xref: /linux/arch/powerpc/mm/fault.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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/sched/task_stack.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/extable.h>
31 #include <linux/kprobes.h>
32 #include <linux/kdebug.h>
33 #include <linux/perf_event.h>
34 #include <linux/ratelimit.h>
35 #include <linux/context_tracking.h>
36 #include <linux/hugetlb.h>
37 #include <linux/uaccess.h>
38 
39 #include <asm/firmware.h>
40 #include <asm/page.h>
41 #include <asm/pgtable.h>
42 #include <asm/mmu.h>
43 #include <asm/mmu_context.h>
44 #include <asm/tlbflush.h>
45 #include <asm/siginfo.h>
46 #include <asm/debug.h>
47 
48 static inline bool notify_page_fault(struct pt_regs *regs)
49 {
50 	bool ret = false;
51 
52 #ifdef CONFIG_KPROBES
53 	/* kprobe_running() needs smp_processor_id() */
54 	if (!user_mode(regs)) {
55 		preempt_disable();
56 		if (kprobe_running() && kprobe_fault_handler(regs, 11))
57 			ret = true;
58 		preempt_enable();
59 	}
60 #endif /* CONFIG_KPROBES */
61 
62 	if (unlikely(debugger_fault_handler(regs)))
63 		ret = true;
64 
65 	return ret;
66 }
67 
68 /*
69  * Check whether the instruction at regs->nip is a store using
70  * an update addressing form which will update r1.
71  */
72 static bool store_updates_sp(struct pt_regs *regs)
73 {
74 	unsigned int inst;
75 
76 	if (get_user(inst, (unsigned int __user *)regs->nip))
77 		return false;
78 	/* check for 1 in the rA field */
79 	if (((inst >> 16) & 0x1f) != 1)
80 		return false;
81 	/* check major opcode */
82 	switch (inst >> 26) {
83 	case 37:	/* stwu */
84 	case 39:	/* stbu */
85 	case 45:	/* sthu */
86 	case 53:	/* stfsu */
87 	case 55:	/* stfdu */
88 		return true;
89 	case 62:	/* std or stdu */
90 		return (inst & 3) == 1;
91 	case 31:
92 		/* check minor opcode */
93 		switch ((inst >> 1) & 0x3ff) {
94 		case 181:	/* stdux */
95 		case 183:	/* stwux */
96 		case 247:	/* stbux */
97 		case 439:	/* sthux */
98 		case 695:	/* stfsux */
99 		case 759:	/* stfdux */
100 			return true;
101 		}
102 	}
103 	return false;
104 }
105 /*
106  * do_page_fault error handling helpers
107  */
108 
109 static int
110 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
111 {
112 	/*
113 	 * If we are in kernel mode, bail out with a SEGV, this will
114 	 * be caught by the assembly which will restore the non-volatile
115 	 * registers before calling bad_page_fault()
116 	 */
117 	if (!user_mode(regs))
118 		return SIGSEGV;
119 
120 	_exception(SIGSEGV, regs, si_code, address);
121 
122 	return 0;
123 }
124 
125 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
126 {
127 	return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
128 }
129 
130 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
131 {
132 	struct mm_struct *mm = current->mm;
133 
134 	/*
135 	 * Something tried to access memory that isn't in our memory map..
136 	 * Fix it, but check if it's kernel or user first..
