1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6 * Derived from "arch/i386/mm/fault.c"
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
9 * Modified by Cort Dougan and Paul Mackerras.
11 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/pagemap.h>
22 #include <linux/ptrace.h>
23 #include <linux/mman.h>
25 #include <linux/interrupt.h>
26 #include <linux/highmem.h>
27 #include <linux/extable.h>
28 #include <linux/kprobes.h>
29 #include <linux/kdebug.h>
30 #include <linux/perf_event.h>
31 #include <linux/ratelimit.h>
32 #include <linux/context_tracking.h>
33 #include <linux/hugetlb.h>
34 #include <linux/uaccess.h>
36 #include <asm/firmware.h>
39 #include <asm/mmu_context.h>
40 #include <asm/siginfo.h>
41 #include <asm/debug.h>
46 * Check whether the instruction inst is a store using
47 * an update addressing form which will update r1.
49 static bool store_updates_sp(struct ppc_inst inst
)
51 /* check for 1 in the rA field */
52 if (((ppc_inst_val(inst
) >> 16) & 0x1f) != 1)
54 /* check major opcode */
55 switch (ppc_inst_primary_opcode(inst
)) {
62 case OP_STD
: /* std or stdu */
63 return (ppc_inst_val(inst
) & 3) == 1;
65 /* check minor opcode */
66 switch ((ppc_inst_val(inst
) >> 1) & 0x3ff) {
71 case OP_31_XOP_STFSUX
:
72 case OP_31_XOP_STFDUX
:
79 * do_page_fault error handling helpers
83 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long address
, int si_code
)
86 * If we are in kernel mode, bail out with a SEGV, this will
87 * be caught by the assembly which will restore the non-volatile
88 * registers before calling bad_page_fault()
93 _exception(SIGSEGV
, regs
, si_code
, address
);
98 static noinline
int bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long address
)
100 return __bad_area_nosemaphore(regs
, address
, SEGV_MAPERR
);
103 static int __bad_area(struct pt_regs
*regs
, unsigned long address
, int si_code
)
105 struct mm_struct
*mm
= current
->mm
;
108 * Something tried to access memory that isn't in our memory map..
109 * Fix it, but check if it's kernel or user first..
111 mmap_read_unlock(mm
);
113 return __bad_area_nosemaphore(regs
, address
, si_code
);
116 static noinline
int bad_area(struct pt_regs
*regs
, unsigned long address
)
118 return __bad_area(regs
, address
, SEGV_MAPERR
);
121 #ifdef CONFIG_PPC_MEM_KEYS
122 static noinline
int bad_access_pkey(struct pt_regs
*regs
, unsigned long address
,
123 struct vm_area_struct
*vma
)
125 struct mm_struct
*mm
= current
->mm
;
129 * We don't try to fetch the pkey from page table because reading
130 * page table without locking doesn't guarantee stable pte value.
131 * Hence the pkey value that we return to userspace can be different
132 * from the pkey that actually caused access error.
134 * It does *not* guarantee that the VMA we find here
135 * was the one that we faulted on.
137 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
138 * 2. T1 : set AMR to deny access to pkey=4, touches, page
140 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
141 * 5. T1 : enters fault handler, takes mmap_sem, etc...
142 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
143 * faulted on a pte with its pkey=4.
