xref: /linux/arch/powerpc/mm/fault.c (revision 5e8d780d745c1619aba81fe7166c5a4b5cad2b84)
1 /*
2  *  PowerPC version
3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4  *
5  *  Derived from "arch/i386/mm/fault.c"
6  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
7  *
8  *  Modified by Cort Dougan and Paul Mackerras.
9  *
10  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
11  *
12  *  This program is free software; you can redistribute it and/or
13  *  modify it under the terms of the GNU General Public License
14  *  as published by the Free Software Foundation; either version
15  *  2 of the License, or (at your option) any later version.
16  */
17 
18 #include <linux/config.h>
19 #include <linux/signal.h>
20 #include <linux/sched.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/string.h>
24 #include <linux/types.h>
25 #include <linux/ptrace.h>
26 #include <linux/mman.h>
27 #include <linux/mm.h>
28 #include <linux/interrupt.h>
29 #include <linux/highmem.h>
30 #include <linux/module.h>
31 #include <linux/kprobes.h>
32 
33 #include <asm/page.h>
34 #include <asm/pgtable.h>
35 #include <asm/mmu.h>
36 #include <asm/mmu_context.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/tlbflush.h>
40 #include <asm/kdebug.h>
41 #include <asm/siginfo.h>
42 
43 #ifdef CONFIG_KPROBES
44 ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
45 
46 /* Hook to register for page fault notifications */
47 int register_page_fault_notifier(struct notifier_block *nb)
48 {
49 	return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
50 }
51 
52 int unregister_page_fault_notifier(struct notifier_block *nb)
53 {
54 	return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
55 }
56 
57 static inline int notify_page_fault(enum die_val val, const char *str,
58 			struct pt_regs *regs, long err, int trap, int sig)
59 {
60 	struct die_args args = {
61 		.regs = regs,
62 		.str = str,
63 		.err = err,
64 		.trapnr = trap,
65 		.signr = sig
66 	};
67 	return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
68 }
69 #else
70 static inline int notify_page_fault(enum die_val val, const char *str,
71 			struct pt_regs *regs, long err, int trap, int sig)
72 {
73 	return NOTIFY_DONE;
74 }
75 #endif
76 
77 /*
78  * Check whether the instruction at regs->nip is a store using
79  * an update addressing form which will update r1.
80  */
81 static int store_updates_sp(struct pt_regs *regs)
82 {
83 	unsigned int inst;
84 
85 	if (get_user(inst, (unsigned int __user *)regs->nip))
86 		return 0;
87 	/* check for 1 in the rA field */
88 	if (((inst >> 16) & 0x1f) != 1)
89 		return 0;
90 	/* check major opcode */
91 	switch (inst >> 26) {
92 	case 37:	/* stwu */
93 	case 39:	/* stbu */
94 	case 45:	/* sthu */
95 	case 53:	/* stfsu */
96 	case 55:	/* stfdu */
97 		return 1;
98 	case 62:	/* std or stdu */
99 		return (inst & 3) == 1;
100 	case 31:
101 		/* check minor opcode */
102 		switch ((inst >> 1) & 0x3ff) {
103 		case 181:	/* stdux */
104 		case 183:	/* stwux */
105 		case 247:	/* stbux */
106 		case 439:	/* sthux */
107 		case 695:	/* stfsux */
108 		case 759:	/* stfdux */
109 			return 1;
110 		}
111 	}
112 	return 0;
113 }
114 
115 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
116 static void do_dabr(struct pt_regs *regs, unsigned long address,
117 		    unsigned long error_code)
118 {
119 	siginfo_t info;
120 
121 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
122 			11, SIGSEGV) == NOTIFY_STOP)
123 		return;
124 
125 	if (debugger_dabr_match(regs))
126 		return;
127 
128 	/* Clear the DABR */
129 	set_dabr(0);
130 
131 	/* Deliver the signal to userspace */
132 	info.si_signo = SIGTRAP;
133 	info.si_errno = 0;
134 	info.si_code = TRAP_HWBKPT;
135 	info.si_addr = (void __user *)address;
136 	force_sig_info(SIGTRAP, &info, current);
137 }
138 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
139 
140 /*
141  * For 600- and 800-family processors, the error_code parameter is DSISR
142  * for a data fault, SRR1 for an instruction fault. For 400-family processors
143  * the error_code parameter is ESR for a data fault, 0 for an instruction
144  * fault.
145  * For 64-bit processors, the error_code parameter is
146  *  - DSISR for a non-SLB data access fault,
147  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
148  *  - 0 any SLB fault.
149  *
150  * The return value is 0 if the fault was handled, or the signal
151  * number if this is a kernel fault that can't be handled here.
152  */
153 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
154 			    unsigned long error_code)
155 {
156 	struct vm_area_struct * vma;
157 	struct mm_struct *mm = current->mm;
158 	siginfo_t info;
159 	int code = SEGV_MAPERR;
160 	int is_write = 0;
161 	int trap = TRAP(regs);
162  	int is_exec = trap == 0x400;
163 
164 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
165 	/*
166 	 * Fortunately the bit assignments in SRR1 for an instruction
167 	 * fault and DSISR for a data fault are mostly the same for the
168 	 * bits we are interested in.  But there are some bits which
169 	 * indicate errors in DSISR but can validly be set in SRR1.