137 	 */
138 	up_read(&mm->mmap_sem);
139 
140 	return __bad_area_nosemaphore(regs, address, si_code);
141 }
142 
143 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
144 {
145 	return __bad_area(regs, address, SEGV_MAPERR);
146 }
147 
148 static int do_sigbus(struct pt_regs *regs, unsigned long address,
149 		     unsigned int fault)
150 {
151 	siginfo_t info;
152 	unsigned int lsb = 0;
153 
154 	if (!user_mode(regs))
155 		return SIGBUS;
156 
157 	current->thread.trap_nr = BUS_ADRERR;
158 	info.si_signo = SIGBUS;
159 	info.si_errno = 0;
160 	info.si_code = BUS_ADRERR;
161 	info.si_addr = (void __user *)address;
162 #ifdef CONFIG_MEMORY_FAILURE
163 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
164 		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
165 			current->comm, current->pid, address);
166 		info.si_code = BUS_MCEERR_AR;
167 	}
168 
169 	if (fault & VM_FAULT_HWPOISON_LARGE)
170 		lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
171 	if (fault & VM_FAULT_HWPOISON)
172 		lsb = PAGE_SHIFT;
173 #endif
174 	info.si_addr_lsb = lsb;
175 	force_sig_info(SIGBUS, &info, current);
176 	return 0;
177 }
178 
179 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
180 {
181 	/*
182 	 * Kernel page fault interrupted by SIGKILL. We have no reason to
183 	 * continue processing.
184 	 */
185 	if (fatal_signal_pending(current) && !user_mode(regs))
186 		return SIGKILL;
187 
188 	/* Out of memory */
189 	if (fault & VM_FAULT_OOM) {
190 		/*
191 		 * We ran out of memory, or some other thing happened to us that
192 		 * made us unable to handle the page fault gracefully.
193 		 */
194 		if (!user_mode(regs))
195 			return SIGSEGV;
196 		pagefault_out_of_memory();
197 	} else {
198 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
199 			     VM_FAULT_HWPOISON_LARGE))
200 			return do_sigbus(regs, addr, fault);
201 		else if (fault & VM_FAULT_SIGSEGV)
202 			return bad_area_nosemaphore(regs, addr);
203 		else
204 			BUG();
205 	}
206 	return 0;
207 }
208 
209 /* Is this a bad kernel fault ? */
210 static bool bad_kernel_fault(bool is_exec, unsigned long error_code,
211 			     unsigned long address)
212 {
213 	if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT))) {
214 		printk_ratelimited(KERN_CRIT "kernel tried to execute"
215 				   " exec-protected page (%lx) -"
216 				   "exploit attempt? (uid: %d)\n",
217 				   address, from_kuid(&init_user_ns,
218 						      current_uid()));
219 	}
220 	return is_exec || (address >= TASK_SIZE);
221 }
222 
223 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
224 				struct vm_area_struct *vma,
225 				bool store_update_sp)
226 {
227 	/*
228 	 * N.B. The POWER/Open ABI allows programs to access up to
229 	 * 288 bytes below the stack pointer.
230 	 * The kernel signal delivery code writes up to about 1.5kB
231 	 * below the stack pointer (r1) before decrementing it.
232 	 * The exec code can write slightly over 640kB to the stack
233 	 * before setting the user r1.  Thus we allow the stack to
234 	 * expand to 1MB without further checks.
235 	 */
236 	if (address + 0x100000 < vma->vm_end) {
237 		/* get user regs even if this fault is in kernel mode */
238 		struct pt_regs *uregs = current->thread.regs;
239 		if (uregs == NULL)
240 			return true;
241 
242 		/*
243 		 * A user-mode access to an address a long way below
244 		 * the stack pointer is only valid if the instruction
245 		 * is one which would update the stack pointer to the
246 		 * address accessed if the instruction completed,
247 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
248 		 * (or the byte, halfword, float or double forms).
249 		 *
250 		 * If we don't check this then any write to the area
251 		 * between the last mapped region and the stack will
252 		 * expand the stack rather than segfaulting.
253 		 */
254 		if (address + 2048 < uregs->gpr[1] && !store_update_sp)
255 			return true;
256 	}
257 	return false;
258 }
259 
260 static bool access_error(bool is_write, bool is_exec,
261 			 struct vm_area_struct *vma)
262 {
263 	/*
264 	 * Allow execution from readable areas if the MMU does not
265 	 * provide separate controls over reading and executing.