145 pkey
= vma_pkey(vma
);
147 mmap_read_unlock(mm
);
150 * If we are in kernel mode, bail out with a SEGV, this will
151 * be caught by the assembly which will restore the non-volatile
152 * registers before calling bad_page_fault()
154 if (!user_mode(regs
))
157 _exception_pkey(regs
, address
, pkey
);
163 static noinline
int bad_access(struct pt_regs
*regs
, unsigned long address
)
165 return __bad_area(regs
, address
, SEGV_ACCERR
);
168 static int do_sigbus(struct pt_regs
*regs
, unsigned long address
,
171 if (!user_mode(regs
))
174 current
->thread
.trap_nr
= BUS_ADRERR
;
175 #ifdef CONFIG_MEMORY_FAILURE
176 if (fault
& (VM_FAULT_HWPOISON
|VM_FAULT_HWPOISON_LARGE
)) {
177 unsigned int lsb
= 0; /* shutup gcc */
179 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
180 current
->comm
, current
->pid
, address
);
182 if (fault
& VM_FAULT_HWPOISON_LARGE
)
183 lsb
= hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault
));
184 if (fault
& VM_FAULT_HWPOISON
)
187 force_sig_mceerr(BUS_MCEERR_AR
, (void __user
*)address
, lsb
);
192 force_sig_fault(SIGBUS
, BUS_ADRERR
, (void __user
*)address
);
196 static int mm_fault_error(struct pt_regs
*regs
, unsigned long addr
,
200 * Kernel page fault interrupted by SIGKILL. We have no reason to
201 * continue processing.
203 if (fatal_signal_pending(current
) && !user_mode(regs
))
207 if (fault
& VM_FAULT_OOM
) {
209 * We ran out of memory, or some other thing happened to us that
210 * made us unable to handle the page fault gracefully.
212 if (!user_mode(regs
))
214 pagefault_out_of_memory();
216 if (fault
& (VM_FAULT_SIGBUS
|VM_FAULT_HWPOISON
|
217 VM_FAULT_HWPOISON_LARGE
))
218 return do_sigbus(regs
, addr
, fault
);
219 else if (fault
& VM_FAULT_SIGSEGV
)
220 return bad_area_nosemaphore(regs
, addr
);
227 /* Is this a bad kernel fault ? */
228 static bool bad_kernel_fault(struct pt_regs
*regs
, unsigned long error_code
,
229 unsigned long address
, bool is_write
)
231 int is_exec
= TRAP(regs
) == 0x400;
233 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
234 if (is_exec
&& (error_code
& (DSISR_NOEXEC_OR_G
| DSISR_KEYFAULT
|
236 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
237 address
>= TASK_SIZE
? "exec-protected" : "user",
239 from_kuid(&init_user_ns
, current_uid()));
241 // Kernel exec fault is always bad
245 if (!is_exec
&& address
< TASK_SIZE
&& (error_code
& DSISR_PROTFAULT
) &&
246 !search_exception_tables(regs
->nip
)) {
247 pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
249 from_kuid(&init_user_ns
, current_uid()));
252 // Kernel fault on kernel address is bad
253 if (address
>= TASK_SIZE
)
256 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
257 if (!search_exception_tables(regs
->nip
))
260 // Read/write fault in a valid region (the exception table search passed
261 // above), but blocked by KUAP is bad, it can never succeed.
262 if (bad_kuap_fault(regs
, address
, is_write
))
265 // What's left? Kernel fault on user in well defined regions (extable
266 // matched), and allowed by KUAP in the faulting context.
270 static bool bad_stack_expansion(struct pt_regs
*regs
, unsigned long address
,
271 struct vm_area_struct
*vma
, unsigned int flags
,
275 * N.B. The POWER/Open ABI allows programs to access up to
276 * 288 bytes below the stack pointer.
277 * The kernel signal delivery code writes up to about 1.5kB
278 * below the stack pointer (r1) before decrementing it.
279 * The exec code can write slightly over 640kB to the stack
280 * before setting the user r1. Thus we allow the stack to
281 * expand to 1MB without further checks.
283 if (address
+ 0x100000 < vma
->vm_end
) {
284 struct ppc_inst __user
*nip
= (struct ppc_inst __user
*)regs
->nip
;
285 /* get user regs even if this fault is in kernel mode */
286 struct pt_regs
*uregs
= current
->thread
.regs
;
291 * A user-mode access to an address a long way below
292 * the stack pointer is only valid if the instruction
293 * is one which would update the stack pointer to the
294 * address accessed if the instruction completed,
295 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
296 * (or the byte, halfword, float or double forms).
