xref: /openbmc/linux/arch/powerpc/mm/fault.c (revision f15cbe6f1a4b4d9df59142fc8e4abb973302cf44)
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 
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/siginfo.h>
41 
42 
43 #ifdef CONFIG_KPROBES
44 static inline int notify_page_fault(struct pt_regs *regs)
45 {
46 	int ret = 0;
47 
48 	/* kprobe_running() needs smp_processor_id() */
49 	if (!user_mode(regs)) {
50 		preempt_disable();
51 		if (kprobe_running() && kprobe_fault_handler(regs, 11))
52 			ret = 1;
53 		preempt_enable();
54 	}
55 
56 	return ret;
57 }
58 #else
59 static inline int notify_page_fault(struct pt_regs *regs)
60 {
61 	return 0;
62 }
63 #endif
64 
65 /*
66  * Check whether the instruction at regs->nip is a store using
67  * an update addressing form which will update r1.
68  */
69 static int store_updates_sp(struct pt_regs *regs)
70 {
71 	unsigned int inst;
72 
73 	if (get_user(inst, (unsigned int __user *)regs->nip))
74 		return 0;
75 	/* check for 1 in the rA field */
76 	if (((inst >> 16) & 0x1f) != 1)
77 		return 0;
78 	/* check major opcode */
79 	switch (inst >> 26) {
80 	case 37:	/* stwu */
81 	case 39:	/* stbu */
82 	case 45:	/* sthu */
83 	case 53:	/* stfsu */
84 	case 55:	/* stfdu */
85 		return 1;
86 	case 62:	/* std or stdu */
87 		return (inst & 3) == 1;
88 	case 31:
89 		/* check minor opcode */
90 		switch ((inst >> 1) & 0x3ff) {
91 		case 181:	/* stdux */
92 		case 183:	/* stwux */
93 		case 247:	/* stbux */
94 		case 439:	/* sthux */
95 		case 695:	/* stfsux */
96 		case 759:	/* stfdux */
97 			return 1;
98 		}
99 	}
100 	return 0;
101 }
102 
103 /*
104  * For 600- and 800-family processors, the error_code parameter is DSISR
105  * for a data fault, SRR1 for an instruction fault. For 400-family processors
106  * the error_code parameter is ESR for a data fault, 0 for an instruction
107  * fault.
108  * For 64-bit processors, the error_code parameter is
109  *  - DSISR for a non-SLB data access fault,
110  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
111  *  - 0 any SLB fault.
112  *
113  * The return value is 0 if the fault was handled, or the signal
114  * number if this is a kernel fault that can't be handled here.
115  */
116 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
117 			    unsigned long error_code)
118 {
119 	struct vm_area_struct * vma;
120 	struct mm_struct *mm = current->mm;
121 	siginfo_t info;
122 	int code = SEGV_MAPERR;
123 	int is_write = 0, ret;
124 	int trap = TRAP(regs);
125  	int is_exec = trap == 0x400;
126 
127 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
128 	/*
129 	 * Fortunately the bit assignments in SRR1 for an instruction
130 	 * fault and DSISR for a data fault are mostly the same for the
131 	 * bits we are interested in.  But there are some bits which
132 	 * indicate errors in DSISR but can validly be set in SRR1.
133 	 */
134 	if (trap == 0x400)
135 		error_code &= 0x48200000;
136 	else
137 		is_write = error_code & DSISR_ISSTORE;
138 #else
139 	is_write = error_code & ESR_DST;
140 #endif /* CONFIG_4xx || CONFIG_BOOKE */
141 
142 	if (notify_page_fault(regs))
143 		return 0;
144 
145 	if (unlikely(debugger_fault_handler(regs)))
146 		return 0;
147 
148 	/* On a kernel SLB miss we can only check for a valid exception entry */
149 	if (!user_mode(regs) && (address >= TASK_SIZE))
150 		return SIGSEGV;
151 
152 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
153   	if (error_code & DSISR_DABRMATCH) {
154 		/* DABR match */
155 		do_dabr(regs, address, error_code);
156 		return 0;
157 	}
158 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
159 
160 	if (in_atomic() || mm == NULL) {
161 		if (!user_mode(regs))
162 			return SIGSEGV;
163 		/* in_atomic() in user mode is really bad,
164 		   as is current->mm == NULL. */
165 		printk(KERN_EMERG "Page fault in user mode with "
166 		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
167 		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
168 		       regs->nip, regs->msr);
169 		die("Weird page fault", regs, SIGSEGV);
170 	}
171 
172 	/* When running in the kernel we expect faults to occur only to
173 	 * addresses in user space.  All other faults represent errors in the
174 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
175 	 * erroneous fault occurring in a code path which already holds mmap_sem
176 	 * we will deadlock attempting to validate the fault against the
177 	 * address space.  Luckily the kernel only validly references user
178 	 * space from well defined areas of code, which are listed in the
179 	 * exceptions table.
