xref: /linux/arch/powerpc/mm/fault.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
1 /*
2  *  arch/ppc/mm/fault.c
3  *
4  *  PowerPC version
5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6  *
7  *  Derived from "arch/i386/mm/fault.c"
8  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
9  *
10  *  Modified by Cort Dougan and Paul Mackerras.
11  *
12  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
13  *
14  *  This program is free software; you can redistribute it and/or
15  *  modify it under the terms of the GNU General Public License
16  *  as published by the Free Software Foundation; either version
17  *  2 of the License, or (at your option) any later version.
18  */
19 
20 #include <linux/config.h>
21 #include <linux/signal.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/string.h>
26 #include <linux/types.h>
27 #include <linux/ptrace.h>
28 #include <linux/mman.h>
29 #include <linux/mm.h>
30 #include <linux/interrupt.h>
31 #include <linux/highmem.h>
32 #include <linux/module.h>
33 #include <linux/kprobes.h>
34 
35 #include <asm/page.h>
36 #include <asm/pgtable.h>
37 #include <asm/mmu.h>
38 #include <asm/mmu_context.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
41 #include <asm/tlbflush.h>
42 #include <asm/kdebug.h>
43 #include <asm/siginfo.h>
44 
45 /*
46  * Check whether the instruction at regs->nip is a store using
47  * an update addressing form which will update r1.
48  */
49 static int store_updates_sp(struct pt_regs *regs)
50 {
51 	unsigned int inst;
52 
53 	if (get_user(inst, (unsigned int __user *)regs->nip))
54 		return 0;
55 	/* check for 1 in the rA field */
56 	if (((inst >> 16) & 0x1f) != 1)
57 		return 0;
58 	/* check major opcode */
59 	switch (inst >> 26) {
60 	case 37:	/* stwu */
61 	case 39:	/* stbu */
62 	case 45:	/* sthu */
63 	case 53:	/* stfsu */
64 	case 55:	/* stfdu */
65 		return 1;
66 	case 62:	/* std or stdu */
67 		return (inst & 3) == 1;
68 	case 31:
69 		/* check minor opcode */
70 		switch ((inst >> 1) & 0x3ff) {
71 		case 181:	/* stdux */
72 		case 183:	/* stwux */
73 		case 247:	/* stbux */
74 		case 439:	/* sthux */
75 		case 695:	/* stfsux */
76 		case 759:	/* stfdux */
77 			return 1;
78 		}
79 	}
80 	return 0;
81 }
82 
83 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
84 static void do_dabr(struct pt_regs *regs, unsigned long address,
85 		    unsigned long error_code)
86 {
87 	siginfo_t info;
88 
89 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
90 			11, SIGSEGV) == NOTIFY_STOP)
91 		return;
92 
93 	if (debugger_dabr_match(regs))
94 		return;
95 
96 	/* Clear the DABR */
97 	set_dabr(0);
98 
99 	/* Deliver the signal to userspace */
100 	info.si_signo = SIGTRAP;
101 	info.si_errno = 0;
102 	info.si_code = TRAP_HWBKPT;
103 	info.si_addr = (void __user *)address;
104 	force_sig_info(SIGTRAP, &info, current);
105 }
106 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
107 
108 /*
109  * For 600- and 800-family processors, the error_code parameter is DSISR
110  * for a data fault, SRR1 for an instruction fault. For 400-family processors
111  * the error_code parameter is ESR for a data fault, 0 for an instruction
112  * fault.
113  * For 64-bit processors, the error_code parameter is
114  *  - DSISR for a non-SLB data access fault,
115  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
116  *  - 0 any SLB fault.
117  *
118  * The return value is 0 if the fault was handled, or the signal
119  * number if this is a kernel fault that can't be handled here.
120  */
121 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
122 			    unsigned long error_code)
123 {
124 	struct vm_area_struct * vma;
125 	struct mm_struct *mm = current->mm;
126 	siginfo_t info;
127 	int code = SEGV_MAPERR;
128 	int is_write = 0;
129 	int trap = TRAP(regs);
130  	int is_exec = trap == 0x400;
131 
132 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
133 	/*
134 	 * Fortunately the bit assignments in SRR1 for an instruction
135 	 * fault and DSISR for a data fault are mostly the same for the
136 	 * bits we are interested in.  But there are some bits which
137 	 * indicate errors in DSISR but can validly be set in SRR1.
