xref: /openbmc/linux/arch/ia64/mm/fault.c (revision b35565bb)
1 /*
2  * MMU fault handling support.
3  *
4  * Copyright (C) 1998-2002 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  */
7 #include <linux/sched/signal.h>
8 #include <linux/kernel.h>
9 #include <linux/mm.h>
10 #include <linux/extable.h>
11 #include <linux/interrupt.h>
12 #include <linux/kprobes.h>
13 #include <linux/kdebug.h>
14 #include <linux/prefetch.h>
15 #include <linux/uaccess.h>
16 
17 #include <asm/pgtable.h>
18 #include <asm/processor.h>
19 #include <asm/exception.h>
20 
21 extern int die(char *, struct pt_regs *, long);
22 
23 #ifdef CONFIG_KPROBES
24 static inline int notify_page_fault(struct pt_regs *regs, int trap)
25 {
26 	int ret = 0;
27 
28 	if (!user_mode(regs)) {
29 		/* kprobe_running() needs smp_processor_id() */
30 		preempt_disable();
31 		if (kprobe_running() && kprobe_fault_handler(regs, trap))
32 			ret = 1;
33 		preempt_enable();
34 	}
35 
36 	return ret;
37 }
38 #else
39 static inline int notify_page_fault(struct pt_regs *regs, int trap)
40 {
41 	return 0;
42 }
43 #endif
44 
45 /*
46  * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
47  * (inside region 5, on ia64) and that page is present.
48  */
49 static int
50 mapped_kernel_page_is_present (unsigned long address)
51 {
52 	pgd_t *pgd;
53 	pud_t *pud;
54 	pmd_t *pmd;
55 	pte_t *ptep, pte;
56 
57 	pgd = pgd_offset_k(address);
58 	if (pgd_none(*pgd) || pgd_bad(*pgd))
59 		return 0;
60 
61 	pud = pud_offset(pgd, address);
62 	if (pud_none(*pud) || pud_bad(*pud))
63 		return 0;
64 
65 	pmd = pmd_offset(pud, address);
66 	if (pmd_none(*pmd) || pmd_bad(*pmd))
67 		return 0;
68 
69 	ptep = pte_offset_kernel(pmd, address);
70 	if (!ptep)
71 		return 0;
72 
73 	pte = *ptep;
74 	return pte_present(pte);
75 }
76 
77 #	define VM_READ_BIT	0
78 #	define VM_WRITE_BIT	1
79 #	define VM_EXEC_BIT	2
80 
81 void __kprobes
82 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
83 {
84 	int signal = SIGSEGV, code = SEGV_MAPERR;
85 	struct vm_area_struct *vma, *prev_vma;
86 	struct mm_struct *mm = current->mm;
87 	struct siginfo si;
88 	unsigned long mask;
89 	int fault;
90 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
91 
92 	mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
93 		| (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
94 
95 	/* mmap_sem is performance critical.... */
96 	prefetchw(&mm->mmap_sem);
97 
98 	/*
99 	 * If we're in an interrupt or have no user context, we must not take the fault..
100 	 */
101 	if (faulthandler_disabled() || !mm)
102 		goto no_context;
103 
104 #ifdef CONFIG_VIRTUAL_MEM_MAP
105 	/*
106 	 * If fault is in region 5 and we are in the kernel, we may already
107 	 * have the mmap_sem (pfn_valid macro is called during mmap). There
108 	 * is no vma for region 5 addr's anyway, so skip getting the semaphore
109 	 * and go directly to the exception handling code.
110 	 */
111 
112 	if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
113 		goto bad_area_no_up;
114 #endif
115 
116 	/*
117 	 * This is to handle the kprobes on user space access instructions
118 	 */
119 	if (notify_page_fault(regs, TRAP_BRKPT))
120 		return;
121 
122 	if (user_mode(regs))
123 		flags |= FAULT_FLAG_USER;
124 	if (mask & VM_WRITE)
125 		flags |= FAULT_FLAG_WRITE;
126 retry:
127 	down_read(&mm->mmap_sem);
128 
129 	vma = find_vma_prev(mm, address, &prev_vma);
130 	if (!vma && !prev_vma )
131 		goto bad_area;
132 
133         /*
134          * find_vma_prev() returns vma such that address < vma->vm_end or NULL
135          *
136          * May find no vma, but could be that the last vm area is the
137          * register backing store that needs to expand upwards, in
138          * this case vma will be null, but prev_vma will ne non-null
139          */
140         if (( !vma && prev_vma ) || (address < vma->vm_start) )
141 		goto check_expansion;
142 
143   good_area:
144 	code = SEGV_ACCERR;
145 
146 	/* OK, we've got a good vm_area for this memory area.  Check the access permissions: */
147 
148 #	if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
149 	    || (1 << VM_EXEC_BIT) != VM_EXEC)
150 #		error File is out of sync with <linux/mm.h>.  Please update.