180 	 *
181 	 * As the vast majority of faults will be valid we will only perform
182 	 * the source reference check when there is a possibility of a deadlock.
183 	 * Attempt to lock the address space, if we cannot we then validate the
184 	 * source.  If this is invalid we can skip the address space check,
185 	 * thus avoiding the deadlock.
186 	 */
187 	if (!down_read_trylock(&mm->mmap_sem)) {
188 		if (!user_mode(regs) && !search_exception_tables(regs->nip))
189 			goto bad_area_nosemaphore;
190 
191 		down_read(&mm->mmap_sem);
192 	}
193 
194 	vma = find_vma(mm, address);
195 	if (!vma)
196 		goto bad_area;
197 	if (vma->vm_start <= address)
198 		goto good_area;
199 	if (!(vma->vm_flags & VM_GROWSDOWN))
200 		goto bad_area;
201 
202 	/*
203 	 * N.B. The POWER/Open ABI allows programs to access up to
204 	 * 288 bytes below the stack pointer.
205 	 * The kernel signal delivery code writes up to about 1.5kB
206 	 * below the stack pointer (r1) before decrementing it.
207 	 * The exec code can write slightly over 640kB to the stack
208 	 * before setting the user r1.  Thus we allow the stack to
209 	 * expand to 1MB without further checks.
210 	 */
211 	if (address + 0x100000 < vma->vm_end) {
212 		/* get user regs even if this fault is in kernel mode */
213 		struct pt_regs *uregs = current->thread.regs;
214 		if (uregs == NULL)
215 			goto bad_area;
216 
217 		/*
218 		 * A user-mode access to an address a long way below
219 		 * the stack pointer is only valid if the instruction
220 		 * is one which would update the stack pointer to the
221 		 * address accessed if the instruction completed,
222 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
223 		 * (or the byte, halfword, float or double forms).
224 		 *
225 		 * If we don't check this then any write to the area
226 		 * between the last mapped region and the stack will
227 		 * expand the stack rather than segfaulting.
228 		 */
229 		if (address + 2048 < uregs->gpr[1]
230 		    && (!user_mode(regs) || !store_updates_sp(regs)))
231 			goto bad_area;
232 	}
233 	if (expand_stack(vma, address))
234 		goto bad_area;
235 
236 good_area:
237 	code = SEGV_ACCERR;
238 #if defined(CONFIG_6xx)
239 	if (error_code & 0x95700000)
240 		/* an error such as lwarx to I/O controller space,
241 		   address matching DABR, eciwx, etc. */
242 		goto bad_area;
243 #endif /* CONFIG_6xx */
244 #if defined(CONFIG_8xx)
245         /* The MPC8xx seems to always set 0x80000000, which is
246          * "undefined".  Of those that can be set, this is the only
247          * one which seems bad.
248          */
249 	if (error_code & 0x10000000)
250                 /* Guarded storage error. */
251 		goto bad_area;
252 #endif /* CONFIG_8xx */
253 
254 	if (is_exec) {
255 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
256 		/* protection fault */
257 		if (error_code & DSISR_PROTFAULT)
258 			goto bad_area;
259 		/*
260 		 * Allow execution from readable areas if the MMU does not
261 		 * provide separate controls over reading and executing.