266 	 *
267 	 * Note: That code used to not be enabled for 4xx/BookE.
268 	 * It is now as I/D cache coherency for these is done at
269 	 * set_pte_at() time and I see no reason why the test
270 	 * below wouldn't be valid on those processors. This -may-
271 	 * break programs compiled with a really old ABI though.
272 	 */
273 	if (is_exec) {
274 		return !(vma->vm_flags & VM_EXEC) &&
275 			(cpu_has_feature(CPU_FTR_NOEXECUTE) ||
276 			 !(vma->vm_flags & (VM_READ | VM_WRITE)));
277 	}
278 
279 	if (is_write) {
280 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
281 			return true;
282 		return false;
283 	}
284 
285 	if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
286 		return true;
287 
288 	return false;
289 }
290 
291 #ifdef CONFIG_PPC_SMLPAR
292 static inline void cmo_account_page_fault(void)
293 {
294 	if (firmware_has_feature(FW_FEATURE_CMO)) {
295 		u32 page_ins;
296 
297 		preempt_disable();
298 		page_ins = be32_to_cpu(get_lppaca()->page_ins);
299 		page_ins += 1 << PAGE_FACTOR;
300 		get_lppaca()->page_ins = cpu_to_be32(page_ins);
301 		preempt_enable();
302 	}
303 }
304 #else
305 static inline void cmo_account_page_fault(void) { }
306 #endif /* CONFIG_PPC_SMLPAR */
307 
308 #ifdef CONFIG_PPC_STD_MMU
309 static void sanity_check_fault(bool is_write, unsigned long error_code)
310 {
311 	/*
312 	 * For hash translation mode, we should never get a
313 	 * PROTFAULT. Any update to pte to reduce access will result in us
314 	 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
315 	 * fault instead of DSISR_PROTFAULT.
316 	 *
317 	 * A pte update to relax the access will not result in a hash page table
318 	 * entry invalidate and hence can result in DSISR_PROTFAULT.
319 	 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
320 	 * the special !is_write in the below conditional.
321 	 *
322 	 * For platforms that doesn't supports coherent icache and do support
323 	 * per page noexec bit, we do setup things such that we do the
324 	 * sync between D/I cache via fault. But that is handled via low level
325 	 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
326 	 * here in such case.
327 	 *
328 	 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
329 	 * check should handle those and hence we should fall to the bad_area
330 	 * handling correctly.
331 	 *
332 	 * For embedded with per page exec support that doesn't support coherent
333 	 * icache we do get PROTFAULT and we handle that D/I cache sync in
334 	 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
335 	 * is conditional for server MMU.
336 	 *
337 	 * For radix, we can get prot fault for autonuma case, because radix
338 	 * page table will have them marked noaccess for user.
339 	 */
340 	if (!radix_enabled() && !is_write)
341 		WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
342 }
343 #else
344 static void sanity_check_fault(bool is_write, unsigned long error_code) { }
345 #endif /* CONFIG_PPC_STD_MMU */
346 
347 /*
348  * Define the correct "is_write" bit in error_code based
349  * on the processor family
350  */
351 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
352 #define page_fault_is_write(__err)	((__err) & ESR_DST)
353 #define page_fault_is_bad(__err)	(0)
354 #else
355 #define page_fault_is_write(__err)	((__err) & DSISR_ISSTORE)
356 #if defined(CONFIG_PPC_8xx)
357 #define page_fault_is_bad(__err)	((__err) & DSISR_NOEXEC_OR_G)
358 #elif defined(CONFIG_PPC64)
359 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_64S)
360 #else
361 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_32S)
362 #endif
363 #endif
364 
365 /*
366  * For 600- and 800-family processors, the error_code parameter is DSISR
367  * for a data fault, SRR1 for an instruction fault. For 400-family processors
368  * the error_code parameter is ESR for a data fault, 0 for an instruction
369  * fault.