298 * If we don't check this then any write to the area
299 * between the last mapped region and the stack will
300 * expand the stack rather than segfaulting.
302 if (address
+ 2048 >= uregs
->gpr
[1])
305 if ((flags
& FAULT_FLAG_WRITE
) && (flags
& FAULT_FLAG_USER
) &&
306 access_ok(nip
, sizeof(*nip
))) {
307 struct ppc_inst inst
;
309 if (!probe_user_read_inst(&inst
, nip
))
310 return !store_updates_sp(inst
);
318 #ifdef CONFIG_PPC_MEM_KEYS
319 static bool access_pkey_error(bool is_write
, bool is_exec
, bool is_pkey
,
320 struct vm_area_struct
*vma
)
323 * Make sure to check the VMA so that we do not perform
324 * faults just to hit a pkey fault as soon as we fill in a
325 * page. Only called for current mm, hence foreign == 0
327 if (!arch_vma_access_permitted(vma
, is_write
, is_exec
, 0))
334 static bool access_error(bool is_write
, bool is_exec
, struct vm_area_struct
*vma
)
337 * Allow execution from readable areas if the MMU does not
338 * provide separate controls over reading and executing.
340 * Note: That code used to not be enabled for 4xx/BookE.
341 * It is now as I/D cache coherency for these is done at
342 * set_pte_at() time and I see no reason why the test
343 * below wouldn't be valid on those processors. This -may-
344 * break programs compiled with a really old ABI though.
347 return !(vma
->vm_flags
& VM_EXEC
) &&
348 (cpu_has_feature(CPU_FTR_NOEXECUTE
) ||
349 !(vma
->vm_flags
& (VM_READ
| VM_WRITE
)));
353 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
358 if (unlikely(!vma_is_accessible(vma
)))
361 * We should ideally do the vma pkey access check here. But in the
362 * fault path, handle_mm_fault() also does the same check. To avoid
363 * these multiple checks, we skip it here and handle access error due
369 #ifdef CONFIG_PPC_SMLPAR
370 static inline void cmo_account_page_fault(void)
372 if (firmware_has_feature(FW_FEATURE_CMO
)) {
376 page_ins
= be32_to_cpu(get_lppaca()->page_ins
);
377 page_ins
+= 1 << PAGE_FACTOR
;
378 get_lppaca()->page_ins
= cpu_to_be32(page_ins
);
383 static inline void cmo_account_page_fault(void) { }
384 #endif /* CONFIG_PPC_SMLPAR */
386 #ifdef CONFIG_PPC_BOOK3S
387 static void sanity_check_fault(bool is_write
, bool is_user
,
388 unsigned long error_code
, unsigned long address
)
391 * Userspace trying to access kernel address, we get PROTFAULT for that.
393 if (is_user
&& address
>= TASK_SIZE
) {
394 if ((long)address
== -1)
397 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
398 current
->comm
, current
->pid
, address
,
399 from_kuid(&init_user_ns
, current_uid()));
404 * For hash translation mode, we should never get a
405 * PROTFAULT. Any update to pte to reduce access will result in us
406 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
407 * fault instead of DSISR_PROTFAULT.
409 * A pte update to relax the access will not result in a hash page table
410 * entry invalidate and hence can result in DSISR_PROTFAULT.
411 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
412 * the special !is_write in the below conditional.
414 * For platforms that doesn't supports coherent icache and do support
415 * per page noexec bit, we do setup things such that we do the
416 * sync between D/I cache via fault. But that is handled via low level
417 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
420 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
421 * check should handle those and hence we should fall to the bad_area
422 * handling correctly.
424 * For embedded with per page exec support that doesn't support coherent
425 * icache we do get PROTFAULT and we handle that D/I cache sync in
426 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
427 * is conditional for server MMU.
429 * For radix, we can get prot fault for autonuma case, because radix
430 * page table will have them marked noaccess for user.
