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/ratelimit.h>
34#include <linux/context_tracking.h>
35#include <linux/hugetlb.h>
36
37#include <asm/firmware.h>
38#include <asm/page.h>
39#include <asm/pgtable.h>
40#include <asm/mmu.h>
41#include <asm/mmu_context.h>
42#include <asm/uaccess.h>
43#include <asm/tlbflush.h>
44#include <asm/siginfo.h>
45#include <asm/debug.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 do_sigbus(struct pt_regs *regs, unsigned long address,
117		     unsigned int fault)
118{
119	siginfo_t info;
120	unsigned int lsb = 0;
121
122	up_read(&current->mm->mmap_sem);
123
124	if (!user_mode(regs))
125		return MM_FAULT_ERR(SIGBUS);
126
127	current->thread.trap_nr = BUS_ADRERR;
128	info.si_signo = SIGBUS;
129	info.si_errno = 0;
130	info.si_code = BUS_ADRERR;
131	info.si_addr = (void __user *)address;
132#ifdef CONFIG_MEMORY_FAILURE
133	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
134		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
135			current->comm, current->pid, address);
136		info.si_code = BUS_MCEERR_AR;
137	}
138
139	if (fault & VM_FAULT_HWPOISON_LARGE)
140		lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
141	if (fault & VM_FAULT_HWPOISON)
142		lsb = PAGE_SHIFT;
143#endif
144	info.si_addr_lsb = lsb;
145	force_sig_info(SIGBUS, &info, current);
146	return MM_FAULT_RETURN;
147}
148
149static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
150{
151	/*
152	 * Pagefault was interrupted by SIGKILL. We have no reason to
153	 * continue the pagefault.
154	 */
155	if (fatal_signal_pending(current)) {
156		/*
157		 * If we have retry set, the mmap semaphore will have
158		 * alrady been released in __lock_page_or_retry(). Else
159		 * we release it now.
160		 */
161		if (!(fault & VM_FAULT_RETRY))
162			up_read(&current->mm->mmap_sem);
163		/* Coming from kernel, we need to deal with uaccess fixups */
164		if (user_mode(regs))
165			return MM_FAULT_RETURN;
166		return MM_FAULT_ERR(SIGKILL);
167	}
168
169	/* No fault: be happy */
170	if (!(fault & VM_FAULT_ERROR))
171		return MM_FAULT_CONTINUE;
172
173	/* Out of memory */
174	if (fault & VM_FAULT_OOM) {
175		up_read(&current->mm->mmap_sem);
176
177		/*
178		 * We ran out of memory, or some other thing happened to us that
179		 * made us unable to handle the page fault gracefully.
180		 */
181		if (!user_mode(regs))
182			return MM_FAULT_ERR(SIGKILL);
183		pagefault_out_of_memory();
184		return MM_FAULT_RETURN;
185	}
186
187	if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE))
188		return do_sigbus(regs, addr, fault);
189
190	/* We don't understand the fault code, this is fatal */
191	BUG();
192	return MM_FAULT_CONTINUE;
193}
194
195/*
196 * For 600- and 800-family processors, the error_code parameter is DSISR
197 * for a data fault, SRR1 for an instruction fault. For 400-family processors
198 * the error_code parameter is ESR for a data fault, 0 for an instruction
199 * fault.
200 * For 64-bit processors, the error_code parameter is
201 *  - DSISR for a non-SLB data access fault,
202 *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
203 *  - 0 any SLB fault.
204 *
205 * The return value is 0 if the fault was handled, or the signal
206 * number if this is a kernel fault that can't be handled here.
207 */
208int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
209			    unsigned long error_code)
210{
211	enum ctx_state prev_state = exception_enter();
212	struct vm_area_struct * vma;
213	struct mm_struct *mm = current->mm;
214	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
215	int code = SEGV_MAPERR;
216	int is_write = 0;
217	int trap = TRAP(regs);
218 	int is_exec = trap == 0x400;
219	int fault;
220	int rc = 0, store_update_sp = 0;
221
222#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
223	/*
224	 * Fortunately the bit assignments in SRR1 for an instruction
225	 * fault and DSISR for a data fault are mostly the same for the
226	 * bits we are interested in.  But there are some bits which
227	 * indicate errors in DSISR but can validly be set in SRR1.
