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_event.h>
33#include <linux/magic.h>
34#include <linux/ratelimit.h>
35
36#include <asm/firmware.h>
37#include <asm/page.h>
38#include <asm/pgtable.h>
39#include <asm/mmu.h>
40#include <asm/mmu_context.h>
41#include <asm/uaccess.h>
42#include <asm/tlbflush.h>
43#include <asm/siginfo.h>
44#include <asm/debug.h>
45#include <mm/mmu_decl.h>
46
47#include "icswx.h"
48
49#ifdef CONFIG_KPROBES
50static inline int notify_page_fault(struct pt_regs *regs)
51{
52	int ret = 0;
53
54	/* kprobe_running() needs smp_processor_id() */
55	if (!user_mode(regs)) {
56		preempt_disable();
57		if (kprobe_running() && kprobe_fault_handler(regs, 11))
58			ret = 1;
59		preempt_enable();
60	}
61
62	return ret;
63}
64#else
65static inline int notify_page_fault(struct pt_regs *regs)
66{
67	return 0;
68}
69#endif
70
71/*
72 * Check whether the instruction at regs->nip is a store using
73 * an update addressing form which will update r1.
74 */
75static int store_updates_sp(struct pt_regs *regs)
76{
77	unsigned int inst;
78
79	if (get_user(inst, (unsigned int __user *)regs->nip))
80		return 0;
81	/* check for 1 in the rA field */
82	if (((inst >> 16) & 0x1f) != 1)
83		return 0;
84	/* check major opcode */
85	switch (inst >> 26) {
86	case 37:	/* stwu */
87	case 39:	/* stbu */
88	case 45:	/* sthu */
89	case 53:	/* stfsu */
90	case 55:	/* stfdu */
91		return 1;
92	case 62:	/* std or stdu */
93		return (inst & 3) == 1;
94	case 31:
95		/* check minor opcode */
96		switch ((inst >> 1) & 0x3ff) {
97		case 181:	/* stdux */
98		case 183:	/* stwux */
99		case 247:	/* stbux */
100		case 439:	/* sthux */
101		case 695:	/* stfsux */
102		case 759:	/* stfdux */
103			return 1;
104		}
105	}
106	return 0;
107}
108/*
109 * do_page_fault error handling helpers
110 */
111
112#define MM_FAULT_RETURN		0
113#define MM_FAULT_CONTINUE	-1
114#define MM_FAULT_ERR(sig)	(sig)
115
116static int out_of_memory(struct pt_regs *regs)
117{
118	/*
119	 * We ran out of memory, or some other thing happened to us that made
120	 * us unable to handle the page fault gracefully.
121	 */
122	up_read(&current->mm->mmap_sem);
123	if (!user_mode(regs))
124		return MM_FAULT_ERR(SIGKILL);
125	pagefault_out_of_memory();
126	return MM_FAULT_RETURN;
127}
128
129static int do_sigbus(struct pt_regs *regs, unsigned long address)
130{
131	siginfo_t info;
132
133	up_read(&current->mm->mmap_sem);
134
135	if (user_mode(regs)) {
136		info.si_signo = SIGBUS;
137		info.si_errno = 0;
138		info.si_code = BUS_ADRERR;
139		info.si_addr = (void __user *)address;
140		force_sig_info(SIGBUS, &info, current);
141		return MM_FAULT_RETURN;
142	}
143	return MM_FAULT_ERR(SIGBUS);
144}
145
146static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
147{
148	/*
149	 * Pagefault was interrupted by SIGKILL. We have no reason to
150	 * continue the pagefault.
151	 */
152	if (fatal_signal_pending(current)) {
153		/*
154		 * If we have retry set, the mmap semaphore will have
155		 * alrady been released in __lock_page_or_retry(). Else
156		 * we release it now.
