3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5 * Derived from "arch/i386/mm/fault.c"
6 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Modified by Cort Dougan and Paul Mackerras.
10 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
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.
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/pagemap.h>
26 #include <linux/ptrace.h>
27 #include <linux/mman.h>
29 #include <linux/interrupt.h>
30 #include <linux/highmem.h>
31 #include <linux/extable.h>
32 #include <linux/kprobes.h>
33 #include <linux/kdebug.h>
34 #include <linux/perf_event.h>
35 #include <linux/ratelimit.h>
36 #include <linux/context_tracking.h>
37 #include <linux/hugetlb.h>
38 #include <linux/uaccess.h>
40 #include <asm/firmware.h>
42 #include <asm/pgtable.h>
44 #include <asm/mmu_context.h>
45 #include <asm/siginfo.h>
46 #include <asm/debug.h>
49 static inline bool notify_page_fault(struct pt_regs
*regs
)
54 /* kprobe_running() needs smp_processor_id() */
55 if (!user_mode(regs
)) {
57 if (kprobe_running() && kprobe_fault_handler(regs
, 11))
61 #endif /* CONFIG_KPROBES */
63 if (unlikely(debugger_fault_handler(regs
)))
70 * Check whether the instruction inst is a store using
71 * an update addressing form which will update r1.
73 static bool store_updates_sp(unsigned int inst
)
75 /* check for 1 in the rA field */
76 if (((inst
>> 16) & 0x1f) != 1)
78 /* check major opcode */
86 case OP_STD
: /* std or stdu */
87 return (inst
& 3) == 1;
89 /* check minor opcode */
90 switch ((inst
>> 1) & 0x3ff) {
95 case OP_31_XOP_STFSUX
:
96 case OP_31_XOP_STFDUX
:
103 * do_page_fault error handling helpers
107 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long address
, int si_code
)
110 * If we are in kernel mode, bail out with a SEGV, this will
111 * be caught by the assembly which will restore the non-volatile
112 * registers before calling bad_page_fault()
114 if (!user_mode(regs
))
117 _exception(SIGSEGV
, regs
, si_code
, address
);
122 static noinline
int bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long address
)
124 return __bad_area_nosemaphore(regs
, address
, SEGV_MAPERR
);
127 static int __bad_area(struct pt_regs
*regs
, unsigned long address
, int si_code
)
129 struct mm_struct
*mm
= current
->mm
;
132 * Something tried to access memory that isn't in our memory map..
133 * Fix it, but check if it's kernel or user first..
135 up_read(&mm
->mmap_sem
);
137 return __bad_area_nosemaphore(regs
, address
, si_code
);
140 static noinline
int bad_area(struct pt_regs
*regs
, unsigned long address
)
142 return __bad_area(regs
, address
, SEGV_MAPERR
);
145 static int bad_key_fault_exception(struct pt_regs
*regs
, unsigned long address
,
149 * If we are in kernel mode, bail out with a SEGV, this will
150 * be caught by the assembly which will restore the non-volatile
151 * registers before calling bad_page_fault()
153 if (!user_mode(regs
))
156 _exception_pkey(regs
, address
, pkey
);
161 static noinline
int bad_access(struct pt_regs
*regs
, unsigned long address
)
163 return __bad_area(regs
, address
, SEGV_ACCERR
);
166 static int do_sigbus(struct pt_regs
*regs
, unsigned long address
,
169 if (!user_mode(regs
))
172 current
->thread
.trap_nr
= BUS_ADRERR
;
173 #ifdef CONFIG_MEMORY_FAILURE
174 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
175 unsigned int lsb
= 0; /* shutup gcc */
177 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
178 current
->comm
, current
->pid
, address
);
180 if (fault
& VM_FAULT_HWPOISON_LARGE
)
181 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
182 if (fault
& VM_FAULT_HWPOISON
)
185 force_sig_mceerr(BUS_MCEERR_AR
, (void __user
*)address
, lsb
);
190 force_sig_fault(SIGBUS
, BUS_ADRERR
, (void __user
*)address
);
194 static int mm_fault_error(struct pt_regs
*regs
, unsigned long addr
,
198 * Kernel page fault interrupted by SIGKILL. We have no reason to
199 * continue processing.
201 if (fatal_signal_pending(current
) && !user_mode(regs
))
205 if (fault
& VM_FAULT_OOM
) {
207 * We ran out of memory, or some other thing happened to us that
208 * made us unable to handle the page fault gracefully.