138 	 */
139 	if (trap == 0x400)
140 		error_code &= 0x48200000;
141 	else
142 		is_write = error_code & DSISR_ISSTORE;
143 #else
144 	is_write = error_code & ESR_DST;
145 #endif /* CONFIG_4xx || CONFIG_BOOKE */
146 
147 	if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
148 				11, SIGSEGV) == NOTIFY_STOP)
149 		return 0;
150 
151 	if (trap == 0x300) {
152 		if (debugger_fault_handler(regs))
153 			return 0;
154 	}
155 
156 	/* On a kernel SLB miss we can only check for a valid exception entry */
157 	if (!user_mode(regs) && (address >= TASK_SIZE))
158 		return SIGSEGV;
159 
160 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
161   	if (error_code & DSISR_DABRMATCH) {
162 		/* DABR match */
163 		do_dabr(regs, address, error_code);
164 		return 0;
165 	}
166 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
167 
168 	if (in_atomic() || mm == NULL) {
169 		if (!user_mode(regs))
170 			return SIGSEGV;
171 		/* in_atomic() in user mode is really bad,
172 		   as is current->mm == NULL. */
173 		printk(KERN_EMERG "Page fault in user mode with"
174 		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
175 		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
176 		       regs->nip, regs->msr);
177 		die("Weird page fault", regs, SIGSEGV);
178 	}
179 
180 	/* When running in the kernel we expect faults to occur only to
181 	 * addresses in user space.  All other faults represent errors in the
182 	 * kernel and should generate an OOPS.  Unfortunatly, in the case of an
183 	 * erroneous fault occuring in a code path which already holds mmap_sem
184 	 * we will deadlock attempting to validate the fault against the
185 	 * address space.  Luckily the kernel only validly references user
186 	 * space from well defined areas of code, which are listed in the
187 	 * exceptions table.
188 	 *
189 	 * As the vast majority of faults will be valid we will only perform
190 	 * the source reference check when there is a possibilty of a deadlock.
191 	 * Attempt to lock the address space, if we cannot we then validate the
192 	 * source.  If this is invalid we can skip the address space check,
193 	 * thus avoiding the deadlock.
194 	 */
195 	if (!down_read_trylock(&mm->mmap_sem)) {
196 		if (!user_mode(regs) && !search_exception_tables(regs->nip))
197 			goto bad_area_nosemaphore;
198 
199 		down_read(&mm->mmap_sem);
200 	}
201 
202 	vma = find_vma(mm, address);
203 	if (!vma)
204 		goto bad_area;
205 	if (vma->vm_start <= address)
206 		goto good_area;
207 	if (!(vma->vm_flags & VM_GROWSDOWN))
208 		goto bad_area;
209 
210 	/*
211 	 * N.B. The POWER/Open ABI allows programs to access up to
212 	 * 288 bytes below the stack pointer.
213 	 * The kernel signal delivery code writes up to about 1.5kB
214 	 * below the stack pointer (r1) before decrementing it.
215 	 * The exec code can write slightly over 640kB to the stack
216 	 * before setting the user r1.  Thus we allow the stack to
217 	 * expand to 1MB without further checks.
218 	 */
219 	if (address + 0x100000 < vma->vm_end) {
220 		/* get user regs even if this fault is in kernel mode */
221 		struct pt_regs *uregs = current->thread.regs;
222 		if (uregs == NULL)
223 			goto bad_area;
224 
225 		/*
226 		 * A user-mode access to an address a long way below
227 		 * the stack pointer is only valid if the instruction
228 		 * is one which would update the stack pointer to the
229 		 * address accessed if the instruction completed,
230 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
231 		 * (or the byte, halfword, float or double forms).
232 		 *
233 		 * If we don't check this then any write to the area
234 		 * between the last mapped region and the stack will
235 		 * expand the stack rather than segfaulting.
236 		 */
237 		if (address + 2048 < uregs->gpr[1]
238 		    && (!user_mode(regs) || !store_updates_sp(regs)))
239 			goto bad_area;
240 	}
241 	if (expand_stack(vma, address))
242 		goto bad_area;
243 
244 good_area:
245 	code = SEGV_ACCERR;
246 #if defined(CONFIG_6xx)
247 	if (error_code & 0x95700000)
248 		/* an error such as lwarx to I/O controller space,
249 		   address matching DABR, eciwx, etc. */
250 		goto bad_area;
251 #endif /* CONFIG_6xx */
252 #if defined(CONFIG_8xx)
253         /* The MPC8xx seems to always set 0x80000000, which is
254          * "undefined".  Of those that can be set, this is the only
255          * one which seems bad.