151 #	endif
152 
153 	if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
154 		goto bad_area;
155 
156 	if ((vma->vm_flags & mask) != mask)
157 		goto bad_area;
158 
159 	/*
160 	 * If for any reason at all we couldn't handle the fault, make
161 	 * sure we exit gracefully rather than endlessly redo the
162 	 * fault.
163 	 */
164 	fault = handle_mm_fault(vma, address, flags);
165 
166 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
167 		return;
168 
169 	if (unlikely(fault & VM_FAULT_ERROR)) {
170 		/*
171 		 * We ran out of memory, or some other thing happened
172 		 * to us that made us unable to handle the page fault
173 		 * gracefully.
174 		 */
175 		if (fault & VM_FAULT_OOM) {
176 			goto out_of_memory;
177 		} else if (fault & VM_FAULT_SIGSEGV) {
178 			goto bad_area;
179 		} else if (fault & VM_FAULT_SIGBUS) {
180 			signal = SIGBUS;
181 			goto bad_area;
182 		}
183 		BUG();
184 	}
185 
186 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
187 		if (fault & VM_FAULT_MAJOR)
188 			current->maj_flt++;
189 		else
190 			current->min_flt++;
191 		if (fault & VM_FAULT_RETRY) {
192 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
193 			flags |= FAULT_FLAG_TRIED;
194 
195 			 /* No need to up_read(&mm->mmap_sem) as we would
196 			 * have already released it in __lock_page_or_retry
197 			 * in mm/filemap.c.
198 			 */
199 
200 			goto retry;
201 		}
202 	}
203 
204 	up_read(&mm->mmap_sem);
205 	return;
206 
207   check_expansion:
208 	if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
209 		if (!vma)
210 			goto bad_area;
211 		if (!(vma->vm_flags & VM_GROWSDOWN))
212 			goto bad_area;
213 		if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
214 		    || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
215 			goto bad_area;
216 		if (expand_stack(vma, address))
217 			goto bad_area;
218 	} else {
219 		vma = prev_vma;
220 		if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
221 		    || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
222 			goto bad_area;
223 		/*
224 		 * Since the register backing store is accessed sequentially,
225 		 * we disallow growing it by more than a page at a time.
226 		 */
227 		if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
228 			goto bad_area;
229 		if (expand_upwards(vma, address))
230 			goto bad_area;
231 	}
232 	goto good_area;
233 
234   bad_area:
235 	up_read(&mm->mmap_sem);
236 #ifdef CONFIG_VIRTUAL_MEM_MAP
237   bad_area_no_up:
238 #endif
239 	if ((isr & IA64_ISR_SP)
240 	    || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
241 	{
242 		/*
243 		 * This fault was due to a speculative load or lfetch.fault, set the "ed"
244 		 * bit in the psr to ensure forward progress.  (Target register will get a
245 		 * NaT for ld.s, lfetch will be canceled.)
246 		 */
247 		ia64_psr(regs)->ed = 1;
248 		return;
249 	}
250 	if (user_mode(regs)) {
251 		si.si_signo = signal;
252 		si.si_errno = 0;
253 		si.si_code = code;
254 		si.si_addr = (void __user *) address;
255 		si.si_isr = isr;
256 		si.si_flags = __ISR_VALID;
257 		force_sig_info(signal, &si, current);
258 		return;
259 	}
260 
261   no_context:
262 	if ((isr & IA64_ISR_SP)
263 	    || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
264 	{
265 		/*
266 		 * This fault was due to a speculative load or lfetch.fault, set the "ed"
267 		 * bit in the psr to ensure forward progress.  (Target register will get a
268 		 * NaT for ld.s, lfetch will be canceled.)
269 		 */
270 		ia64_psr(regs)->ed = 1;
271 		return;
272 	}
273 
274 	/*
275 	 * Since we have no vma's for region 5, we might get here even if the address is
276 	 * valid, due to the VHPT walker inserting a non present translation that becomes
277 	 * stale. If that happens, the non present fault handler already purged the stale
278 	 * translation, which fixed the problem. So, we check to see if the translation is
279 	 * valid, and return if it is.
280 	 */
281 	if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
282 		return;
283 
284 	if (ia64_done_with_exception(regs))
285 		return;
286 
287 	/*
288 	 * Oops. The kernel tried to access some bad page. We'll have to terminate things
289 	 * with extreme prejudice.
290 	 */
291 	bust_spinlocks(1);
292 
293 	if (address < PAGE_SIZE)
294 		printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
295 	else
296 		printk(KERN_ALERT "Unable to handle kernel paging request at "
297 		       "virtual address %016lx\n", address);
298 	if (die("Oops", regs, isr))
299 		regs = NULL;
300 	bust_spinlocks(0);
301 	if (regs)
302 		do_exit(SIGKILL);
303 	return;
304 
305   out_of_memory:
306 	up_read(&mm->mmap_sem);
307 	if (!user_mode(regs))
308 		goto no_context;
309 	pagefault_out_of_memory();
310 }
311