xref: /openbmc/linux/arch/powerpc/mm/fault.c (revision 78c99ba1)
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/signal.h>
19 #include <linux/sched.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/string.h>
23 #include <linux/types.h>
24 #include <linux/ptrace.h>
25 #include <linux/mman.h>
26 #include <linux/mm.h>
27 #include <linux/interrupt.h>
28 #include <linux/highmem.h>
29 #include <linux/module.h>
30 #include <linux/kprobes.h>
31 #include <linux/kdebug.h>
32 #include <linux/perf_counter.h>
33 
34 #include <asm/firmware.h>
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/siginfo.h>
43 
44 
45 #ifdef CONFIG_KPROBES
46 static inline int notify_page_fault(struct pt_regs *regs)
47 {
48 	int ret = 0;
49 
50 	/* kprobe_running() needs smp_processor_id() */
51 	if (!user_mode(regs)) {
52 		preempt_disable();
53 		if (kprobe_running() && kprobe_fault_handler(regs, 11))
54 			ret = 1;
55 		preempt_enable();
56 	}
57 
58 	return ret;
59 }
60 #else
61 static inline int notify_page_fault(struct pt_regs *regs)
62 {
63 	return 0;
64 }
65 #endif
66 
67 /*
68  * Check whether the instruction at regs->nip is a store using
69  * an update addressing form which will update r1.
70  */
71 static int store_updates_sp(struct pt_regs *regs)
72 {
73 	unsigned int inst;
74 
75 	if (get_user(inst, (unsigned int __user *)regs->nip))
76 		return 0;
77 	/* check for 1 in the rA field */
78 	if (((inst >> 16) & 0x1f) != 1)
79 		return 0;
80 	/* check major opcode */
81 	switch (inst >> 26) {
82 	case 37:	/* stwu */
83 	case 39:	/* stbu */
84 	case 45:	/* sthu */
85 	case 53:	/* stfsu */
86 	case 55:	/* stfdu */
87 		return 1;
88 	case 62:	/* std or stdu */
89 		return (inst & 3) == 1;
90 	case 31:
91 		/* check minor opcode */
92 		switch ((inst >> 1) & 0x3ff) {
93 		case 181:	/* stdux */
94 		case 183:	/* stwux */
95 		case 247:	/* stbux */
96 		case 439:	/* sthux */
97 		case 695:	/* stfsux */
98 		case 759:	/* stfdux */
99 			return 1;
100 		}
101 	}
102 	return 0;
103 }
104 
105 /*
106  * For 600- and 800-family processors, the error_code parameter is DSISR
107  * for a data fault, SRR1 for an instruction fault. For 400-family processors
108  * the error_code parameter is ESR for a data fault, 0 for an instruction
109  * fault.
110  * For 64-bit processors, the error_code parameter is
111  *  - DSISR for a non-SLB data access fault,
112  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
113  *  - 0 any SLB fault.
114  *
115  * The return value is 0 if the fault was handled, or the signal
116  * number if this is a kernel fault that can't be handled here.
117  */
118 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
119 			    unsigned long error_code)
120 {
121 	struct vm_area_struct * vma;
122 	struct mm_struct *mm = current->mm;
123 	siginfo_t info;
124 	int code = SEGV_MAPERR;
125 	int is_write = 0, ret;
126 	int trap = TRAP(regs);
127  	int is_exec = trap == 0x400;
128 
129 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
130 	/*
131 	 * Fortunately the bit assignments in SRR1 for an instruction
132 	 * fault and DSISR for a data fault are mostly the same for the
133 	 * bits we are interested in.  But there are some bits which
134 	 * indicate errors in DSISR but can validly be set in SRR1.
135 	 */
136 	if (trap == 0x400)
137 		error_code &= 0x48200000;
138 	else
139 		is_write = error_code & DSISR_ISSTORE;
140 #else
141 	is_write = error_code & ESR_DST;
142 #endif /* CONFIG_4xx || CONFIG_BOOKE */
143 
144 	if (notify_page_fault(regs))
145 		return 0;
146 
147 	if (unlikely(debugger_fault_handler(regs)))
148 		return 0;
149 
150 	/* On a kernel SLB miss we can only check for a valid exception entry */
151 	if (!user_mode(regs) && (address >= TASK_SIZE))
152 		return SIGSEGV;
153 
154 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
155   	if (error_code & DSISR_DABRMATCH) {
156 		/* DABR match */
157 		do_dabr(regs, address, error_code);
158 		return 0;
159 	}
160 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
161 
162 	if (in_atomic() || mm == NULL) {
163 		if (!user_mode(regs))
164 			return SIGSEGV;
165 		/* in_atomic() in user mode is really bad,
166 		   as is current->mm == NULL. */
167 		printk(KERN_EMERG "Page fault in user mode with "
168 		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
169 		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
170 		       regs->nip, regs->msr);
171 		die("Weird page fault", regs, SIGSEGV);
172 	}
173 
174 	perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
175 
176 	/* When running in the kernel we expect faults to occur only to
177 	 * addresses in user space.  All other faults represent errors in the
178 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
179 	 * erroneous fault occurring in a code path which already holds mmap_sem
180 	 * we will deadlock attempting to validate the fault against the
181 	 * address space.  Luckily the kernel only validly references user
182 	 * space from well defined areas of code, which are listed in the
183 	 * exceptions table.