228	 */
229	if (trap == 0x400)
230		error_code &= 0x48200000;
231	else
232		is_write = error_code & DSISR_ISSTORE;
233#else
234	is_write = error_code & ESR_DST;
235#endif /* CONFIG_4xx || CONFIG_BOOKE */
236
237#ifdef CONFIG_PPC_ICSWX
238	/*
239	 * we need to do this early because this "data storage
240	 * interrupt" does not update the DAR/DEAR so we don't want to
241	 * look at it
242	 */
243	if (error_code & ICSWX_DSI_UCT) {
244		rc = acop_handle_fault(regs, address, error_code);
245		if (rc)
246			goto bail;
247	}
248#endif /* CONFIG_PPC_ICSWX */
249
250	if (notify_page_fault(regs))
251		goto bail;
252
253	if (unlikely(debugger_fault_handler(regs)))
254		goto bail;
255
256	/* On a kernel SLB miss we can only check for a valid exception entry */
257	if (!user_mode(regs) && (address >= TASK_SIZE)) {
258		rc = SIGSEGV;
259		goto bail;
260	}
261
262#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
263			     defined(CONFIG_PPC_BOOK3S_64))
264  	if (error_code & DSISR_DABRMATCH) {
265		/* breakpoint match */
266		do_break(regs, address, error_code);
267		goto bail;
268	}
269#endif
270
271	/* We restore the interrupt state now */
272	if (!arch_irq_disabled_regs(regs))
273		local_irq_enable();
274
275	if (in_atomic() || mm == NULL) {
276		if (!user_mode(regs)) {
277			rc = SIGSEGV;
278			goto bail;
279		}
280		/* in_atomic() in user mode is really bad,
281		   as is current->mm == NULL. */
282		printk(KERN_EMERG "Page fault in user mode with "
283		       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
284		printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
285		       regs->nip, regs->msr);
286		die("Weird page fault", regs, SIGSEGV);
287	}
288
289	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
290
291	/*
292	 * We want to do this outside mmap_sem, because reading code around nip
293	 * can result in fault, which will cause a deadlock when called with
294	 * mmap_sem held
295	 */
296	if (user_mode(regs))
297		store_update_sp = store_updates_sp(regs);
298
299	if (user_mode(regs))
300		flags |= FAULT_FLAG_USER;
301
302	/* When running in the kernel we expect faults to occur only to
303	 * addresses in user space.  All other faults represent errors in the
304	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
305	 * erroneous fault occurring in a code path which already holds mmap_sem
306	 * we will deadlock attempting to validate the fault against the
307	 * address space.  Luckily the kernel only validly references user
308	 * space from well defined areas of code, which are listed in the
309	 * exceptions table.
310	 *
311	 * As the vast majority of faults will be valid we will only perform
312	 * the source reference check when there is a possibility of a deadlock.
313	 * Attempt to lock the address space, if we cannot we then validate the
314	 * source.  If this is invalid we can skip the address space check,
315	 * thus avoiding the deadlock.
316	 */
317	if (!down_read_trylock(&mm->mmap_sem)) {
318		if (!user_mode(regs) && !search_exception_tables(regs->nip))
319			goto bad_area_nosemaphore;
320
321retry:
322		down_read(&mm->mmap_sem);
323	} else {
324		/*
325		 * The above down_read_trylock() might have succeeded in
326		 * which case we'll have missed the might_sleep() from
327		 * down_read():
328		 */
329		might_sleep();
330	}
331
332	vma = find_vma(mm, address);
333	if (!vma)
334		goto bad_area;
335	if (vma->vm_start <= address)
336		goto good_area;
337	if (!(vma->vm_flags & VM_GROWSDOWN))
338		goto bad_area;
339
340	/*
341	 * N.B. The POWER/Open ABI allows programs to access up to
342	 * 288 bytes below the stack pointer.
343	 * The kernel signal delivery code writes up to about 1.5kB
344	 * below the stack pointer (r1) before decrementing it.
345	 * The exec code can write slightly over 640kB to the stack
346	 * before setting the user r1.  Thus we allow the stack to
347	 * expand to 1MB without further checks.
348	 */
349	if (address + 0x100000 < vma->vm_end) {
350		/* get user regs even if this fault is in kernel mode */
351		struct pt_regs *uregs = current->thread.regs;
352		if (uregs == NULL)
353			goto bad_area;
354
355		/*
356		 * A user-mode access to an address a long way below
357		 * the stack pointer is only valid if the instruction
358		 * is one which would update the stack pointer to the
359		 * address accessed if the instruction completed,
360		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
361		 * (or the byte, halfword, float or double forms).
362		 *
363		 * If we don't check this then any write to the area
364		 * between the last mapped region and the stack will
365		 * expand the stack rather than segfaulting.
366		 */
367		if (address + 2048 < uregs->gpr[1] && !store_update_sp)
368			goto bad_area;
369	}
370	if (expand_stack(vma, address))
371		goto bad_area;
372
373good_area:
374	code = SEGV_ACCERR;
375#if defined(CONFIG_6xx)
376	if (error_code & 0x95700000)
377		/* an error such as lwarx to I/O controller space,
378		   address matching DABR, eciwx, etc. */
379		goto bad_area;
380#endif /* CONFIG_6xx */
381#if defined(CONFIG_8xx)
382        /* The MPC8xx seems to always set 0x80000000, which is
383         * "undefined".  Of those that can be set, this is the only
384         * one which seems bad.
