Loading...
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
48static 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 */
72static 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
109static int
110__bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code,
111 int pkey)
112{
113 /*
114 * If we are in kernel mode, bail out with a SEGV, this will
115 * be caught by the assembly which will restore the non-volatile
116 * registers before calling bad_page_fault()
117 */
118 if (!user_mode(regs))
119 return SIGSEGV;
120
121 _exception_pkey(SIGSEGV, regs, si_code, address, pkey);
122
123 return 0;
124}
125
126static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
127{
128 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR, 0);
129}
130
131static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code,
132 int pkey)
133{
134 struct mm_struct *mm = current->mm;
135
136 /*
137 * Something tried to access memory that isn't in our memory map..
138 * Fix it, but check if it's kernel or user first..
139 */
140 up_read(&mm->mmap_sem);
141
142 return __bad_area_nosemaphore(regs, address, si_code, pkey);
143}
144
145static noinline int bad_area(struct pt_regs *regs, unsigned long address)
146{
147 return __bad_area(regs, address, SEGV_MAPERR, 0);
148}
149
150static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address,
151 int pkey)
152{
153 return __bad_area_nosemaphore(regs, address, SEGV_PKUERR, pkey);
154}
155
156static noinline int bad_access(struct pt_regs *regs, unsigned long address)
157{
158 return __bad_area(regs, address, SEGV_ACCERR, 0);
159}
160
161static int do_sigbus(struct pt_regs *regs, unsigned long address,
162 unsigned int fault)
163{
164 siginfo_t info;
165 unsigned int lsb = 0;
166
167 if (!user_mode(regs))
168 return SIGBUS;
169
170 current->thread.trap_nr = BUS_ADRERR;
171 info.si_signo = SIGBUS;
172 info.si_errno = 0;
173 info.si_code = BUS_ADRERR;
174 info.si_addr = (void __user *)address;
175#ifdef CONFIG_MEMORY_FAILURE
176 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
177 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
178 current->comm, current->pid, address);
179 info.si_code = BUS_MCEERR_AR;
180 }
181
182 if (fault & VM_FAULT_HWPOISON_LARGE)
183 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
184 if (fault & VM_FAULT_HWPOISON)
185 lsb = PAGE_SHIFT;
186#endif
187 info.si_addr_lsb = lsb;
188 force_sig_info(SIGBUS, &info, current);
189 return 0;
190}
191
192static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
193{
194 /*
195 * Kernel page fault interrupted by SIGKILL. We have no reason to
196 * continue processing.
197 */
198 if (fatal_signal_pending(current) && !user_mode(regs))
199 return SIGKILL;
200
201 /* Out of memory */
202 if (fault & VM_FAULT_OOM) {
203 /*
204 * We ran out of memory, or some other thing happened to us that
205 * made us unable to handle the page fault gracefully.
206 */
207 if (!user_mode(regs))
208 return SIGSEGV;
209 pagefault_out_of_memory();
210 } else {
211 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
212 VM_FAULT_HWPOISON_LARGE))
213 return do_sigbus(regs, addr, fault);
214 else if (fault & VM_FAULT_SIGSEGV)
215 return bad_area_nosemaphore(regs, addr);
216 else
217 BUG();
218 }
219 return 0;
220}
221
222/* Is this a bad kernel fault ? */
223static bool bad_kernel_fault(bool is_exec, unsigned long error_code,
224 unsigned long address)
225{
226 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT))) {
227 printk_ratelimited(KERN_CRIT "kernel tried to execute"
228 " exec-protected page (%lx) -"
229 "exploit attempt? (uid: %d)\n",
230 address, from_kuid(&init_user_ns,
231 current_uid()));
232 }
233 return is_exec || (address >= TASK_SIZE);
234}
235
236static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
237 struct vm_area_struct *vma,
238 bool store_update_sp)
239{
240 /*
241 * N.B. The POWER/Open ABI allows programs to access up to
242 * 288 bytes below the stack pointer.
243 * The kernel signal delivery code writes up to about 1.5kB
244 * below the stack pointer (r1) before decrementing it.
245 * The exec code can write slightly over 640kB to the stack
246 * before setting the user r1. Thus we allow the stack to
247 * expand to 1MB without further checks.
