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