170 	 */
171 	if (trap == 0x400)
172 		error_code &= 0x48200000;
173 	else
174 		is_write = error_code & DSISR_ISSTORE;
175 #else
176 	is_write = error_code & ESR_DST;
177 #endif /* CONFIG_4xx || CONFIG_BOOKE */
178 
179 	if (notify_page_fault(DIE_PAGE_FAULT, "page_fault", regs, error_code,
180 				11, SIGSEGV) == NOTIFY_STOP)
181 		return 0;
182 
183 	if (trap == 0x300) {
184 		if (debugger_fault_handler(regs))
185 			return 0;
186 	}
187 
188 	/* On a kernel SLB miss we can only check for a valid exception entry */
189 	if (!user_mode(regs) && (address >= TASK_SIZE))
190 		return SIGSEGV;
191 
192 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
193   	if (error_code & DSISR_DABRMATCH) {
194 		/* DABR match */
195 		do_dabr(regs, address, error_code);
196 		return 0;
197 	}
198 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
199 
200 	if (in_atomic() || mm == NULL) {
201 		if (!user_mode(regs))
202 			return SIGSEGV;
203 		/* in_atomic() in user mode is really bad,
204 		   as is current->mm == NULL. */
205 		printk(KERN_EMERG "Page fault in user mode with"
206 		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
207 		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
208 		       regs->nip, regs->msr);
209 		die("Weird page fault", regs, SIGSEGV);
210 	}
211 
212 	/* When running in the kernel we expect faults to occur only to
213 	 * addresses in user space.  All other faults represent errors in the
214 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
215 	 * erroneous fault occurring in a code path which already holds mmap_sem
216 	 * we will deadlock attempting to validate the fault against the
217 	 * address space.  Luckily the kernel only validly references user
218 	 * space from well defined areas of code, which are listed in the
219 	 * exceptions table.
220 	 *
221 	 * As the vast majority of faults will be valid we will only perform
222 	 * the source reference check when there is a possibility of a deadlock.
223 	 * Attempt to lock the address space, if we cannot we then validate the
224 	 * source.  If this is invalid we can skip the address space check,
225 	 * thus avoiding the deadlock.
226 	 */
227 	if (!down_read_trylock(&mm->mmap_sem)) {
228 		if (!user_mode(regs) && !search_exception_tables(regs->nip))
229 			goto bad_area_nosemaphore;
230 
231 		down_read(&mm->mmap_sem);
232 	}
233 
234 	vma = find_vma(mm, address);
235 	if (!vma)
236 		goto bad_area;
237 	if (vma->vm_start <= address)
238 		goto good_area;
239 	if (!(vma->vm_flags & VM_GROWSDOWN))
240 		goto bad_area;
241 
242 	/*
243 	 * N.B. The POWER/Open ABI allows programs to access up to
244 	 * 288 bytes below the stack pointer.
245 	 * The kernel signal delivery code writes up to about 1.5kB
246 	 * below the stack pointer (r1) before decrementing it.
247 	 * The exec code can write slightly over 640kB to the stack
248 	 * before setting the user r1.  Thus we allow the stack to
249 	 * expand to 1MB without further checks.
250 	 */
251 	if (address + 0x100000 < vma->vm_end) {
252 		/* get user regs even if this fault is in kernel mode */
253 		struct pt_regs *uregs = current->thread.regs;
254 		if (uregs == NULL)
255 			goto bad_area;
256 
257 		/*
258 		 * A user-mode access to an address a long way below
259 		 * the stack pointer is only valid if the instruction
260 		 * is one which would update the stack pointer to the
261 		 * address accessed if the instruction completed,
262 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
263 		 * (or the byte, halfword, float or double forms).
264 		 *
265 		 * If we don't check this then any write to the area
266 		 * between the last mapped region and the stack will
267 		 * expand the stack rather than segfaulting.
268 		 */
269 		if (address + 2048 < uregs->gpr[1]
270 		    && (!user_mode(regs) || !store_updates_sp(regs)))
271 			goto bad_area;
272 	}
273 	if (expand_stack(vma, address))
274 		goto bad_area;
275 
276 good_area:
277 	code = SEGV_ACCERR;
278 #if defined(CONFIG_6xx)
279 	if (error_code & 0x95700000)
280 		/* an error such as lwarx to I/O controller space,
281 		   address matching DABR, eciwx, etc. */
282 		goto bad_area;
283 #endif /* CONFIG_6xx */
284 #if defined(CONFIG_8xx)
285         /* The MPC8xx seems to always set 0x80000000, which is
286          * "undefined".  Of those that can be set, this is the only
287          * one which seems bad.