248 */
249 if (address + 0x100000 < vma->vm_end) {
250 /* get user regs even if this fault is in kernel mode */
251 struct pt_regs *uregs = current->thread.regs;
252 if (uregs == NULL)
253 return true;
254
255 /*
256 * A user-mode access to an address a long way below
257 * the stack pointer is only valid if the instruction
258 * is one which would update the stack pointer to the
259 * address accessed if the instruction completed,
260 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
261 * (or the byte, halfword, float or double forms).
262 *
263 * If we don't check this then any write to the area
264 * between the last mapped region and the stack will
265 * expand the stack rather than segfaulting.
266 */
267 if (address + 2048 < uregs->gpr[1] && !store_update_sp)
268 return true;
269 }
270 return false;
271}
272
273static bool access_error(bool is_write, bool is_exec,
274 struct vm_area_struct *vma)
275{
276 /*
277 * Allow execution from readable areas if the MMU does not
278 * provide separate controls over reading and executing.
279 *
280 * Note: That code used to not be enabled for 4xx/BookE.
281 * It is now as I/D cache coherency for these is done at
282 * set_pte_at() time and I see no reason why the test
283 * below wouldn't be valid on those processors. This -may-
284 * break programs compiled with a really old ABI though.
285 */
286 if (is_exec) {
287 return !(vma->vm_flags & VM_EXEC) &&
288 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
289 !(vma->vm_flags & (VM_READ | VM_WRITE)));
290 }
291
292 if (is_write) {
293 if (unlikely(!(vma->vm_flags & VM_WRITE)))
294 return true;
295 return false;
296 }
297
298 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
299 return true;
300 /*
301 * We should ideally do the vma pkey access check here. But in the
302 * fault path, handle_mm_fault() also does the same check. To avoid
303 * these multiple checks, we skip it here and handle access error due
304 * to pkeys later.
305 */
306 return false;
307}
308
309#ifdef CONFIG_PPC_SMLPAR
310static inline void cmo_account_page_fault(void)
311{
312 if (firmware_has_feature(FW_FEATURE_CMO)) {
313 u32 page_ins;
314
315 preempt_disable();
316 page_ins = be32_to_cpu(get_lppaca()->page_ins);
317 page_ins += 1 << PAGE_FACTOR;
318 get_lppaca()->page_ins = cpu_to_be32(page_ins);
319 preempt_enable();
320 }
321}
322#else
323static inline void cmo_account_page_fault(void) { }
324#endif /* CONFIG_PPC_SMLPAR */
325
326#ifdef CONFIG_PPC_STD_MMU
327static void sanity_check_fault(bool is_write, unsigned long error_code)
328{
329 /*
330 * For hash translation mode, we should never get a
331 * PROTFAULT. Any update to pte to reduce access will result in us
332 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
333 * fault instead of DSISR_PROTFAULT.
334 *
335 * A pte update to relax the access will not result in a hash page table
336 * entry invalidate and hence can result in DSISR_PROTFAULT.
337 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
338 * the special !is_write in the below conditional.
339 *
340 * For platforms that doesn't supports coherent icache and do support
341 * per page noexec bit, we do setup things such that we do the
342 * sync between D/I cache via fault. But that is handled via low level
343 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
344 * here in such case.
345 *
346 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
347 * check should handle those and hence we should fall to the bad_area
348 * handling correctly.
349 *
350 * For embedded with per page exec support that doesn't support coherent
351 * icache we do get PROTFAULT and we handle that D/I cache sync in
352 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
353 * is conditional for server MMU.
354 *
355 * For radix, we can get prot fault for autonuma case, because radix
356 * page table will have them marked noaccess for user.
357 */
358 if (!radix_enabled() && !is_write)
359 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
360}
361#else
362static void sanity_check_fault(bool is_write, unsigned long error_code) { }
363#endif /* CONFIG_PPC_STD_MMU */
364
365/*
366 * Define the correct "is_write" bit in error_code based
367 * on the processor family
368 */
369#if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
370#define page_fault_is_write(__err) ((__err) & ESR_DST)
371#define page_fault_is_bad(__err) (0)
372#else
373#define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
374#if defined(CONFIG_PPC_8xx)
375#define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
376#elif defined(CONFIG_PPC64)
377#define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
378#else
379#define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
380#endif
381#endif
382
383/*
384 * For 600- and 800-family processors, the error_code parameter is DSISR
385 * for a data fault, SRR1 for an instruction fault. For 400-family processors
386 * the error_code parameter is ESR for a data fault, 0 for an instruction
387 * fault.