262 		 */
263 		if (!(vma->vm_flags & VM_EXEC) &&
264 		    (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
265 		     !(vma->vm_flags & (VM_READ | VM_WRITE))))
266 			goto bad_area;
267 #else
268 		pte_t *ptep;
269 		pmd_t *pmdp;
270 
271 		/* Since 4xx/Book-E supports per-page execute permission,
272 		 * we lazily flush dcache to icache. */
273 		ptep = NULL;
274 		if (get_pteptr(mm, address, &ptep, &pmdp)) {
275 			spinlock_t *ptl = pte_lockptr(mm, pmdp);
276 			spin_lock(ptl);
277 			if (pte_present(*ptep)) {
278 				struct page *page = pte_page(*ptep);
279 
280 				if (!test_bit(PG_arch_1, &page->flags)) {
281 					flush_dcache_icache_page(page);
282 					set_bit(PG_arch_1, &page->flags);
283 				}
284 				pte_update(ptep, 0, _PAGE_HWEXEC |
285 					   _PAGE_ACCESSED);
286 				_tlbie(address, mm->context.id);
287 				pte_unmap_unlock(ptep, ptl);
288 				up_read(&mm->mmap_sem);
289 				return 0;
290 			}
291 			pte_unmap_unlock(ptep, ptl);
292 		}
293 #endif
294 	/* a write */
295 	} else if (is_write) {
296 		if (!(vma->vm_flags & VM_WRITE))
297 			goto bad_area;
298 	/* a read */
299 	} else {
300 		/* protection fault */
301 		if (error_code & 0x08000000)
302 			goto bad_area;
303 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
304 			goto bad_area;
305 	}
306 
307 	/*
308 	 * If for any reason at all we couldn't handle the fault,
309 	 * make sure we exit gracefully rather than endlessly redo
310 	 * the fault.
311 	 */
312  survive:
313 	ret = handle_mm_fault(mm, vma, address, is_write);
314 	if (unlikely(ret & VM_FAULT_ERROR)) {
315 		if (ret & VM_FAULT_OOM)
316 			goto out_of_memory;
317 		else if (ret & VM_FAULT_SIGBUS)
318 			goto do_sigbus;
319 		BUG();
320 	}
321 	if (ret & VM_FAULT_MAJOR)
322 		current->maj_flt++;
323 	else
324 		current->min_flt++;
325 	up_read(&mm->mmap_sem);
326 	return 0;
327 
328 bad_area:
329 	up_read(&mm->mmap_sem);
330 
331 bad_area_nosemaphore:
332 	/* User mode accesses cause a SIGSEGV */
333 	if (user_mode(regs)) {
334 		_exception(SIGSEGV, regs, code, address);
335 		return 0;
336 	}
337 
338 	if (is_exec && (error_code & DSISR_PROTFAULT)
339 	    && printk_ratelimit())
340 		printk(KERN_CRIT "kernel tried to execute NX-protected"
341 		       " page (%lx) - exploit attempt? (uid: %d)\n",
342 		       address, current->uid);
343 
344 	return SIGSEGV;
345 
346 /*
347  * We ran out of memory, or some other thing happened to us that made
348  * us unable to handle the page fault gracefully.
349  */
350 out_of_memory:
351 	up_read(&mm->mmap_sem);
352 	if (is_global_init(current)) {
353 		yield();
354 		down_read(&mm->mmap_sem);
355 		goto survive;
356 	}
357 	printk("VM: killing process %s\n", current->comm);
358 	if (user_mode(regs))
359 		do_group_exit(SIGKILL);
360 	return SIGKILL;
361 
362 do_sigbus:
363 	up_read(&mm->mmap_sem);
364 	if (user_mode(regs)) {
365 		info.si_signo = SIGBUS;
366 		info.si_errno = 0;
367 		info.si_code = BUS_ADRERR;
368 		info.si_addr = (void __user *)address;
369 		force_sig_info(SIGBUS, &info, current);
370 		return 0;
371 	}
372 	return SIGBUS;
373 }
374 
375 /*
376  * bad_page_fault is called when we have a bad access from the kernel.
377  * It is called from the DSI and ISI handlers in head.S and from some
378  * of the procedures in traps.c.
379  */
380 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
381 {
382 	const struct exception_table_entry *entry;
383 
384 	/* Are we prepared to handle this fault?  */
385 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
386 		regs->nip = entry->fixup;
387 		return;
388 	}
389 
390 	/* kernel has accessed a bad area */
391 
392 	switch (regs->trap) {
393 	case 0x300:
394 	case 0x380:
395 		printk(KERN_ALERT "Unable to handle kernel paging request for "
396 			"data at address 0x%08lx\n", regs->dar);
397 		break;
398 	case 0x400:
399 	case 0x480:
400 		printk(KERN_ALERT "Unable to handle kernel paging request for "
401 			"instruction fetch\n");
402 		break;
403 	default:
404 		printk(KERN_ALERT "Unable to handle kernel paging request for "
405 			"unknown fault\n");
406 		break;
407 	}
408 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
409 		regs->nip);
410 
411 	die("Kernel access of bad area", regs, sig);
412 }
413