157		 */
158		if (!(fault & VM_FAULT_RETRY))
159			up_read(&current->mm->mmap_sem);
160		/* Coming from kernel, we need to deal with uaccess fixups */
161		if (user_mode(regs))
162			return MM_FAULT_RETURN;
163		return MM_FAULT_ERR(SIGKILL);
164	}
165
166	/* No fault: be happy */
167	if (!(fault & VM_FAULT_ERROR))
168		return MM_FAULT_CONTINUE;
169
170	/* Out of memory */
171	if (fault & VM_FAULT_OOM)
172		return out_of_memory(regs);
173
174	/* Bus error. x86 handles HWPOISON here, we'll add this if/when
175	 * we support the feature in HW
176	 */
177	if (fault & VM_FAULT_SIGBUS)
178		return do_sigbus(regs, addr);
179
180	/* We don't understand the fault code, this is fatal */
181	BUG();
182	return MM_FAULT_CONTINUE;
183}
184
185/*
186 * For 600- and 800-family processors, the error_code parameter is DSISR
187 * for a data fault, SRR1 for an instruction fault. For 400-family processors
188 * the error_code parameter is ESR for a data fault, 0 for an instruction
189 * fault.
190 * For 64-bit processors, the error_code parameter is
191 *  - DSISR for a non-SLB data access fault,
192 *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
193 *  - 0 any SLB fault.
194 *
195 * The return value is 0 if the fault was handled, or the signal
196 * number if this is a kernel fault that can't be handled here.
197 */
198int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
199			    unsigned long error_code)
200{
201	struct vm_area_struct * vma;
202	struct mm_struct *mm = current->mm;
203	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
204	int code = SEGV_MAPERR;
205	int is_write = 0;
206	int trap = TRAP(regs);
207 	int is_exec = trap == 0x400;
208	int fault;
209
210#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
211	/*
212	 * Fortunately the bit assignments in SRR1 for an instruction
213	 * fault and DSISR for a data fault are mostly the same for the
214	 * bits we are interested in.  But there are some bits which
215	 * indicate errors in DSISR but can validly be set in SRR1.
216	 */
217	if (trap == 0x400)
218		error_code &= 0x48200000;
219	else
220		is_write = error_code & DSISR_ISSTORE;
221#else
222	is_write = error_code & ESR_DST;
223#endif /* CONFIG_4xx || CONFIG_BOOKE */
224
225	if (is_write)
226		flags |= FAULT_FLAG_WRITE;
227
228#ifdef CONFIG_PPC_ICSWX
229	/*
230	 * we need to do this early because this "data storage
231	 * interrupt" does not update the DAR/DEAR so we don't want to
232	 * look at it
233	 */
234	if (error_code & ICSWX_DSI_UCT) {
235		int rc = acop_handle_fault(regs, address, error_code);
236		if (rc)
237			return rc;
238	}
239#endif /* CONFIG_PPC_ICSWX */
240
241	if (notify_page_fault(regs))
242		return 0;
243
244	if (unlikely(debugger_fault_handler(regs)))
245		return 0;
246
247	/* On a kernel SLB miss we can only check for a valid exception entry */
248	if (!user_mode(regs) && (address >= TASK_SIZE))
249		return SIGSEGV;
250
251#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
252			     defined(CONFIG_PPC_BOOK3S_64))
253  	if (error_code & DSISR_DABRMATCH) {
254		/* DABR match */
255		do_dabr(regs, address, error_code);
256		return 0;
257	}
258#endif
259
260	/* We restore the interrupt state now */
261	if (!arch_irq_disabled_regs(regs))
262		local_irq_enable();
263
264	if (in_atomic() || mm == NULL) {
265		if (!user_mode(regs))
266			return SIGSEGV;
267		/* in_atomic() in user mode is really bad,
268		   as is current->mm == NULL. */
269		printk(KERN_EMERG "Page fault in user mode with "
270		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
271		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
272		       regs->nip, regs->msr);
273		die("Weird page fault", regs, SIGSEGV);
274	}
275
276	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
277
278	/* When running in the kernel we expect faults to occur only to
279	 * addresses in user space.  All other faults represent errors in the
280	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
281	 * erroneous fault occurring in a code path which already holds mmap_sem
282	 * we will deadlock attempting to validate the fault against the
283	 * address space.  Luckily the kernel only validly references user
284	 * space from well defined areas of code, which are listed in the
285	 * exceptions table.