388 * For 64-bit processors, the error_code parameter is
389 * - DSISR for a non-SLB data access fault,
390 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
391 * - 0 any SLB fault.
392 *
393 * The return value is 0 if the fault was handled, or the signal
394 * number if this is a kernel fault that can't be handled here.
395 */
396static int __do_page_fault(struct pt_regs *regs, unsigned long address,
397 unsigned long error_code)
398{
399 struct vm_area_struct * vma;
400 struct mm_struct *mm = current->mm;
401 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
402 int is_exec = TRAP(regs) == 0x400;
403 int is_user = user_mode(regs);
404 int is_write = page_fault_is_write(error_code);
405 int fault, major = 0;
406 bool store_update_sp = false;
407
408 if (notify_page_fault(regs))
409 return 0;
410
411 if (unlikely(page_fault_is_bad(error_code))) {
412 if (is_user) {
413 _exception(SIGBUS, regs, BUS_OBJERR, address);
414 return 0;
415 }
416 return SIGBUS;
417 }
418
419 /* Additional sanity check(s) */
420 sanity_check_fault(is_write, error_code);
421
422 /*
423 * The kernel should never take an execute fault nor should it
424 * take a page fault to a kernel address.
425 */
426 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
427 return SIGSEGV;
428
429 /*
430 * If we're in an interrupt, have no user context or are running
431 * in a region with pagefaults disabled then we must not take the fault
432 */
433 if (unlikely(faulthandler_disabled() || !mm)) {
434 if (is_user)
435 printk_ratelimited(KERN_ERR "Page fault in user mode"
436 " with faulthandler_disabled()=%d"
437 " mm=%p\n",
438 faulthandler_disabled(), mm);
439 return bad_area_nosemaphore(regs, address);
440 }
441
442 /* We restore the interrupt state now */
443 if (!arch_irq_disabled_regs(regs))
444 local_irq_enable();
445
446 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
447
448 if (error_code & DSISR_KEYFAULT)
449 return bad_key_fault_exception(regs, address,
450 get_mm_addr_key(mm, address));
451
452 /*
453 * We want to do this outside mmap_sem, because reading code around nip
454 * can result in fault, which will cause a deadlock when called with
455 * mmap_sem held
456 */
457 if (is_write && is_user)
458 store_update_sp = store_updates_sp(regs);
459
460 if (is_user)
461 flags |= FAULT_FLAG_USER;
462 if (is_write)
463 flags |= FAULT_FLAG_WRITE;
464 if (is_exec)
465 flags |= FAULT_FLAG_INSTRUCTION;
466
467 /* When running in the kernel we expect faults to occur only to
468 * addresses in user space. All other faults represent errors in the
469 * kernel and should generate an OOPS. Unfortunately, in the case of an
470 * erroneous fault occurring in a code path which already holds mmap_sem
471 * we will deadlock attempting to validate the fault against the
472 * address space. Luckily the kernel only validly references user
473 * space from well defined areas of code, which are listed in the
474 * exceptions table.
475 *
476 * As the vast majority of faults will be valid we will only perform
477 * the source reference check when there is a possibility of a deadlock.
478 * Attempt to lock the address space, if we cannot we then validate the
479 * source. If this is invalid we can skip the address space check,
480 * thus avoiding the deadlock.
481 */
482 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
483 if (!is_user && !search_exception_tables(regs->nip))
484 return bad_area_nosemaphore(regs, address);
485
486retry:
487 down_read(&mm->mmap_sem);
488 } else {
489 /*
490 * The above down_read_trylock() might have succeeded in
491 * which case we'll have missed the might_sleep() from
492 * down_read():
493 */
494 might_sleep();
495 }
496
497 vma = find_vma(mm, address);
498 if (unlikely(!vma))
499 return bad_area(regs, address);
500 if (likely(vma->vm_start <= address))
501 goto good_area;
502 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
503 return bad_area(regs, address);
504
505 /* The stack is being expanded, check if it's valid */
506 if (unlikely(bad_stack_expansion(regs, address, vma, store_update_sp)))
507 return bad_area(regs, address);
508
509 /* Try to expand it */
510 if (unlikely(expand_stack(vma, address)))
511 return bad_area(regs, address);
512
513good_area:
514 if (unlikely(access_error(is_write, is_exec, vma)))
515 return bad_access(regs, address);
516
517 /*
518 * If for any reason at all we couldn't handle the fault,
519 * make sure we exit gracefully rather than endlessly redo
520 * the fault.
