xref: /openbmc/linux/arch/xtensa/mm/fault.c (revision e7f127b2)
1 // TODO VM_EXEC flag work-around, cache aliasing
2 /*
3  * arch/xtensa/mm/fault.c
4  *
5  * This file is subject to the terms and conditions of the GNU General Public
6  * License.  See the file "COPYING" in the main directory of this archive
7  * for more details.
8  *
9  * Copyright (C) 2001 - 2010 Tensilica Inc.
10  *
11  * Chris Zankel <chris@zankel.net>
12  * Joe Taylor	<joe@tensilica.com, joetylr@yahoo.com>
13  */
14 
15 #include <linux/mm.h>
16 #include <linux/extable.h>
17 #include <linux/hardirq.h>
18 #include <linux/perf_event.h>
19 #include <linux/uaccess.h>
20 #include <asm/mmu_context.h>
21 #include <asm/cacheflush.h>
22 #include <asm/hardirq.h>
23 
24 DEFINE_PER_CPU(unsigned long, asid_cache) = ASID_USER_FIRST;
25 void bad_page_fault(struct pt_regs*, unsigned long, int);
26 
27 /*
28  * This routine handles page faults.  It determines the address,
29  * and the problem, and then passes it off to one of the appropriate
30  * routines.
31  *
32  * Note: does not handle Miss and MultiHit.
33  */
34 
35 void do_page_fault(struct pt_regs *regs)
36 {
37 	struct vm_area_struct * vma;
38 	struct mm_struct *mm = current->mm;
39 	unsigned int exccause = regs->exccause;
40 	unsigned int address = regs->excvaddr;
41 	int code;
42 
43 	int is_write, is_exec;
44 	vm_fault_t fault;
45 	unsigned int flags = FAULT_FLAG_DEFAULT;
46 
47 	code = SEGV_MAPERR;
48 
49 	/* We fault-in kernel-space virtual memory on-demand. The
50 	 * 'reference' page table is init_mm.pgd.
51 	 */
52 	if (address >= TASK_SIZE && !user_mode(regs))
53 		goto vmalloc_fault;
54 
55 	/* If we're in an interrupt or have no user
56 	 * context, we must not take the fault..
57 	 */
58 	if (faulthandler_disabled() || !mm) {
59 		bad_page_fault(regs, address, SIGSEGV);
60 		return;
61 	}
62 
63 	is_write = (exccause == EXCCAUSE_STORE_CACHE_ATTRIBUTE) ? 1 : 0;
64 	is_exec =  (exccause == EXCCAUSE_ITLB_PRIVILEGE ||
65 		    exccause == EXCCAUSE_ITLB_MISS ||
66 		    exccause == EXCCAUSE_FETCH_CACHE_ATTRIBUTE) ? 1 : 0;
67 
68 	pr_debug("[%s:%d:%08x:%d:%08lx:%s%s]\n",
69 		 current->comm, current->pid,
70 		 address, exccause, regs->pc,
71 		 is_write ? "w" : "", is_exec ? "x" : "");
72 
73 	if (user_mode(regs))
74 		flags |= FAULT_FLAG_USER;
75 
76 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
77 
78 retry:
79 	mmap_read_lock(mm);
80 	vma = find_vma(mm, address);
81 
82 	if (!vma)
83 		goto bad_area;
84 	if (vma->vm_start <= address)
85 		goto good_area;
86 	if (!(vma->vm_flags & VM_GROWSDOWN))
87 		goto bad_area;
88 	if (expand_stack(vma, address))
89 		goto bad_area;
90 
91 	/* Ok, we have a good vm_area for this memory access, so
92 	 * we can handle it..
93 	 */
94 
95 good_area:
96 	code = SEGV_ACCERR;
97 
98 	if (is_write) {
99 		if (!(vma->vm_flags & VM_WRITE))
100 			goto bad_area;
101 		flags |= FAULT_FLAG_WRITE;
102 	} else if (is_exec) {
103 		if (!(vma->vm_flags & VM_EXEC))
104 			goto bad_area;
105 	} else	/* Allow read even from write-only pages. */
106 		if (!(vma->vm_flags & (VM_READ | VM_WRITE)))
107 			goto bad_area;
108 
109 	/* If for any reason at all we couldn't handle the fault,
110 	 * make sure we exit gracefully rather than endlessly redo
111 	 * the fault.
112 	 */
113 	fault = handle_mm_fault(vma, address, flags, regs);
114 
115 	if (fault_signal_pending(fault, regs)) {
116 		if (!user_mode(regs))
117 			goto bad_page_fault;
118 		return;
119 	}
120 
121 	if (unlikely(fault & VM_FAULT_ERROR)) {
122 		if (fault & VM_FAULT_OOM)
123 			goto out_of_memory;
124 		else if (fault & VM_FAULT_SIGSEGV)
125 			goto bad_area;
126 		else if (fault & VM_FAULT_SIGBUS)
127 			goto do_sigbus;
128 		BUG();
129 	}
130 
131 	if (fault & VM_FAULT_RETRY) {
132 		flags |= FAULT_FLAG_TRIED;
133 
134 		/* No need to mmap_read_unlock(mm) as we would
135 		 * have already released it in __lock_page_or_retry
136 		 * in mm/filemap.c.