210 if (!user_mode(regs
))
212 pagefault_out_of_memory();
214 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
215 VM_FAULT_HWPOISON_LARGE
))
216 return do_sigbus(regs
, addr
, fault
);
217 else if (fault
& VM_FAULT_SIGSEGV
)
218 return bad_area_nosemaphore(regs
, addr
);
225 /* Is this a bad kernel fault ? */
226 static bool bad_kernel_fault(struct pt_regs
*regs
, unsigned long error_code
,
227 unsigned long address
, bool is_write
)
229 int is_exec
= TRAP(regs
) == 0x400;
231 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
232 if (is_exec
&& (error_code
& (DSISR_NOEXEC_OR_G
| DSISR_KEYFAULT
|
234 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
235 address
>= TASK_SIZE
? "exec-protected" : "user",
237 from_kuid(&init_user_ns
, current_uid()));
239 // Kernel exec fault is always bad
243 if (!is_exec
&& address
< TASK_SIZE
&& (error_code
& DSISR_PROTFAULT
) &&
244 !search_exception_tables(regs
->nip
)) {
245 pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
247 from_kuid(&init_user_ns
, current_uid()));
250 // Kernel fault on kernel address is bad
251 if (address
>= TASK_SIZE
)
254 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
255 if (!search_exception_tables(regs
->nip
))
258 // Read/write fault in a valid region (the exception table search passed
259 // above), but blocked by KUAP is bad, it can never succeed.
260 if (bad_kuap_fault(regs
, is_write
))
263 // What's left? Kernel fault on user in well defined regions (extable
264 // matched), and allowed by KUAP in the faulting context.
268 static bool bad_stack_expansion(struct pt_regs
*regs
, unsigned long address
,
269 struct vm_area_struct
*vma
, unsigned int flags
,
273 * N.B. The POWER/Open ABI allows programs to access up to
274 * 288 bytes below the stack pointer.
275 * The kernel signal delivery code writes up to about 1.5kB
276 * below the stack pointer (r1) before decrementing it.
277 * The exec code can write slightly over 640kB to the stack
278 * before setting the user r1. Thus we allow the stack to
279 * expand to 1MB without further checks.
281 if (address
+ 0x100000 < vma
->vm_end
) {
282 unsigned int __user
*nip
= (unsigned int __user
*)regs
->nip
;
283 /* get user regs even if this fault is in kernel mode */
284 struct pt_regs
*uregs
= current
->thread
.regs
;
289 * A user-mode access to an address a long way below
290 * the stack pointer is only valid if the instruction
291 * is one which would update the stack pointer to the
292 * address accessed if the instruction completed,
293 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
294 * (or the byte, halfword, float or double forms).
296 * If we don't check this then any write to the area
297 * between the last mapped region and the stack will
298 * expand the stack rather than segfaulting.
300 if (address
+ 2048 >= uregs
->gpr
[1])
303 if ((flags
& FAULT_FLAG_WRITE
) && (flags
& FAULT_FLAG_USER
) &&
304 access_ok(nip
, sizeof(*nip
))) {
309 res
= __get_user_inatomic(inst
, nip
);
312 return !store_updates_sp(inst
);
320 static bool access_error(bool is_write
, bool is_exec
,
321 struct vm_area_struct
*vma
)
324 * Allow execution from readable areas if the MMU does not
325 * provide separate controls over reading and executing.
327 * Note: That code used to not be enabled for 4xx/BookE.
328 * It is now as I/D cache coherency for these is done at
329 * set_pte_at() time and I see no reason why the test
330 * below wouldn't be valid on those processors. This -may-
331 * break programs compiled with a really old ABI though.
334 return !(vma
->vm_flags
& VM_EXEC
) &&
335 (cpu_has_feature(CPU_FTR_NOEXECUTE
) ||
336 !(vma
->vm_flags
& (VM_READ
| VM_WRITE
)));
340 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
345 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
348 * We should ideally do the vma pkey access check here. But in the
349 * fault path, handle_mm_fault() also does the same check. To avoid
350 * these multiple checks, we skip it here and handle access error due
356 #ifdef CONFIG_PPC_SMLPAR
357 static inline void cmo_account_page_fault(void)
359 if (firmware_has_feature(FW_FEATURE_CMO
)) {
363 page_ins
= be32_to_cpu(get_lppaca()->page_ins
);
364 page_ins
+= 1 << PAGE_FACTOR
;
365 get_lppaca()->page_ins
= cpu_to_be32(page_ins
);
370 static inline void cmo_account_page_fault(void) { }
371 #endif /* CONFIG_PPC_SMLPAR */
373 #ifdef CONFIG_PPC_BOOK3S
374 static void sanity_check_fault(bool is_write
, bool is_user
,
375 unsigned long error_code
, unsigned long address
)
378 * Userspace trying to access kernel address, we get PROTFAULT for that.