521 */
522 fault = handle_mm_fault(vma, address, flags);
523
524#ifdef CONFIG_PPC_MEM_KEYS
525 /*
526 * we skipped checking for access error due to key earlier.
527 * Check that using handle_mm_fault error return.
528 */
529 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
530 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
531
532 int pkey = vma_pkey(vma);
533
534 up_read(&mm->mmap_sem);
535 return bad_key_fault_exception(regs, address, pkey);
536 }
537#endif /* CONFIG_PPC_MEM_KEYS */
538
539 major |= fault & VM_FAULT_MAJOR;
540
541 /*
542 * Handle the retry right now, the mmap_sem has been released in that
543 * case.
544 */
545 if (unlikely(fault & VM_FAULT_RETRY)) {
546 /* We retry only once */
547 if (flags & FAULT_FLAG_ALLOW_RETRY) {
548 /*
549 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
550 * of starvation.
551 */
552 flags &= ~FAULT_FLAG_ALLOW_RETRY;
553 flags |= FAULT_FLAG_TRIED;
554 if (!fatal_signal_pending(current))
555 goto retry;
556 }
557
558 /*
559 * User mode? Just return to handle the fatal exception otherwise
560 * return to bad_page_fault
561 */
562 return is_user ? 0 : SIGBUS;
563 }
564
565 up_read(¤t->mm->mmap_sem);
566
567 if (unlikely(fault & VM_FAULT_ERROR))
568 return mm_fault_error(regs, address, fault);
569
570 /*
571 * Major/minor page fault accounting.
572 */
573 if (major) {
574 current->maj_flt++;
575 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
576 cmo_account_page_fault();
577 } else {
578 current->min_flt++;
579 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
580 }
581 return 0;
582}
583NOKPROBE_SYMBOL(__do_page_fault);
584
585int do_page_fault(struct pt_regs *regs, unsigned long address,
586 unsigned long error_code)
587{
588 enum ctx_state prev_state = exception_enter();
589 int rc = __do_page_fault(regs, address, error_code);
590 exception_exit(prev_state);
591 return rc;
592}
593NOKPROBE_SYMBOL(do_page_fault);
594
595/*
596 * bad_page_fault is called when we have a bad access from the kernel.
597 * It is called from the DSI and ISI handlers in head.S and from some
598 * of the procedures in traps.c.
599 */
600void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
601{
602 const struct exception_table_entry *entry;
603
604 /* Are we prepared to handle this fault? */
605 if ((entry = search_exception_tables(regs->nip)) != NULL) {
606 regs->nip = extable_fixup(entry);
607 return;
608 }
609
610 /* kernel has accessed a bad area */
611
612 switch (TRAP(regs)) {
613 case 0x300:
614 case 0x380:
615 printk(KERN_ALERT "Unable to handle kernel paging request for "
616 "data at address 0x%08lx\n", regs->dar);
617 break;
618 case 0x400:
619 case 0x480:
620 printk(KERN_ALERT "Unable to handle kernel paging request for "
621 "instruction fetch\n");
622 break;
623 case 0x600:
624 printk(KERN_ALERT "Unable to handle kernel paging request for "
625 "unaligned access at address 0x%08lx\n", regs->dar);
626 break;
627 default:
628 printk(KERN_ALERT "Unable to handle kernel paging request for "
629 "unknown fault\n");
630 break;
631 }
632 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
633 regs->nip);
634
635 if (task_stack_end_corrupted(current))
636 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
637
638 die("Kernel access of bad area", regs, sig);
639}
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#include <linux/context_tracking.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#include <mm/mmu_decl.h>
47
48#include "icswx.h"
49
50#ifdef CONFIG_KPROBES
51static inline int notify_page_fault(struct pt_regs *regs)
52{
53 int ret = 0;
54
55 /* kprobe_running() needs smp_processor_id() */
56 if (!user_mode(regs)) {
57 preempt_disable();
58 if (kprobe_running() && kprobe_fault_handler(regs, 11))
59 ret = 1;
60 preempt_enable();
61 }
62
63 return ret;
64}
65#else
66static inline int notify_page_fault(struct pt_regs *regs)
67{
68 return 0;
69}
70#endif
71
72/*
73 * Check whether the instruction at regs->nip is a store using
74 * an update addressing form which will update r1.