288          */
289 	if (error_code & 0x10000000)
290                 /* Guarded storage error. */
291 		goto bad_area;
292 #endif /* CONFIG_8xx */
293 
294 	if (is_exec) {
295 #ifdef CONFIG_PPC64
296 		/* protection fault */
297 		if (error_code & DSISR_PROTFAULT)
298 			goto bad_area;
299 		if (!(vma->vm_flags & VM_EXEC))
300 			goto bad_area;
301 #endif
302 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
303 		pte_t *ptep;
304 		pmd_t *pmdp;
305 
306 		/* Since 4xx/Book-E supports per-page execute permission,
307 		 * we lazily flush dcache to icache. */
308 		ptep = NULL;
309 		if (get_pteptr(mm, address, &ptep, &pmdp)) {
310 			spinlock_t *ptl = pte_lockptr(mm, pmdp);
311 			spin_lock(ptl);
312 			if (pte_present(*ptep)) {
313 				struct page *page = pte_page(*ptep);
314 
315 				if (!test_bit(PG_arch_1, &page->flags)) {
316 					flush_dcache_icache_page(page);
317 					set_bit(PG_arch_1, &page->flags);
318 				}
319 				pte_update(ptep, 0, _PAGE_HWEXEC);
320 				_tlbie(address);
321 				pte_unmap_unlock(ptep, ptl);
322 				up_read(&mm->mmap_sem);
323 				return 0;
324 			}
325 			pte_unmap_unlock(ptep, ptl);
326 		}
327 #endif
328 	/* a write */
329 	} else if (is_write) {
330 		if (!(vma->vm_flags & VM_WRITE))
331 			goto bad_area;
332 	/* a read */
333 	} else {
334 		/* protection fault */
335 		if (error_code & 0x08000000)
336 			goto bad_area;
337 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
338 			goto bad_area;
339 	}
340 
341 	/*
342 	 * If for any reason at all we couldn't handle the fault,
343 	 * make sure we exit gracefully rather than endlessly redo
344 	 * the fault.
345 	 */
346  survive:
347 	switch (handle_mm_fault(mm, vma, address, is_write)) {
348 
349 	case VM_FAULT_MINOR:
350 		current->min_flt++;
351 		break;
352 	case VM_FAULT_MAJOR:
353 		current->maj_flt++;
354 		break;
355 	case VM_FAULT_SIGBUS:
356 		goto do_sigbus;
357 	case VM_FAULT_OOM:
358 		goto out_of_memory;
359 	default:
360 		BUG();
361 	}
362 
363 	up_read(&mm->mmap_sem);
364 	return 0;
365 
366 bad_area:
367 	up_read(&mm->mmap_sem);
368 
369 bad_area_nosemaphore:
370 	/* User mode accesses cause a SIGSEGV */
371 	if (user_mode(regs)) {
372 		_exception(SIGSEGV, regs, code, address);
373 		return 0;
374 	}
375 
376 	if (is_exec && (error_code & DSISR_PROTFAULT)
377 	    && printk_ratelimit())
378 		printk(KERN_CRIT "kernel tried to execute NX-protected"
379 		       " page (%lx) - exploit attempt? (uid: %d)\n",
380 		       address, current->uid);
381 
382 	return SIGSEGV;
383 
384 /*
385  * We ran out of memory, or some other thing happened to us that made
386  * us unable to handle the page fault gracefully.
387  */
388 out_of_memory:
389 	up_read(&mm->mmap_sem);
390 	if (current->pid == 1) {
391 		yield();
392 		down_read(&mm->mmap_sem);
393 		goto survive;
394 	}
395 	printk("VM: killing process %s\n", current->comm);
396 	if (user_mode(regs))
397 		do_exit(SIGKILL);
398 	return SIGKILL;
399 
400 do_sigbus:
401 	up_read(&mm->mmap_sem);
402 	if (user_mode(regs)) {
403 		info.si_signo = SIGBUS;
404 		info.si_errno = 0;
405 		info.si_code = BUS_ADRERR;
406 		info.si_addr = (void __user *)address;
407 		force_sig_info(SIGBUS, &info, current);
408 		return 0;
409 	}
410 	return SIGBUS;
411 }
412 
413 /*
414  * bad_page_fault is called when we have a bad access from the kernel.
415  * It is called from the DSI and ISI handlers in head.S and from some
416  * of the procedures in traps.c.
417  */
418 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
419 {
420 	const struct exception_table_entry *entry;
421 
422 	/* Are we prepared to handle this fault?  */
423 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
424 		regs->nip = entry->fixup;
425 		return;
426 	}
427 
428 	/* kernel has accessed a bad area */
429 
430 	printk(KERN_ALERT "Unable to handle kernel paging request for ");
431 	switch (regs->trap) {
432 		case 0x300:
433 		case 0x380:
434 			printk("data at address 0x%08lx\n", regs->dar);
435 			break;
436 		case 0x400:
437 		case 0x480:
438 			printk("instruction fetch\n");
439 			break;
440 		default:
441 			printk("unknown fault\n");
442 	}
443 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
444 		regs->nip);
445 
446 	die("Kernel access of bad area", regs, sig);
447 }
448