385         */
386	if (error_code & 0x10000000)
387                /* Guarded storage error. */
388		goto bad_area;
389#endif /* CONFIG_8xx */
390
391	if (is_exec) {
392		/*
393		 * Allow execution from readable areas if the MMU does not
394		 * provide separate controls over reading and executing.
395		 *
396		 * Note: That code used to not be enabled for 4xx/BookE.
397		 * It is now as I/D cache coherency for these is done at
398		 * set_pte_at() time and I see no reason why the test
399		 * below wouldn't be valid on those processors. This -may-
400		 * break programs compiled with a really old ABI though.
401		 */
402		if (!(vma->vm_flags & VM_EXEC) &&
403		    (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
404		     !(vma->vm_flags & (VM_READ | VM_WRITE))))
405			goto bad_area;
406#ifdef CONFIG_PPC_STD_MMU
407		/*
408		 * protfault should only happen due to us
409		 * mapping a region readonly temporarily. PROT_NONE
410		 * is also covered by the VMA check above.
411		 */
412		WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
413#endif /* CONFIG_PPC_STD_MMU */
414	/* a write */
415	} else if (is_write) {
416		if (!(vma->vm_flags & VM_WRITE))
417			goto bad_area;
418		flags |= FAULT_FLAG_WRITE;
419	/* a read */
420	} else {
421		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
422			goto bad_area;
423		WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
424	}
425
426	/*
427	 * If for any reason at all we couldn't handle the fault,
428	 * make sure we exit gracefully rather than endlessly redo
429	 * the fault.
430	 */
431	fault = handle_mm_fault(mm, vma, address, flags);
432	if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
433		if (fault & VM_FAULT_SIGSEGV)
434			goto bad_area;
435		rc = mm_fault_error(regs, address, fault);
436		if (rc >= MM_FAULT_RETURN)
437			goto bail;
438		else
439			rc = 0;
440	}
441
442	/*
443	 * Major/minor page fault accounting is only done on the
444	 * initial attempt. If we go through a retry, it is extremely
445	 * likely that the page will be found in page cache at that point.
446	 */
447	if (flags & FAULT_FLAG_ALLOW_RETRY) {
448		if (fault & VM_FAULT_MAJOR) {
449			current->maj_flt++;
450			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
451				      regs, address);
452#ifdef CONFIG_PPC_SMLPAR
453			if (firmware_has_feature(FW_FEATURE_CMO)) {
454				u32 page_ins;
455
456				preempt_disable();
457				page_ins = be32_to_cpu(get_lppaca()->page_ins);
458				page_ins += 1 << PAGE_FACTOR;
459				get_lppaca()->page_ins = cpu_to_be32(page_ins);
460				preempt_enable();
461			}
462#endif /* CONFIG_PPC_SMLPAR */
463		} else {
464			current->min_flt++;
465			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
466				      regs, address);
467		}
468		if (fault & VM_FAULT_RETRY) {
469			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
470			 * of starvation. */
471			flags &= ~FAULT_FLAG_ALLOW_RETRY;
472			flags |= FAULT_FLAG_TRIED;
473			goto retry;
474		}
475	}
476
477	up_read(&mm->mmap_sem);
478	goto bail;
479
480bad_area:
481	up_read(&mm->mmap_sem);
482
483bad_area_nosemaphore:
484	/* User mode accesses cause a SIGSEGV */
485	if (user_mode(regs)) {
486		_exception(SIGSEGV, regs, code, address);
487		goto bail;
488	}
489
490	if (is_exec && (error_code & DSISR_PROTFAULT))
491		printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
492				   " page (%lx) - exploit attempt? (uid: %d)\n",
493				   address, from_kuid(&init_user_ns, current_uid()));
494
495	rc = SIGSEGV;
496
497bail:
498	exception_exit(prev_state);
499	return rc;
500
501}
502
503/*
504 * bad_page_fault is called when we have a bad access from the kernel.
505 * It is called from the DSI and ISI handlers in head.S and from some
506 * of the procedures in traps.c.
507 */
508void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
509{
510	const struct exception_table_entry *entry;
511
512	/* Are we prepared to handle this fault?  */
513	if ((entry = search_exception_tables(regs->nip)) != NULL) {
514		regs->nip = entry->fixup;
515		return;
516	}
517
518	/* kernel has accessed a bad area */
519
520	switch (regs->trap) {
521	case 0x300:
522	case 0x380:
523		printk(KERN_ALERT "Unable to handle kernel paging request for "
524			"data at address 0x%08lx\n", regs->dar);
525		break;
526	case 0x400:
527	case 0x480:
528		printk(KERN_ALERT "Unable to handle kernel paging request for "
529			"instruction fetch\n");
530		break;
531	default:
532		printk(KERN_ALERT "Unable to handle kernel paging request for "
533			"unknown fault\n");
534		break;
535	}
536	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
537		regs->nip);
538
539	if (task_stack_end_corrupted(current))
540		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
541
542	die("Kernel access of bad area", regs, sig);
543}
544