370  * For 64-bit processors, the error_code parameter is
371  *  - DSISR for a non-SLB data access fault,
372  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
373  *  - 0 any SLB fault.
374  *
375  * The return value is 0 if the fault was handled, or the signal
376  * number if this is a kernel fault that can't be handled here.
377  */
378 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
379 			   unsigned long error_code)
380 {
381 	struct vm_area_struct * vma;
382 	struct mm_struct *mm = current->mm;
383 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
384  	int is_exec = TRAP(regs) == 0x400;
385 	int is_user = user_mode(regs);
386 	int is_write = page_fault_is_write(error_code);
387 	int fault, major = 0;
388 	bool store_update_sp = false;
389 
390 	if (notify_page_fault(regs))
391 		return 0;
392 
393 	if (unlikely(page_fault_is_bad(error_code))) {
394 		if (is_user) {
395 			_exception(SIGBUS, regs, BUS_OBJERR, address);
396 			return 0;
397 		}
398 		return SIGBUS;
399 	}
400 
401 	/* Additional sanity check(s) */
402 	sanity_check_fault(is_write, error_code);
403 
404 	/*
405 	 * The kernel should never take an execute fault nor should it
406 	 * take a page fault to a kernel address.
407 	 */
408 	if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
409 		return SIGSEGV;
410 
411 	/*
412 	 * If we're in an interrupt, have no user context or are running
413 	 * in a region with pagefaults disabled then we must not take the fault
414 	 */
415 	if (unlikely(faulthandler_disabled() || !mm)) {
416 		if (is_user)
417 			printk_ratelimited(KERN_ERR "Page fault in user mode"
418 					   " with faulthandler_disabled()=%d"
419 					   " mm=%p\n",
420 					   faulthandler_disabled(), mm);
421 		return bad_area_nosemaphore(regs, address);
422 	}
423 
424 	/* We restore the interrupt state now */
425 	if (!arch_irq_disabled_regs(regs))
426 		local_irq_enable();
427 
428 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
429 
430 	/*
431 	 * We want to do this outside mmap_sem, because reading code around nip
432 	 * can result in fault, which will cause a deadlock when called with
433 	 * mmap_sem held
434 	 */
435 	if (is_write && is_user)
436 		store_update_sp = store_updates_sp(regs);
437 
438 	if (is_user)
439 		flags |= FAULT_FLAG_USER;
440 	if (is_write)
441 		flags |= FAULT_FLAG_WRITE;
442 	if (is_exec)
443 		flags |= FAULT_FLAG_INSTRUCTION;
444 
445 	/* When running in the kernel we expect faults to occur only to
446 	 * addresses in user space.  All other faults represent errors in the
447 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
448 	 * erroneous fault occurring in a code path which already holds mmap_sem
449 	 * we will deadlock attempting to validate the fault against the
450 	 * address space.  Luckily the kernel only validly references user
451 	 * space from well defined areas of code, which are listed in the
452 	 * exceptions table.
453 	 *
454 	 * As the vast majority of faults will be valid we will only perform
455 	 * the source reference check when there is a possibility of a deadlock.
456 	 * Attempt to lock the address space, if we cannot we then validate the
457 	 * source.  If this is invalid we can skip the address space check,
458 	 * thus avoiding the deadlock.