75 */
76static int store_updates_sp(struct pt_regs *regs)
77{
78 unsigned int inst;
79
80 if (get_user(inst, (unsigned int __user *)regs->nip))
81 return 0;
82 /* check for 1 in the rA field */
83 if (((inst >> 16) & 0x1f) != 1)
84 return 0;
85 /* check major opcode */
86 switch (inst >> 26) {
87 case 37: /* stwu */
88 case 39: /* stbu */
89 case 45: /* sthu */
90 case 53: /* stfsu */
91 case 55: /* stfdu */
92 return 1;
93 case 62: /* std or stdu */
94 return (inst & 3) == 1;
95 case 31:
96 /* check minor opcode */
97 switch ((inst >> 1) & 0x3ff) {
98 case 181: /* stdux */
99 case 183: /* stwux */
100 case 247: /* stbux */
101 case 439: /* sthux */
102 case 695: /* stfsux */
103 case 759: /* stfdux */
104 return 1;
105 }
106 }
107 return 0;
108}
109/*
110 * do_page_fault error handling helpers
111 */
112
113#define MM_FAULT_RETURN 0
114#define MM_FAULT_CONTINUE -1
115#define MM_FAULT_ERR(sig) (sig)
116
117static int do_sigbus(struct pt_regs *regs, unsigned long address)
118{
119 siginfo_t info;
120
121 up_read(¤t->mm->mmap_sem);
122
123 if (user_mode(regs)) {
124 current->thread.trap_nr = BUS_ADRERR;
125 info.si_signo = SIGBUS;
126 info.si_errno = 0;
127 info.si_code = BUS_ADRERR;
128 info.si_addr = (void __user *)address;
129 force_sig_info(SIGBUS, &info, current);
130 return MM_FAULT_RETURN;
131 }
132 return MM_FAULT_ERR(SIGBUS);
133}
134
135static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
136{
137 /*
138 * Pagefault was interrupted by SIGKILL. We have no reason to
139 * continue the pagefault.
140 */
141 if (fatal_signal_pending(current)) {
142 /*
143 * If we have retry set, the mmap semaphore will have
144 * alrady been released in __lock_page_or_retry(). Else
145 * we release it now.
146 */
147 if (!(fault & VM_FAULT_RETRY))
148 up_read(¤t->mm->mmap_sem);
149 /* Coming from kernel, we need to deal with uaccess fixups */
150 if (user_mode(regs))
151 return MM_FAULT_RETURN;
152 return MM_FAULT_ERR(SIGKILL);
153 }
154
155 /* No fault: be happy */
156 if (!(fault & VM_FAULT_ERROR))
157 return MM_FAULT_CONTINUE;
158
159 /* Out of memory */
160 if (fault & VM_FAULT_OOM) {
161 up_read(¤t->mm->mmap_sem);
162
163 /*
164 * We ran out of memory, or some other thing happened to us that
165 * made us unable to handle the page fault gracefully.
166 */
167 if (!user_mode(regs))
168 return MM_FAULT_ERR(SIGKILL);
169 pagefault_out_of_memory();
170 return MM_FAULT_RETURN;
171 }
172
173 /* Bus error. x86 handles HWPOISON here, we'll add this if/when
174 * we support the feature in HW
175 */
176 if (fault & VM_FAULT_SIGBUS)
177 return do_sigbus(regs, addr);
178
179 /* We don't understand the fault code, this is fatal */
180 BUG();
181 return MM_FAULT_CONTINUE;
182}
183
184/*
185 * For 600- and 800-family processors, the error_code parameter is DSISR
186 * for a data fault, SRR1 for an instruction fault. For 400-family processors
187 * the error_code parameter is ESR for a data fault, 0 for an instruction
188 * fault.
189 * For 64-bit processors, the error_code parameter is
190 * - DSISR for a non-SLB data access fault,
191 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
192 * - 0 any SLB fault.