286	 *
287	 * As the vast majority of faults will be valid we will only perform
288	 * the source reference check when there is a possibility of a deadlock.
289	 * Attempt to lock the address space, if we cannot we then validate the
290	 * source.  If this is invalid we can skip the address space check,
291	 * thus avoiding the deadlock.
292	 */
293	if (!down_read_trylock(&mm->mmap_sem)) {
294		if (!user_mode(regs) && !search_exception_tables(regs->nip))
295			goto bad_area_nosemaphore;
296
297retry:
298		down_read(&mm->mmap_sem);
299	} else {
300		/*
301		 * The above down_read_trylock() might have succeeded in
302		 * which case we'll have missed the might_sleep() from
303		 * down_read():
304		 */
305		might_sleep();
306	}
307
308	vma = find_vma(mm, address);
309	if (!vma)
310		goto bad_area;
311	if (vma->vm_start <= address)
312		goto good_area;
313	if (!(vma->vm_flags & VM_GROWSDOWN))
314		goto bad_area;
315
316	/*
317	 * N.B. The POWER/Open ABI allows programs to access up to
318	 * 288 bytes below the stack pointer.
319	 * The kernel signal delivery code writes up to about 1.5kB
320	 * below the stack pointer (r1) before decrementing it.
321	 * The exec code can write slightly over 640kB to the stack
322	 * before setting the user r1.  Thus we allow the stack to
323	 * expand to 1MB without further checks.
324	 */
325	if (address + 0x100000 < vma->vm_end) {
326		/* get user regs even if this fault is in kernel mode */
327		struct pt_regs *uregs = current->thread.regs;
328		if (uregs == NULL)
329			goto bad_area;
330
331		/*
332		 * A user-mode access to an address a long way below
333		 * the stack pointer is only valid if the instruction
334		 * is one which would update the stack pointer to the
335		 * address accessed if the instruction completed,
336		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
337		 * (or the byte, halfword, float or double forms).
338		 *
339		 * If we don't check this then any write to the area
340		 * between the last mapped region and the stack will
341		 * expand the stack rather than segfaulting.
342		 */
343		if (address + 2048 < uregs->gpr[1]
344		    && (!user_mode(regs) || !store_updates_sp(regs)))
345			goto bad_area;
346	}
347	if (expand_stack(vma, address))
348		goto bad_area;
349
350good_area:
351	code = SEGV_ACCERR;
352#if defined(CONFIG_6xx)
353	if (error_code & 0x95700000)
354		/* an error such as lwarx to I/O controller space,
355		   address matching DABR, eciwx, etc. */
356		goto bad_area;
357#endif /* CONFIG_6xx */
358#if defined(CONFIG_8xx)
359	/* 8xx sometimes need to load a invalid/non-present TLBs.
360	 * These must be invalidated separately as linux mm don't.
361	 */
362	if (error_code & 0x40000000) /* no translation? */
363		_tlbil_va(address, 0, 0, 0);
364
365        /* The MPC8xx seems to always set 0x80000000, which is
366         * "undefined".  Of those that can be set, this is the only
367         * one which seems bad.
368         */
369	if (error_code & 0x10000000)
370                /* Guarded storage error. */
371		goto bad_area;
372#endif /* CONFIG_8xx */
373
374	if (is_exec) {
375#ifdef CONFIG_PPC_STD_MMU
376		/* Protection fault on exec go straight to failure on
377		 * Hash based MMUs as they either don't support per-page
378		 * execute permission, or if they do, it's handled already
379		 * at the hash level. This test would probably have to
380		 * be removed if we change the way this works to make hash
381		 * processors use the same I/D cache coherency mechanism
382		 * as embedded.
383		 */
384		if (error_code & DSISR_PROTFAULT)
385			goto bad_area;
386#endif /* CONFIG_PPC_STD_MMU */
387
388		/*
389		 * Allow execution from readable areas if the MMU does not
390		 * provide separate controls over reading and executing.