432 if (radix_enabled() || is_write
)
435 WARN_ON_ONCE(error_code
& DSISR_PROTFAULT
);
438 static void sanity_check_fault(bool is_write
, bool is_user
,
439 unsigned long error_code
, unsigned long address
) { }
440 #endif /* CONFIG_PPC_BOOK3S */
443 * Define the correct "is_write" bit in error_code based
444 * on the processor family
446 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
447 #define page_fault_is_write(__err) ((__err) & ESR_DST)
448 #define page_fault_is_bad(__err) (0)
450 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
451 #if defined(CONFIG_PPC_8xx)
452 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
453 #elif defined(CONFIG_PPC64)
454 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
456 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
461 * For 600- and 800-family processors, the error_code parameter is DSISR
462 * for a data fault, SRR1 for an instruction fault. For 400-family processors
463 * the error_code parameter is ESR for a data fault, 0 for an instruction
465 * For 64-bit processors, the error_code parameter is
466 * - DSISR for a non-SLB data access fault,
467 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
470 * The return value is 0 if the fault was handled, or the signal
471 * number if this is a kernel fault that can't be handled here.
473 static int __do_page_fault(struct pt_regs
*regs
, unsigned long address
,
474 unsigned long error_code
)
476 struct vm_area_struct
* vma
;
477 struct mm_struct
*mm
= current
->mm
;
478 unsigned int flags
= FAULT_FLAG_DEFAULT
;
479 int is_exec
= TRAP(regs
) == 0x400;
480 int is_user
= user_mode(regs
);
481 int is_write
= page_fault_is_write(error_code
);
482 vm_fault_t fault
, major
= 0;
483 bool must_retry
= false;
484 bool kprobe_fault
= kprobe_page_fault(regs
, 11);
486 if (unlikely(debugger_fault_handler(regs
) || kprobe_fault
))
489 if (unlikely(page_fault_is_bad(error_code
))) {
491 _exception(SIGBUS
, regs
, BUS_OBJERR
, address
);
497 /* Additional sanity check(s) */
498 sanity_check_fault(is_write
, is_user
, error_code
, address
);
501 * The kernel should never take an execute fault nor should it
502 * take a page fault to a kernel address or a page fault to a user
503 * address outside of dedicated places
505 if (unlikely(!is_user
&& bad_kernel_fault(regs
, error_code
, address
, is_write
)))
509 * If we're in an interrupt, have no user context or are running
510 * in a region with pagefaults disabled then we must not take the fault
512 if (unlikely(faulthandler_disabled() || !mm
)) {
514 printk_ratelimited(KERN_ERR
"Page fault in user mode"
515 " with faulthandler_disabled()=%d"
517 faulthandler_disabled(), mm
);
518 return bad_area_nosemaphore(regs
, address
);
521 /* We restore the interrupt state now */
522 if (!arch_irq_disabled_regs(regs
))
525 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
528 * We want to do this outside mmap_sem, because reading code around nip
529 * can result in fault, which will cause a deadlock when called with
533 flags
|= FAULT_FLAG_USER
;
535 flags
|= FAULT_FLAG_WRITE
;
537 flags
|= FAULT_FLAG_INSTRUCTION
;
539 /* When running in the kernel we expect faults to occur only to
540 * addresses in user space. All other faults represent errors in the
541 * kernel and should generate an OOPS. Unfortunately, in the case of an
542 * erroneous fault occurring in a code path which already holds mmap_sem
543 * we will deadlock attempting to validate the fault against the
544 * address space. Luckily the kernel only validly references user
545 * space from well defined areas of code, which are listed in the
548 * As the vast majority of faults will be valid we will only perform
549 * the source reference check when there is a possibility of a deadlock.
550 * Attempt to lock the address space, if we cannot we then validate the
551 * source. If this is invalid we can skip the address space check,
552 * thus avoiding the deadlock.