193 *
194 * The return value is 0 if the fault was handled, or the signal
195 * number if this is a kernel fault that can't be handled here.
196 */
197int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
198 unsigned long error_code)
199{
200 enum ctx_state prev_state = exception_enter();
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 int rc = 0, store_update_sp = 0;
210
211#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
212 /*
213 * Fortunately the bit assignments in SRR1 for an instruction
214 * fault and DSISR for a data fault are mostly the same for the
215 * bits we are interested in. But there are some bits which
216 * indicate errors in DSISR but can validly be set in SRR1.
217 */
218 if (trap == 0x400)
219 error_code &= 0x48200000;
220 else
221 is_write = error_code & DSISR_ISSTORE;
222#else
223 is_write = error_code & ESR_DST;
224#endif /* CONFIG_4xx || CONFIG_BOOKE */
225
226#ifdef CONFIG_PPC_ICSWX
227 /*
228 * we need to do this early because this "data storage
229 * interrupt" does not update the DAR/DEAR so we don't want to
230 * look at it
231 */
232 if (error_code & ICSWX_DSI_UCT) {
233 rc = acop_handle_fault(regs, address, error_code);
234 if (rc)
235 goto bail;
236 }
237#endif /* CONFIG_PPC_ICSWX */
238
239 if (notify_page_fault(regs))
240 goto bail;
241
242 if (unlikely(debugger_fault_handler(regs)))
243 goto bail;
244
245 /* On a kernel SLB miss we can only check for a valid exception entry */
246 if (!user_mode(regs) && (address >= TASK_SIZE)) {
247 rc = SIGSEGV;
248 goto bail;
249 }
250
251#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
252 defined(CONFIG_PPC_BOOK3S_64))
253 if (error_code & DSISR_DABRMATCH) {
254 /* breakpoint match */
255 do_break(regs, address, error_code);
256 goto bail;
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 rc = SIGSEGV;
267 goto bail;
268 }
269 /* in_atomic() in user mode is really bad,
270 as is current->mm == NULL. */
271 printk(KERN_EMERG "Page fault in user mode with "
272 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
273 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
274 regs->nip, regs->msr);
275 die("Weird page fault", regs, SIGSEGV);
276 }
277
278 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
279
280 /*
281 * We want to do this outside mmap_sem, because reading code around nip
282 * can result in fault, which will cause a deadlock when called with
283 * mmap_sem held
284 */
285 if (user_mode(regs))
286 store_update_sp = store_updates_sp(regs);
287
288 if (user_mode(regs))
289 flags |= FAULT_FLAG_USER;
290
291 /* When running in the kernel we expect faults to occur only to
292 * addresses in user space. All other faults represent errors in the
293 * kernel and should generate an OOPS. Unfortunately, in the case of an
294 * erroneous fault occurring in a code path which already holds mmap_sem
295 * we will deadlock attempting to validate the fault against the
296 * address space. Luckily the kernel only validly references user
297 * space from well defined areas of code, which are listed in the
298 * exceptions table.
299 *
300 * As the vast majority of faults will be valid we will only perform
301 * the source reference check when there is a possibility of a deadlock.
302 * Attempt to lock the address space, if we cannot we then validate the
303 * source. If this is invalid we can skip the address space check,
304 * thus avoiding the deadlock.
305 */
306 if (!down_read_trylock(&mm->mmap_sem)) {
307 if (!user_mode(regs) && !search_exception_tables(regs->nip))
308 goto bad_area_nosemaphore;
309
310retry:
311 down_read(&mm->mmap_sem);
312 } else {
313 /*
314 * The above down_read_trylock() might have succeeded in
315 * which case we'll have missed the might_sleep() from
316 * down_read():
317 */
318 might_sleep();
319 }
320
321 vma = find_vma(mm, address);
322 if (!vma)
323 goto bad_area;
324 if (vma->vm_start <= address)
325 goto good_area;
326 if (!(vma->vm_flags & VM_GROWSDOWN))
327 goto bad_area;
328
329 /*
330 * N.B. The POWER/Open ABI allows programs to access up to
331 * 288 bytes below the stack pointer.
332 * The kernel signal delivery code writes up to about 1.5kB
333 * below the stack pointer (r1) before decrementing it.
334 * The exec code can write slightly over 640kB to the stack
335 * before setting the user r1. Thus we allow the stack to
336 * expand to 1MB without further checks.