391		 *
392		 * Note: That code used to not be enabled for 4xx/BookE.
393		 * It is now as I/D cache coherency for these is done at
394		 * set_pte_at() time and I see no reason why the test
395		 * below wouldn't be valid on those processors. This -may-
396		 * break programs compiled with a really old ABI though.
397		 */
398		if (!(vma->vm_flags & VM_EXEC) &&
399		    (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
400		     !(vma->vm_flags & (VM_READ | VM_WRITE))))
401			goto bad_area;
402	/* a write */
403	} else if (is_write) {
404		if (!(vma->vm_flags & VM_WRITE))
405			goto bad_area;
406	/* a read */
407	} else {
408		/* protection fault */
409		if (error_code & 0x08000000)
410			goto bad_area;
411		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
412			goto bad_area;
413	}
414
415	/*
416	 * If for any reason at all we couldn't handle the fault,
417	 * make sure we exit gracefully rather than endlessly redo
418	 * the fault.
419	 */
420	fault = handle_mm_fault(mm, vma, address, flags);
421	if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
422		int rc = mm_fault_error(regs, address, fault);
423		if (rc >= MM_FAULT_RETURN)
424			return rc;
425	}
426
427	/*
428	 * Major/minor page fault accounting is only done on the
429	 * initial attempt. If we go through a retry, it is extremely
430	 * likely that the page will be found in page cache at that point.
431	 */
432	if (flags & FAULT_FLAG_ALLOW_RETRY) {
433		if (fault & VM_FAULT_MAJOR) {
434			current->maj_flt++;
435			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
436				      regs, address);
437#ifdef CONFIG_PPC_SMLPAR
438			if (firmware_has_feature(FW_FEATURE_CMO)) {
439				preempt_disable();
440				get_lppaca()->page_ins += (1 << PAGE_FACTOR);
441				preempt_enable();
442			}
443#endif /* CONFIG_PPC_SMLPAR */
444		} else {
445			current->min_flt++;
446			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
447				      regs, address);
448		}
449		if (fault & VM_FAULT_RETRY) {
450			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
451			 * of starvation. */
452			flags &= ~FAULT_FLAG_ALLOW_RETRY;
453			goto retry;
454		}
455	}
456
457	up_read(&mm->mmap_sem);
458	return 0;
459
460bad_area:
461	up_read(&mm->mmap_sem);
462
463bad_area_nosemaphore:
464	/* User mode accesses cause a SIGSEGV */
465	if (user_mode(regs)) {
466		_exception(SIGSEGV, regs, code, address);
467		return 0;
468	}
469
470	if (is_exec && (error_code & DSISR_PROTFAULT))
471		printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
472				   " page (%lx) - exploit attempt? (uid: %d)\n",
473				   address, current_uid());
474
475	return SIGSEGV;
476
477}
478
479/*
480 * bad_page_fault is called when we have a bad access from the kernel.
481 * It is called from the DSI and ISI handlers in head.S and from some
482 * of the procedures in traps.c.
483 */
484void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
485{
486	const struct exception_table_entry *entry;
487	unsigned long *stackend;
488
489	/* Are we prepared to handle this fault?  */
490	if ((entry = search_exception_tables(regs->nip)) != NULL) {
491		regs->nip = entry->fixup;
492		return;
493	}
494
495	/* kernel has accessed a bad area */
496
497	switch (regs->trap) {
498	case 0x300:
499	case 0x380:
500		printk(KERN_ALERT "Unable to handle kernel paging request for "
501			"data at address 0x%08lx\n", regs->dar);
502		break;
503	case 0x400:
504	case 0x480:
505		printk(KERN_ALERT "Unable to handle kernel paging request for "
506			"instruction fetch\n");
507		break;
508	default:
509		printk(KERN_ALERT "Unable to handle kernel paging request for "
510			"unknown fault\n");
511		break;
512	}
513	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
514		regs->nip);
515
516	stackend = end_of_stack(current);
517	if (current != &init_task && *stackend != STACK_END_MAGIC)
518		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
519
520	die("Kernel access of bad area", regs, sig);
521}
522