459 	 */
460 	if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
461 		if (!is_user && !search_exception_tables(regs->nip))
462 			return bad_area_nosemaphore(regs, address);
463 
464 retry:
465 		down_read(&mm->mmap_sem);
466 	} else {
467 		/*
468 		 * The above down_read_trylock() might have succeeded in
469 		 * which case we'll have missed the might_sleep() from
470 		 * down_read():
471 		 */
472 		might_sleep();
473 	}
474 
475 	vma = find_vma(mm, address);
476 	if (unlikely(!vma))
477 		return bad_area(regs, address);
478 	if (likely(vma->vm_start <= address))
479 		goto good_area;
480 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
481 		return bad_area(regs, address);
482 
483 	/* The stack is being expanded, check if it's valid */
484 	if (unlikely(bad_stack_expansion(regs, address, vma, store_update_sp)))
485 		return bad_area(regs, address);
486 
487 	/* Try to expand it */
488 	if (unlikely(expand_stack(vma, address)))
489 		return bad_area(regs, address);
490 
491 good_area:
492 	if (unlikely(access_error(is_write, is_exec, vma)))
493 		return bad_area(regs, address);
494 
495 	/*
496 	 * If for any reason at all we couldn't handle the fault,
497 	 * make sure we exit gracefully rather than endlessly redo
498 	 * the fault.
499 	 */
500 	fault = handle_mm_fault(vma, address, flags);
501 	major |= fault & VM_FAULT_MAJOR;
502 
503 	/*
504 	 * Handle the retry right now, the mmap_sem has been released in that
505 	 * case.
506 	 */
507 	if (unlikely(fault & VM_FAULT_RETRY)) {
508 		/* We retry only once */
509 		if (flags & FAULT_FLAG_ALLOW_RETRY) {
510 			/*
511 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
512 			 * of starvation.
513 			 */
514 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
515 			flags |= FAULT_FLAG_TRIED;
516 			if (!fatal_signal_pending(current))
517 				goto retry;
518 		}
519 
520 		/*
521 		 * User mode? Just return to handle the fatal exception otherwise
522 		 * return to bad_page_fault
523 		 */
524 		return is_user ? 0 : SIGBUS;
525 	}
526 
527 	up_read(&current->mm->mmap_sem);
528 
529 	if (unlikely(fault & VM_FAULT_ERROR))
530 		return mm_fault_error(regs, address, fault);
531 
532 	/*
533 	 * Major/minor page fault accounting.
534 	 */
535 	if (major) {
536 		current->maj_flt++;
537 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
538 		cmo_account_page_fault();
539 	} else {
540 		current->min_flt++;
541 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
542 	}
543 	return 0;
544 }
545 NOKPROBE_SYMBOL(__do_page_fault);
546 
547 int do_page_fault(struct pt_regs *regs, unsigned long address,
548 		  unsigned long error_code)
549 {
550 	enum ctx_state prev_state = exception_enter();
551 	int rc = __do_page_fault(regs, address, error_code);
552 	exception_exit(prev_state);
553 	return rc;
554 }
555 NOKPROBE_SYMBOL(do_page_fault);
556 
557 /*
558  * bad_page_fault is called when we have a bad access from the kernel.
559  * It is called from the DSI and ISI handlers in head.S and from some
560  * of the procedures in traps.c.
561  */
562 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
563 {
564 	const struct exception_table_entry *entry;
565 
566 	/* Are we prepared to handle this fault?  */
567 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
568 		regs->nip = extable_fixup(entry);
569 		return;
570 	}
571 
572 	/* kernel has accessed a bad area */
573 
574 	switch (regs->trap) {
575 	case 0x300:
576 	case 0x380:
577 		printk(KERN_ALERT "Unable to handle kernel paging request for "
578 			"data at address 0x%08lx\n", regs->dar);
579 		break;
580 	case 0x400:
581 	case 0x480:
582 		printk(KERN_ALERT "Unable to handle kernel paging request for "
583 			"instruction fetch\n");
584 		break;
585 	case 0x600:
586 		printk(KERN_ALERT "Unable to handle kernel paging request for "
587 			"unaligned access at address 0x%08lx\n", regs->dar);
588 		break;
589 	default:
590 		printk(KERN_ALERT "Unable to handle kernel paging request for "
591 			"unknown fault\n");
592 		break;
593 	}
594 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
595 		regs->nip);
596 
597 	if (task_stack_end_corrupted(current))
598 		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
599 
600 	die("Kernel access of bad area", regs, sig);
601 }
602