137 		 */
138 
139 		goto retry;
140 	}
141 
142 	mmap_read_unlock(mm);
143 	return;
144 
145 	/* Something tried to access memory that isn't in our memory map..
146 	 * Fix it, but check if it's kernel or user first..
147 	 */
148 bad_area:
149 	mmap_read_unlock(mm);
150 	if (user_mode(regs)) {
151 		current->thread.bad_vaddr = address;
152 		current->thread.error_code = is_write;
153 		force_sig_fault(SIGSEGV, code, (void *) address);
154 		return;
155 	}
156 	bad_page_fault(regs, address, SIGSEGV);
157 	return;
158 
159 
160 	/* We ran out of memory, or some other thing happened to us that made
161 	 * us unable to handle the page fault gracefully.
162 	 */
163 out_of_memory:
164 	mmap_read_unlock(mm);
165 	if (!user_mode(regs))
166 		bad_page_fault(regs, address, SIGKILL);
167 	else
168 		pagefault_out_of_memory();
169 	return;
170 
171 do_sigbus:
172 	mmap_read_unlock(mm);
173 
174 	/* Send a sigbus, regardless of whether we were in kernel
175 	 * or user mode.
176 	 */
177 	current->thread.bad_vaddr = address;
178 	force_sig_fault(SIGBUS, BUS_ADRERR, (void *) address);
179 
180 	/* Kernel mode? Handle exceptions or die */
181 	if (!user_mode(regs))
182 		bad_page_fault(regs, address, SIGBUS);
183 	return;
184 
185 vmalloc_fault:
186 	{
187 		/* Synchronize this task's top level page-table
188 		 * with the 'reference' page table.
189 		 */
190 		struct mm_struct *act_mm = current->active_mm;
191 		int index = pgd_index(address);
192 		pgd_t *pgd, *pgd_k;
193 		p4d_t *p4d, *p4d_k;
194 		pud_t *pud, *pud_k;
195 		pmd_t *pmd, *pmd_k;
196 		pte_t *pte_k;
197 
198 		if (act_mm == NULL)
199 			goto bad_page_fault;
200 
201 		pgd = act_mm->pgd + index;
202 		pgd_k = init_mm.pgd + index;
203 
204 		if (!pgd_present(*pgd_k))
205 			goto bad_page_fault;
206 
207 		pgd_val(*pgd) = pgd_val(*pgd_k);
208 
209 		p4d = p4d_offset(pgd, address);
210 		p4d_k = p4d_offset(pgd_k, address);
211 		if (!p4d_present(*p4d) || !p4d_present(*p4d_k))
212 			goto bad_page_fault;
213 
214 		pud = pud_offset(p4d, address);
215 		pud_k = pud_offset(p4d_k, address);
216 		if (!pud_present(*pud) || !pud_present(*pud_k))
217 			goto bad_page_fault;
218 
219 		pmd = pmd_offset(pud, address);
220 		pmd_k = pmd_offset(pud_k, address);
221 		if (!pmd_present(*pmd) || !pmd_present(*pmd_k))
222 			goto bad_page_fault;
223 
224 		pmd_val(*pmd) = pmd_val(*pmd_k);
225 		pte_k = pte_offset_kernel(pmd_k, address);
226 
227 		if (!pte_present(*pte_k))
228 			goto bad_page_fault;
229 		return;
230 	}
231 bad_page_fault:
232 	bad_page_fault(regs, address, SIGKILL);
233 	return;
234 }
235 
236 
237 void
238 bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
239 {
240 	extern void __noreturn die(const char*, struct pt_regs*, long);
241 	const struct exception_table_entry *entry;
242 
243 	/* Are we prepared to handle this kernel fault?  */
244 	if ((entry = search_exception_tables(regs->pc)) != NULL) {
245 		pr_debug("%s: Exception at pc=%#010lx (%lx)\n",
246 			 current->comm, regs->pc, entry->fixup);
247 		current->thread.bad_uaddr = address;
248 		regs->pc = entry->fixup;
249 		return;
250 	}
251 
252 	/* Oops. The kernel tried to access some bad page. We'll have to
253 	 * terminate things with extreme prejudice.
254 	 */
255 	pr_alert("Unable to handle kernel paging request at virtual "
256 		 "address %08lx\n pc = %08lx, ra = %08lx\n",
257 		 address, regs->pc, regs->areg[0]);
258 	die("Oops", regs, sig);
259 }
260