 
324 		p4d = p4d_offset(pgd, address);
325 		p4d_k = p4d_offset(pgd_k, address);
326 		if (!p4d_present(*p4d_k))
327 			goto no_context;
328 
329 		pud = pud_offset(p4d, address);
330 		pud_k = pud_offset(p4d_k, address);
331 		if (!pud_present(*pud_k))
332 			goto no_context;
333 
334 		pmd = pmd_offset(pud, address);
335 		pmd_k = pmd_offset(pud_k, address);
336 
337 		if (!pmd_present(*pmd_k))
338 			goto bad_area_nosemaphore;
339 
340 		set_pmd(pmd, *pmd_k);
341 
342 		/* Make sure the actual PTE exists as well to
343 		 * catch kernel vmalloc-area accesses to non-mapped
344 		 * addresses. If we don't do this, this will just
345 		 * silently loop forever.
346 		 */
347 
348 		pte_k = pte_offset_kernel(pmd_k, address);
349 		if (!pte_present(*pte_k))
350 			goto no_context;
351 
352 		return;
353 	}
354 }
355