337 */
338 if (address + 0x100000 < vma->vm_end) {
339 /* get user regs even if this fault is in kernel mode */
340 struct pt_regs *uregs = current->thread.regs;
341 if (uregs == NULL)
342 goto bad_area;
343
344 /*
345 * A user-mode access to an address a long way below
346 * the stack pointer is only valid if the instruction
347 * is one which would update the stack pointer to the
348 * address accessed if the instruction completed,
349 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
350 * (or the byte, halfword, float or double forms).
351 *
352 * If we don't check this then any write to the area
353 * between the last mapped region and the stack will
354 * expand the stack rather than segfaulting.
355 */
356 if (address + 2048 < uregs->gpr[1] && !store_update_sp)
357 goto bad_area;
358 }
359 if (expand_stack(vma, address))
360 goto bad_area;
361
362good_area:
363 code = SEGV_ACCERR;
364#if defined(CONFIG_6xx)
365 if (error_code & 0x95700000)
366 /* an error such as lwarx to I/O controller space,
367 address matching DABR, eciwx, etc. */
368 goto bad_area;
369#endif /* CONFIG_6xx */
370#if defined(CONFIG_8xx)
371 /* 8xx sometimes need to load a invalid/non-present TLBs.
372 * These must be invalidated separately as linux mm don't.
373 */
374 if (error_code & 0x40000000) /* no translation? */
375 _tlbil_va(address, 0, 0, 0);
376
377 /* The MPC8xx seems to always set 0x80000000, which is
378 * "undefined". Of those that can be set, this is the only
379 * one which seems bad.
380 */
381 if (error_code & 0x10000000)
382 /* Guarded storage error. */
383 goto bad_area;
384#endif /* CONFIG_8xx */
385
386 if (is_exec) {
387#ifdef CONFIG_PPC_STD_MMU
388 /* Protection fault on exec go straight to failure on
389 * Hash based MMUs as they either don't support per-page
390 * execute permission, or if they do, it's handled already
391 * at the hash level. This test would probably have to
392 * be removed if we change the way this works to make hash
393 * processors use the same I/D cache coherency mechanism
394 * as embedded.
395 */
396 if (error_code & DSISR_PROTFAULT)
397 goto bad_area;
398#endif /* CONFIG_PPC_STD_MMU */
399
400 /*
401 * Allow execution from readable areas if the MMU does not
402 * provide separate controls over reading and executing.
403 *
404 * Note: That code used to not be enabled for 4xx/BookE.
405 * It is now as I/D cache coherency for these is done at
406 * set_pte_at() time and I see no reason why the test
407 * below wouldn't be valid on those processors. This -may-
408 * break programs compiled with a really old ABI though.
409 */
410 if (!(vma->vm_flags & VM_EXEC) &&
411 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
412 !(vma->vm_flags & (VM_READ | VM_WRITE))))
413 goto bad_area;
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 /* protection fault */
422 if (error_code & 0x08000000)
423 goto bad_area;
424 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
425 goto bad_area;
426 }
427
428 /*
429 * If for any reason at all we couldn't handle the fault,
430 * make sure we exit gracefully rather than endlessly redo
431 * the fault.
432 */
433 fault = handle_mm_fault(mm, vma, address, flags);
434 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
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 unsigned long *stackend;
512
513 /* Are we prepared to handle this fault? */
514 if ((entry = search_exception_tables(regs->nip)) != NULL) {
515 regs->nip = entry->fixup;
516 return;
517 }
518
519 /* kernel has accessed a bad area */
520
521 switch (regs->trap) {
522 case 0x300:
523 case 0x380:
524 printk(KERN_ALERT "Unable to handle kernel paging request for "
525 "data at address 0x%08lx\n", regs->dar);
526 break;
527 case 0x400:
528 case 0x480:
529 printk(KERN_ALERT "Unable to handle kernel paging request for "
530 "instruction fetch\n");
531 break;
532 default:
533 printk(KERN_ALERT "Unable to handle kernel paging request for "
534 "unknown fault\n");
535 break;
536 }
537 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
538 regs->nip);
539
540 stackend = end_of_stack(current);
541 if (current != &init_task && *stackend != STACK_END_MAGIC)
542 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
543
544 die("Kernel access of bad area", regs, sig);
545}