554 if (unlikely(!mmap_read_trylock(mm
))) {
555 if (!is_user
&& !search_exception_tables(regs
->nip
))
556 return bad_area_nosemaphore(regs
, address
);
562 * The above down_read_trylock() might have succeeded in
563 * which case we'll have missed the might_sleep() from
569 vma
= find_vma(mm
, address
);
571 return bad_area(regs
, address
);
572 if (likely(vma
->vm_start
<= address
))
574 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
)))
575 return bad_area(regs
, address
);
577 /* The stack is being expanded, check if it's valid */
578 if (unlikely(bad_stack_expansion(regs
, address
, vma
, flags
,
581 return bad_area(regs
, address
);
583 mmap_read_unlock(mm
);
584 if (fault_in_pages_readable((const char __user
*)regs
->nip
,
585 sizeof(unsigned int)))
586 return bad_area_nosemaphore(regs
, address
);
590 /* Try to expand it */
591 if (unlikely(expand_stack(vma
, address
)))
592 return bad_area(regs
, address
);
596 #ifdef CONFIG_PPC_MEM_KEYS
597 if (unlikely(access_pkey_error(is_write
, is_exec
,
598 (error_code
& DSISR_KEYFAULT
), vma
)))
599 return bad_access_pkey(regs
, address
, vma
);
600 #endif /* CONFIG_PPC_MEM_KEYS */
602 if (unlikely(access_error(is_write
, is_exec
, vma
)))
603 return bad_access(regs
, address
);
606 * If for any reason at all we couldn't handle the fault,
607 * make sure we exit gracefully rather than endlessly redo
610 fault
= handle_mm_fault(vma
, address
, flags
);
612 major
|= fault
& VM_FAULT_MAJOR
;
614 if (fault_signal_pending(fault
, regs
))
615 return user_mode(regs
) ? 0 : SIGBUS
;
618 * Handle the retry right now, the mmap_sem has been released in that
621 if (unlikely(fault
& VM_FAULT_RETRY
)) {
622 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
623 flags
|= FAULT_FLAG_TRIED
;
628 mmap_read_unlock(current
->mm
);
630 if (unlikely(fault
& VM_FAULT_ERROR
))
631 return mm_fault_error(regs
, address
, fault
);
634 * Major/minor page fault accounting.
638 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
, address
);
639 cmo_account_page_fault();
642 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
, address
);
646 NOKPROBE_SYMBOL(__do_page_fault
);
648 int do_page_fault(struct pt_regs
*regs
, unsigned long address
,
649 unsigned long error_code
)
651 enum ctx_state prev_state
= exception_enter();
652 int rc
= __do_page_fault(regs
, address
, error_code
);
653 exception_exit(prev_state
);
656 NOKPROBE_SYMBOL(do_page_fault
);
659 * bad_page_fault is called when we have a bad access from the kernel.
660 * It is called from the DSI and ISI handlers in head.S and from some
661 * of the procedures in traps.c.
663 void bad_page_fault(struct pt_regs
*regs
, unsigned long address
, int sig
)
665 const struct exception_table_entry
*entry
;
666 int is_write
= page_fault_is_write(regs
->dsisr
);
668 /* Are we prepared to handle this fault? */
669 if ((entry
= search_exception_tables(regs
->nip
)) != NULL
) {
670 regs
->nip
= extable_fixup(entry
);
674 /* kernel has accessed a bad area */
676 switch (TRAP(regs
)) {
680 pr_alert("BUG: %s on %s at 0x%08lx\n",
681 regs
->dar
< PAGE_SIZE
? "Kernel NULL pointer dereference" :
682 "Unable to handle kernel data access",
683 is_write
? "write" : "read", regs
->dar
);
687 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
688 regs
->nip
< PAGE_SIZE
? " (NULL pointer?)\n" : "\n");
691 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
695 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
699 printk(KERN_ALERT
"Faulting instruction address: 0x%08lx\n",
702 if (task_stack_end_corrupted(current
))
703 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
705 die("Kernel access of bad area", regs
, sig
);