xref: /openbmc/linux/arch/ia64/mm/fault.c (revision 5a0e3ad6)
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 
14 #include <asm/pgtable.h>
15 #include <asm/processor.h>
16 #include <asm/system.h>
17 #include <asm/uaccess.h>
18 
19 extern int die(char *, struct pt_regs *, long);
20 
21 #ifdef CONFIG_KPROBES
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
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
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 void __kprobes
76 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
77 {
78 	int signal = SIGSEGV, code = SEGV_MAPERR;
79 	struct vm_area_struct *vma, *prev_vma;
80 	struct mm_struct *mm = current->mm;
81 	struct siginfo si;
82 	unsigned long mask;
83 	int fault;
84 
85 	/* mmap_sem is performance critical.... */
86 	prefetchw(&mm->mmap_sem);
87 
88 	/*
89 	 * If we're in an interrupt or have no user context, we must not take the fault..
90 	 */
91 	if (in_atomic() || !mm)
92 		goto no_context;
93 
94 #ifdef CONFIG_VIRTUAL_MEM_MAP
95 	/*
96 	 * If fault is in region 5 and we are in the kernel, we may already
97 	 * have the mmap_sem (pfn_valid macro is called during mmap). There
98 	 * is no vma for region 5 addr's anyway, so skip getting the semaphore
99 	 * and go directly to the exception handling code.
100 	 */
101 
102 	if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
103 		goto bad_area_no_up;
104 #endif
105 
106 	/*
107 	 * This is to handle the kprobes on user space access instructions
108 	 */
109 	if (notify_page_fault(regs, TRAP_BRKPT))
110 		return;
111 
112 	down_read(&mm->mmap_sem);
113 
114 	vma = find_vma_prev(mm, address, &prev_vma);
115 	if (!vma && !prev_vma )
116 		goto bad_area;
117 
118         /*
119          * find_vma_prev() returns vma such that address < vma->vm_end or NULL
120          *
121          * May find no vma, but could be that the last vm area is the
122          * register backing store that needs to expand upwards, in
123          * this case vma will be null, but prev_vma will ne non-null
124          */
125         if (( !vma && prev_vma ) || (address < vma->vm_start) )
126 		goto check_expansion;
127 
128   good_area:
129 	code = SEGV_ACCERR;
130 
131 	/* OK, we've got a good vm_area for this memory area.  Check the access permissions: */
132 
133 #	define VM_READ_BIT	0
134 #	define VM_WRITE_BIT	1
135 #	define VM_EXEC_BIT	2
136 
137 #	if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
138 	    || (1 << VM_EXEC_BIT) != VM_EXEC)
139 #		error File is out of sync with <linux/mm.h>.  Please update.
140 #	endif
141 
142 	if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
143 		goto bad_area;
144 
145 	mask = (  (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
146 		| (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
147 
148 	if ((vma->vm_flags & mask) != mask)
149 		goto bad_area;
150 
151   survive:
152 	/*
153 	 * If for any reason at all we couldn't handle the fault, make
154 	 * sure we exit gracefully rather than endlessly redo the
155 	 * fault.
156 	 */
157 	fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0);
158 	if (unlikely(fault & VM_FAULT_ERROR)) {
159 		/*
160 		 * We ran out of memory, or some other thing happened
161 		 * to us that made us unable to handle the page fault
162 		 * gracefully.
163 		 */
164 		if (fault & VM_FAULT_OOM) {
165 			goto out_of_memory;
166 		} else if (fault & VM_FAULT_SIGBUS) {
167 			signal = SIGBUS;
168 			goto bad_area;
169 		}
170 		BUG();
171 	}
172 	if (fault & VM_FAULT_MAJOR)
173 		current->maj_flt++;
174 	else
175 		current->min_flt++;
176 	up_read(&mm->mmap_sem);
177 	return;
178 
179   check_expansion:
180 	if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
181 		if (!vma)
182 			goto bad_area;
183 		if (!(vma->vm_flags & VM_GROWSDOWN))
184 			goto bad_area;
185 		if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
186 		    || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
187 			goto bad_area;
188 		if (expand_stack(vma, address))
189 			goto bad_area;
190 	} else {
191 		vma = prev_vma;
192 		if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
193 		    || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
194 			goto bad_area;
195 		/*
196 		 * Since the register backing store is accessed sequentially,
197 		 * we disallow growing it by more than a page at a time.
198 		 */
199 		if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
200 			goto bad_area;
201 		if (expand_upwards(vma, address))
202 			goto bad_area;
203 	}
204 	goto good_area;
205 
206   bad_area:
207 	up_read(&mm->mmap_sem);
208 #ifdef CONFIG_VIRTUAL_MEM_MAP
209   bad_area_no_up:
210 #endif
211 	if ((isr & IA64_ISR_SP)
212 	    || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
213 	{
214 		/*
215 		 * This fault was due to a speculative load or lfetch.fault, set the "ed"
216 		 * bit in the psr to ensure forward progress.  (Target register will get a
217 		 * NaT for ld.s, lfetch will be canceled.)
218 		 */
219 		ia64_psr(regs)->ed = 1;
220 		return;
221 	}
222 	if (user_mode(regs)) {
223 		si.si_signo = signal;
224 		si.si_errno = 0;
225 		si.si_code = code;
226 		si.si_addr = (void __user *) address;
227 		si.si_isr = isr;
228 		si.si_flags = __ISR_VALID;
229 		force_sig_info(signal, &si, current);
230 		return;
231 	}
232 
233   no_context:
234 	if ((isr & IA64_ISR_SP)
235 	    || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
236 	{
237 		/*
238 		 * This fault was due to a speculative load or lfetch.fault, set the "ed"
239 		 * bit in the psr to ensure forward progress.  (Target register will get a
240 		 * NaT for ld.s, lfetch will be canceled.)
241 		 */
242 		ia64_psr(regs)->ed = 1;
243 		return;
244 	}
245 
246 	/*
247 	 * Since we have no vma's for region 5, we might get here even if the address is
248 	 * valid, due to the VHPT walker inserting a non present translation that becomes
249 	 * stale. If that happens, the non present fault handler already purged the stale
250 	 * translation, which fixed the problem. So, we check to see if the translation is
251 	 * valid, and return if it is.
252 	 */
253 	if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
254 		return;
255 
256 	if (ia64_done_with_exception(regs))
257 		return;
258 
259 	/*
260 	 * Oops. The kernel tried to access some bad page. We'll have to terminate things
261 	 * with extreme prejudice.
262 	 */
263 	bust_spinlocks(1);
264 
265 	if (address < PAGE_SIZE)
266 		printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
267 	else
268 		printk(KERN_ALERT "Unable to handle kernel paging request at "
269 		       "virtual address %016lx\n", address);
270 	if (die("Oops", regs, isr))
271 		regs = NULL;
272 	bust_spinlocks(0);
273 	if (regs)
274 		do_exit(SIGKILL);
275 	return;
276 
277   out_of_memory:
278 	up_read(&mm->mmap_sem);
279 	if (is_global_init(current)) {
280 		yield();
281 		down_read(&mm->mmap_sem);
282 		goto survive;
283 	}
284 	printk(KERN_CRIT "VM: killing process %s\n", current->comm);
285 	if (user_mode(regs))
286 		do_group_exit(SIGKILL);
287 	goto no_context;
288 }
289