256          */
257 	if (error_code & 0x10000000)
258                 /* Guarded storage error. */
259 		goto bad_area;
260 #endif /* CONFIG_8xx */
261 
262 	if (is_exec) {
263 #ifdef CONFIG_PPC64
264 		/* protection fault */
265 		if (error_code & DSISR_PROTFAULT)
266 			goto bad_area;
267 		if (!(vma->vm_flags & VM_EXEC))
268 			goto bad_area;
269 #endif
270 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
271 		pte_t *ptep;
272 
273 		/* Since 4xx/Book-E supports per-page execute permission,
274 		 * we lazily flush dcache to icache. */
275 		ptep = NULL;
276 		if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
277 			struct page *page = pte_page(*ptep);
278 
279 			if (! test_bit(PG_arch_1, &page->flags)) {
280 				flush_dcache_icache_page(page);
281 				set_bit(PG_arch_1, &page->flags);
282 			}
283 			pte_update(ptep, 0, _PAGE_HWEXEC);
284 			_tlbie(address);
285 			pte_unmap(ptep);
286 			up_read(&mm->mmap_sem);
287 			return 0;
288 		}
289 		if (ptep != NULL)
290 			pte_unmap(ptep);
291 #endif
292 	/* a write */
293 	} else if (is_write) {
294 		if (!(vma->vm_flags & VM_WRITE))
295 			goto bad_area;
296 	/* a read */
297 	} else {
298 		/* protection fault */
299 		if (error_code & 0x08000000)
300 			goto bad_area;
301 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
302 			goto bad_area;
303 	}
304 
305 	/*
306 	 * If for any reason at all we couldn't handle the fault,
307 	 * make sure we exit gracefully rather than endlessly redo
308 	 * the fault.
309 	 */
310  survive:
311 	switch (handle_mm_fault(mm, vma, address, is_write)) {
312 
313 	case VM_FAULT_MINOR:
314 		current->min_flt++;
315 		break;
316 	case VM_FAULT_MAJOR:
317 		current->maj_flt++;
318 		break;
319 	case VM_FAULT_SIGBUS:
320 		goto do_sigbus;
321 	case VM_FAULT_OOM:
322 		goto out_of_memory;
323 	default:
324 		BUG();
325 	}
326 
327 	up_read(&mm->mmap_sem);
328 	return 0;
329 
330 bad_area:
331 	up_read(&mm->mmap_sem);
332 
333 bad_area_nosemaphore:
334 	/* User mode accesses cause a SIGSEGV */
335 	if (user_mode(regs)) {
336 		_exception(SIGSEGV, regs, code, address);
337 		return 0;
338 	}
339 
340 	if (is_exec && (error_code & DSISR_PROTFAULT)
341 	    && printk_ratelimit())
342 		printk(KERN_CRIT "kernel tried to execute NX-protected"
343 		       " page (%lx) - exploit attempt? (uid: %d)\n",
344 		       address, current->uid);
345 
346 	return SIGSEGV;
347 
348 /*
349  * We ran out of memory, or some other thing happened to us that made
350  * us unable to handle the page fault gracefully.
351  */
352 out_of_memory:
353 	up_read(&mm->mmap_sem);
354 	if (current->pid == 1) {
355 		yield();
356 		down_read(&mm->mmap_sem);
357 		goto survive;
358 	}
359 	printk("VM: killing process %s\n", current->comm);
360 	if (user_mode(regs))
361 		do_exit(SIGKILL);
362 	return SIGKILL;
363 
364 do_sigbus:
365 	up_read(&mm->mmap_sem);
366 	if (user_mode(regs)) {
367 		info.si_signo = SIGBUS;
368 		info.si_errno = 0;
369 		info.si_code = BUS_ADRERR;
370 		info.si_addr = (void __user *)address;
371 		force_sig_info(SIGBUS, &info, current);
372 		return 0;
373 	}
374 	return SIGBUS;
375 }
376 
377 /*
378  * bad_page_fault is called when we have a bad access from the kernel.
379  * It is called from the DSI and ISI handlers in head.S and from some
380  * of the procedures in traps.c.
381  */
382 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
383 {
384 	const struct exception_table_entry *entry;
385 
386 	/* Are we prepared to handle this fault?  */
387 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
388 		regs->nip = entry->fixup;
389 		return;
390 	}
391 
392 	/* kernel has accessed a bad area */
393 
394 	printk(KERN_ALERT "Unable to handle kernel paging request for ");
395 	switch (regs->trap) {
396 		case 0x300:
397 		case 0x380:
398 			printk("data at address 0x%08lx\n", regs->dar);
399 			break;
400 		case 0x400:
401 		case 0x480:
402 			printk("instruction fetch\n");
403 			break;
404 		default:
405 			printk("unknown fault\n");
406 	}
407 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
408 		regs->nip);
409 
410 	die("Kernel access of bad area", regs, sig);
411 }
412