184 	 *
185 	 * As the vast majority of faults will be valid we will only perform
186 	 * the source reference check when there is a possibility of a deadlock.
187 	 * Attempt to lock the address space, if we cannot we then validate the
188 	 * source.  If this is invalid we can skip the address space check,
189 	 * thus avoiding the deadlock.
190 	 */
191 	if (!down_read_trylock(&mm->mmap_sem)) {
192 		if (!user_mode(regs) && !search_exception_tables(regs->nip))
193 			goto bad_area_nosemaphore;
194 
195 		down_read(&mm->mmap_sem);
196 	}
197 
198 	vma = find_vma(mm, address);
199 	if (!vma)
200 		goto bad_area;
201 	if (vma->vm_start <= address)
202 		goto good_area;
203 	if (!(vma->vm_flags & VM_GROWSDOWN))
204 		goto bad_area;
205 
206 	/*
207 	 * N.B. The POWER/Open ABI allows programs to access up to
208 	 * 288 bytes below the stack pointer.
209 	 * The kernel signal delivery code writes up to about 1.5kB
210 	 * below the stack pointer (r1) before decrementing it.
211 	 * The exec code can write slightly over 640kB to the stack
212 	 * before setting the user r1.  Thus we allow the stack to
213 	 * expand to 1MB without further checks.
214 	 */
215 	if (address + 0x100000 < vma->vm_end) {
216 		/* get user regs even if this fault is in kernel mode */
217 		struct pt_regs *uregs = current->thread.regs;
218 		if (uregs == NULL)
219 			goto bad_area;
220 
221 		/*
222 		 * A user-mode access to an address a long way below
223 		 * the stack pointer is only valid if the instruction
224 		 * is one which would update the stack pointer to the
225 		 * address accessed if the instruction completed,
226 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
227 		 * (or the byte, halfword, float or double forms).
228 		 *
229 		 * If we don't check this then any write to the area
230 		 * between the last mapped region and the stack will
231 		 * expand the stack rather than segfaulting.
232 		 */
233 		if (address + 2048 < uregs->gpr[1]
234 		    && (!user_mode(regs) || !store_updates_sp(regs)))
235 			goto bad_area;
236 	}
237 	if (expand_stack(vma, address))
238 		goto bad_area;
239 
240 good_area:
241 	code = SEGV_ACCERR;
242 #if defined(CONFIG_6xx)
243 	if (error_code & 0x95700000)
244 		/* an error such as lwarx to I/O controller space,
245 		   address matching DABR, eciwx, etc. */
246 		goto bad_area;
247 #endif /* CONFIG_6xx */
248 #if defined(CONFIG_8xx)
249         /* The MPC8xx seems to always set 0x80000000, which is
250          * "undefined".  Of those that can be set, this is the only
251          * one which seems bad.
252          */
253 	if (error_code & 0x10000000)
254                 /* Guarded storage error. */
255 		goto bad_area;
256 #endif /* CONFIG_8xx */
257 
258 	if (is_exec) {
259 #ifdef CONFIG_PPC_STD_MMU
260 		/* Protection fault on exec go straight to failure on
261 		 * Hash based MMUs as they either don't support per-page
262 		 * execute permission, or if they do, it's handled already
263 		 * at the hash level. This test would probably have to
264 		 * be removed if we change the way this works to make hash
265 		 * processors use the same I/D cache coherency mechanism
266 		 * as embedded.
267 		 */
268 		if (error_code & DSISR_PROTFAULT)
269 			goto bad_area;
270 #endif /* CONFIG_PPC_STD_MMU */
271 
272 		/*
273 		 * Allow execution from readable areas if the MMU does not
274 		 * provide separate controls over reading and executing.
275 		 *
276 		 * Note: That code used to not be enabled for 4xx/BookE.