380 if (is_user
&& address
>= TASK_SIZE
) {
381 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
382 current
->comm
, current
->pid
, address
,
383 from_kuid(&init_user_ns
, current_uid()));
388 * For hash translation mode, we should never get a
389 * PROTFAULT. Any update to pte to reduce access will result in us
390 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
391 * fault instead of DSISR_PROTFAULT.
393 * A pte update to relax the access will not result in a hash page table
394 * entry invalidate and hence can result in DSISR_PROTFAULT.
395 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
396 * the special !is_write in the below conditional.
398 * For platforms that doesn't supports coherent icache and do support
399 * per page noexec bit, we do setup things such that we do the
400 * sync between D/I cache via fault. But that is handled via low level
401 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
404 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
405 * check should handle those and hence we should fall to the bad_area
406 * handling correctly.
408 * For embedded with per page exec support that doesn't support coherent
409 * icache we do get PROTFAULT and we handle that D/I cache sync in
410 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
411 * is conditional for server MMU.
413 * For radix, we can get prot fault for autonuma case, because radix
414 * page table will have them marked noaccess for user.
416 if (radix_enabled() || is_write
)
419 WARN_ON_ONCE(error_code
& DSISR_PROTFAULT
);
422 static void sanity_check_fault(bool is_write
, bool is_user
,
423 unsigned long error_code
, unsigned long address
) { }
424 #endif /* CONFIG_PPC_BOOK3S */
427 * Define the correct "is_write" bit in error_code based
428 * on the processor family
430 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
431 #define page_fault_is_write(__err) ((__err) & ESR_DST)
432 #define page_fault_is_bad(__err) (0)
434 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
435 #if defined(CONFIG_PPC_8xx)
436 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
437 #elif defined(CONFIG_PPC64)
438 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
440 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
445 * For 600- and 800-family processors, the error_code parameter is DSISR
446 * for a data fault, SRR1 for an instruction fault. For 400-family processors
447 * the error_code parameter is ESR for a data fault, 0 for an instruction
449 * For 64-bit processors, the error_code parameter is
450 * - DSISR for a non-SLB data access fault,
451 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
454 * The return value is 0 if the fault was handled, or the signal
455 * number if this is a kernel fault that can't be handled here.
457 static int __do_page_fault(struct pt_regs
*regs
, unsigned long address
,
458 unsigned long error_code
)
460 struct vm_area_struct
* vma
;
461 struct mm_struct
*mm
= current
->mm
;
462 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
463 int is_exec
= TRAP(regs
) == 0x400;
464 int is_user
= user_mode(regs
);
465 int is_write
= page_fault_is_write(error_code
);
466 vm_fault_t fault
, major
= 0;
467 bool must_retry
= false;
469 if (notify_page_fault(regs
))
472 if (unlikely(page_fault_is_bad(error_code
))) {
474 _exception(SIGBUS
, regs
, BUS_OBJERR
, address
);
480 /* Additional sanity check(s) */
481 sanity_check_fault(is_write
, is_user
, error_code
, address
);
484 * The kernel should never take an execute fault nor should it
485 * take a page fault to a kernel address or a page fault to a user
486 * address outside of dedicated places
488 if (unlikely(!is_user
&& bad_kernel_fault(regs
, error_code
, address
, is_write
)))
492 * If we're in an interrupt, have no user context or are running
493 * in a region with pagefaults disabled then we must not take the fault
495 if (unlikely(faulthandler_disabled() || !mm
)) {
497 printk_ratelimited(KERN_ERR
"Page fault in user mode"
498 " with faulthandler_disabled()=%d"
500 faulthandler_disabled(), mm
);
501 return bad_area_nosemaphore(regs
, address
);
504 /* We restore the interrupt state now */
505 if (!arch_irq_disabled_regs(regs
))
508 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
510 if (error_code
& DSISR_KEYFAULT
)
511 return bad_key_fault_exception(regs
, address
,
512 get_mm_addr_key(mm
, address
));
515 * We want to do this outside mmap_sem, because reading code around nip
516 * can result in fault, which will cause a deadlock when called with
520 flags
|= FAULT_FLAG_USER
;
522 flags
|= FAULT_FLAG_WRITE
;
524 flags
|= FAULT_FLAG_INSTRUCTION
;
526 /* When running in the kernel we expect faults to occur only to
527 * addresses in user space. All other faults represent errors in the
528 * kernel and should generate an OOPS. Unfortunately, in the case of an
529 * erroneous fault occurring in a code path which already holds mmap_sem
530 * we will deadlock attempting to validate the fault against the
531 * address space. Luckily the kernel only validly references user
532 * space from well defined areas of code, which are listed in the
535 * As the vast majority of faults will be valid we will only perform
536 * the source reference check when there is a possibility of a deadlock.
537 * Attempt to lock the address space, if we cannot we then validate the
538 * source. If this is invalid we can skip the address space check,
539 * thus avoiding the deadlock.
541 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
542 if (!is_user
&& !search_exception_tables(regs
->nip
))
543 return bad_area_nosemaphore(regs
, address
);
546 down_read(&mm
->mmap_sem
);
549 * The above down_read_trylock() might have succeeded in
550 * which case we'll have missed the might_sleep() from
556 vma
= find_vma(mm
, address
);
558 return bad_area(regs
, address
);
559 if (likely(vma
->vm_start
<= address
))
561 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
)))
562 return bad_area(regs
, address
);
564 /* The stack is being expanded, check if it's valid */
565 if (unlikely(bad_stack_expansion(regs
, address
, vma
, flags
,
568 return bad_area(regs
, address
);
570 up_read(&mm
->mmap_sem
);
571 if (fault_in_pages_readable((const char __user
*)regs
->nip
,
572 sizeof(unsigned int)))
573 return bad_area_nosemaphore(regs
, address
);
577 /* Try to expand it */
578 if (unlikely(expand_stack(vma
, address
)))
579 return bad_area(regs
, address
);
582 if (unlikely(access_error(is_write
, is_exec
, vma
)))
583 return bad_access(regs
, address
);
586 * If for any reason at all we couldn't handle the fault,
587 * make sure we exit gracefully rather than endlessly redo
590 fault
= handle_mm_fault(vma
, address
, flags
);
592 #ifdef CONFIG_PPC_MEM_KEYS
594 * we skipped checking for access error due to key earlier.
595 * Check that using handle_mm_fault error return.
597 if (unlikely(fault
& VM_FAULT_SIGSEGV
) &&
598 !arch_vma_access_permitted(vma
, is_write
, is_exec
, 0)) {
600 int pkey
= vma_pkey(vma
);
602 up_read(&mm
->mmap_sem
);
603 return bad_key_fault_exception(regs
, address
, pkey
);
605 #endif /* CONFIG_PPC_MEM_KEYS */
607 major
|= fault
& VM_FAULT_MAJOR
;
610 * Handle the retry right now, the mmap_sem has been released in that
613 if (unlikely(fault
& VM_FAULT_RETRY
)) {
614 /* We retry only once */
615 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
617 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
620 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
621 flags
|= FAULT_FLAG_TRIED
;
622 if (!fatal_signal_pending(current
))
627 * User mode? Just return to handle the fatal exception otherwise
628 * return to bad_page_fault
630 return is_user
? 0 : SIGBUS
;
633 up_read(¤t
->mm
->mmap_sem
);
635 if (unlikely(fault
& VM_FAULT_ERROR
))
636 return mm_fault_error(regs
, address
, fault
);
639 * Major/minor page fault accounting.
643 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
, address
);
644 cmo_account_page_fault();
647 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
, address
);
651 NOKPROBE_SYMBOL(__do_page_fault
);
653 int do_page_fault(struct pt_regs
*regs
, unsigned long address
,
654 unsigned long error_code
)
656 enum ctx_state prev_state
= exception_enter();
657 int rc
= __do_page_fault(regs
, address
, error_code
);
658 exception_exit(prev_state
);
661 NOKPROBE_SYMBOL(do_page_fault
);
664 * bad_page_fault is called when we have a bad access from the kernel.
665 * It is called from the DSI and ISI handlers in head.S and from some
666 * of the procedures in traps.c.
668 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
670 const struct exception_table_entry
*entry
;
672 /* Are we prepared to handle this fault? */
673 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
674 regs
->nip
= extable_fixup(entry
);
678 /* kernel has accessed a bad area */
680 switch (TRAP(regs
)) {
684 pr_alert("BUG: %s at 0x%08lx\n",
685 regs
->dar
< PAGE_SIZE
? "Kernel NULL pointer dereference" :
686 "Unable to handle kernel data access", regs
->dar
);
690 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
691 regs
->nip
< PAGE_SIZE
? " (NULL pointer?)\n" : "\n");
694 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
698 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
702 printk(KERN_ALERT
"Faulting instruction address: 0x%08lx\n",
705 if (task_stack_end_corrupted(current
))
706 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
708 die("Kernel access of bad area", regs
, sig
);