277 		 * It is now as I/D cache coherency for these is done at
278 		 * set_pte_at() time and I see no reason why the test
279 		 * below wouldn't be valid on those processors. This -may-
280 		 * break programs compiled with a really old ABI though.
281 		 */
282 		if (!(vma->vm_flags & VM_EXEC) &&
283 		    (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
284 		     !(vma->vm_flags & (VM_READ | VM_WRITE))))
285 			goto bad_area;
286 	/* a write */
287 	} else if (is_write) {
288 		if (!(vma->vm_flags & VM_WRITE))
289 			goto bad_area;
290 	/* a read */
291 	} else {
292 		/* protection fault */
293 		if (error_code & 0x08000000)
294 			goto bad_area;
295 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
296 			goto bad_area;
297 	}
298 
299 	/*
300 	 * If for any reason at all we couldn't handle the fault,
301 	 * make sure we exit gracefully rather than endlessly redo
302 	 * the fault.
303 	 */
304  survive:
305 	ret = handle_mm_fault(mm, vma, address, is_write);
306 	if (unlikely(ret & VM_FAULT_ERROR)) {
307 		if (ret & VM_FAULT_OOM)
308 			goto out_of_memory;
309 		else if (ret & VM_FAULT_SIGBUS)
310 			goto do_sigbus;
311 		BUG();
312 	}
313 	if (ret & VM_FAULT_MAJOR) {
314 		current->maj_flt++;
315 		perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
316 				     regs, address);
317 #ifdef CONFIG_PPC_SMLPAR
318 		if (firmware_has_feature(FW_FEATURE_CMO)) {
319 			preempt_disable();
320 			get_lppaca()->page_ins += (1 << PAGE_FACTOR);
321 			preempt_enable();
322 		}
323 #endif
324 	} else {
325 		current->min_flt++;
326 		perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
327 				     regs, address);
328 	}
329 	up_read(&mm->mmap_sem);
330 	return 0;
331 
332 bad_area:
333 	up_read(&mm->mmap_sem);
334 
335 bad_area_nosemaphore:
336 	/* User mode accesses cause a SIGSEGV */
337 	if (user_mode(regs)) {
338 		_exception(SIGSEGV, regs, code, address);
339 		return 0;
340 	}
341 
342 	if (is_exec && (error_code & DSISR_PROTFAULT)
343 	    && printk_ratelimit())
344 		printk(KERN_CRIT "kernel tried to execute NX-protected"
345 		       " page (%lx) - exploit attempt? (uid: %d)\n",
346 		       address, current_uid());
347 
348 	return SIGSEGV;
349 
350 /*
351  * We ran out of memory, or some other thing happened to us that made
352  * us unable to handle the page fault gracefully.
353  */
354 out_of_memory:
355 	up_read(&mm->mmap_sem);
356 	if (is_global_init(current)) {
357 		yield();
358 		down_read(&mm->mmap_sem);
359 		goto survive;
360 	}
361 	printk("VM: killing process %s\n", current->comm);
362 	if (user_mode(regs))
363 		do_group_exit(SIGKILL);
364 	return SIGKILL;
365 
366 do_sigbus:
367 	up_read(&mm->mmap_sem);
368 	if (user_mode(regs)) {
369 		info.si_signo = SIGBUS;
370 		info.si_errno = 0;
371 		info.si_code = BUS_ADRERR;
372 		info.si_addr = (void __user *)address;
373 		force_sig_info(SIGBUS, &info, current);
374 		return 0;
375 	}
376 	return SIGBUS;
377 }
378 
379 /*
380  * bad_page_fault is called when we have a bad access from the kernel.
381  * It is called from the DSI and ISI handlers in head.S and from some
382  * of the procedures in traps.c.
383  */
384 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
385 {
386 	const struct exception_table_entry *entry;
387 
388 	/* Are we prepared to handle this fault?  */
389 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
390 		regs->nip = entry->fixup;
391 		return;
392 	}
393 
394 	/* kernel has accessed a bad area */
395 
396 	switch (regs->trap) {
397 	case 0x300:
398 	case 0x380:
399 		printk(KERN_ALERT "Unable to handle kernel paging request for "
400 			"data at address 0x%08lx\n", regs->dar);
401 		break;
402 	case 0x400:
403 	case 0x480:
404 		printk(KERN_ALERT "Unable to handle kernel paging request for "
405 			"instruction fetch\n");
406 		break;
407 	default:
408 		printk(KERN_ALERT "Unable to handle kernel paging request for "
409 			"unknown fault\n");
410 		break;
411 	}
412 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
413 		regs->nip);
414 
415 	die("Kernel access of bad area", regs, sig);
416 }
417