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.
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>
48 static inline bool notify_page_fault(struct pt_regs *regs)
53 /* kprobe_running() needs smp_processor_id() */
54 if (!user_mode(regs)) {
56 if (kprobe_running() && kprobe_fault_handler(regs, 11))
60 #endif /* CONFIG_KPROBES */
62 if (unlikely(debugger_fault_handler(regs)))
69 * Check whether the instruction inst is a store using
70 * an update addressing form which will update r1.
72 static bool store_updates_sp(unsigned int inst)
74 /* check for 1 in the rA field */
75 if (((inst >> 16) & 0x1f) != 1)
77 /* check major opcode */
85 case OP_STD: /* std or stdu */
86 return (inst & 3) == 1;
88 /* check minor opcode */
89 switch ((inst >> 1) & 0x3ff) {
94 case OP_31_XOP_STFSUX:
95 case OP_31_XOP_STFDUX:
102 * do_page_fault error handling helpers
106 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
109 * If we are in kernel mode, bail out with a SEGV, this will
110 * be caught by the assembly which will restore the non-volatile
111 * registers before calling bad_page_fault()
113 if (!user_mode(regs))
116 _exception(SIGSEGV, regs, si_code, address);
121 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
123 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
126 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
128 struct mm_struct *mm = current->mm;
131 * Something tried to access memory that isn't in our memory map..
132 * Fix it, but check if it's kernel or user first..
134 up_read(&mm->mmap_sem);
136 return __bad_area_nosemaphore(regs, address, si_code);
139 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
141 return __bad_area(regs, address, SEGV_MAPERR);
144 static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address,
148 * If we are in kernel mode, bail out with a SEGV, this will
149 * be caught by the assembly which will restore the non-volatile
150 * registers before calling bad_page_fault()
152 if (!user_mode(regs))
155 _exception_pkey(SIGSEGV, regs, SEGV_PKUERR, address, pkey);
160 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
162 return __bad_area(regs, address, SEGV_ACCERR);
165 static int do_sigbus(struct pt_regs *regs, unsigned long address,
170 if (!user_mode(regs))
173 current->thread.trap_nr = BUS_ADRERR;
174 clear_siginfo(&info);
175 info.si_signo = SIGBUS;
177 info.si_code = BUS_ADRERR;
178 info.si_addr = (void __user *)address;
179 #ifdef CONFIG_MEMORY_FAILURE
180 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
181 unsigned int lsb = 0; /* shutup gcc */
183 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
184 current->comm, current->pid, address);
186 if (fault & VM_FAULT_HWPOISON_LARGE)
187 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
188 if (fault & VM_FAULT_HWPOISON)
191 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb,
197 force_sig_info(SIGBUS, &info, current);
201 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
205 * Kernel page fault interrupted by SIGKILL. We have no reason to
206 * continue processing.
208 if (fatal_signal_pending(current) && !user_mode(regs))
212 if (fault & VM_FAULT_OOM) {
214 * We ran out of memory, or some other thing happened to us that
215 * made us unable to handle the page fault gracefully.
217 if (!user_mode(regs))
219 pagefault_out_of_memory();
221 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
222 VM_FAULT_HWPOISON_LARGE))
223 return do_sigbus(regs, addr, fault);
224 else if (fault & VM_FAULT_SIGSEGV)
225 return bad_area_nosemaphore(regs, addr);
232 /* Is this a bad kernel fault ? */
233 static bool bad_kernel_fault(bool is_exec, unsigned long error_code,
234 unsigned long address)
236 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT))) {
237 printk_ratelimited(KERN_CRIT "kernel tried to execute"
238 " exec-protected page (%lx) -"
239 "exploit attempt? (uid: %d)\n",
240 address, from_kuid(&init_user_ns,
243 return is_exec || (address >= TASK_SIZE);
246 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
247 struct vm_area_struct *vma, unsigned int flags,
251 * N.B. The POWER/Open ABI allows programs to access up to
252 * 288 bytes below the stack pointer.
253 * The kernel signal delivery code writes up to about 1.5kB
254 * below the stack pointer (r1) before decrementing it.
255 * The exec code can write slightly over 640kB to the stack
256 * before setting the user r1. Thus we allow the stack to
257 * expand to 1MB without further checks.
259 if (address + 0x100000 < vma->vm_end) {
260 unsigned int __user *nip = (unsigned int __user *)regs->nip;
261 /* get user regs even if this fault is in kernel mode */
262 struct pt_regs *uregs = current->thread.regs;
267 * A user-mode access to an address a long way below
268 * the stack pointer is only valid if the instruction
269 * is one which would update the stack pointer to the
270 * address accessed if the instruction completed,
271 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
272 * (or the byte, halfword, float or double forms).
274 * If we don't check this then any write to the area
275 * between the last mapped region and the stack will
276 * expand the stack rather than segfaulting.
278 if (address + 2048 >= uregs->gpr[1])
281 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
282 access_ok(VERIFY_READ, nip, sizeof(*nip))) {
287 res = __get_user_inatomic(inst, nip);
290 return !store_updates_sp(inst);
298 static bool access_error(bool is_write, bool is_exec,
299 struct vm_area_struct *vma)
302 * Allow execution from readable areas if the MMU does not
303 * provide separate controls over reading and executing.
305 * Note: That code used to not be enabled for 4xx/BookE.
306 * It is now as I/D cache coherency for these is done at
307 * set_pte_at() time and I see no reason why the test
308 * below wouldn't be valid on those processors. This -may-
309 * break programs compiled with a really old ABI though.
312 return !(vma->vm_flags & VM_EXEC) &&
313 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
314 !(vma->vm_flags & (VM_READ | VM_WRITE)));
318 if (unlikely(!(vma->vm_flags & VM_WRITE)))
323 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
326 * We should ideally do the vma pkey access check here. But in the
327 * fault path, handle_mm_fault() also does the same check. To avoid
328 * these multiple checks, we skip it here and handle access error due
334 #ifdef CONFIG_PPC_SMLPAR
335 static inline void cmo_account_page_fault(void)
337 if (firmware_has_feature(FW_FEATURE_CMO)) {
341 page_ins = be32_to_cpu(get_lppaca()->page_ins);
342 page_ins += 1 << PAGE_FACTOR;
343 get_lppaca()->page_ins = cpu_to_be32(page_ins);
348 static inline void cmo_account_page_fault(void) { }
349 #endif /* CONFIG_PPC_SMLPAR */
351 #ifdef CONFIG_PPC_STD_MMU
352 static void sanity_check_fault(bool is_write, unsigned long error_code)
355 * For hash translation mode, we should never get a
356 * PROTFAULT. Any update to pte to reduce access will result in us
357 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
358 * fault instead of DSISR_PROTFAULT.
360 * A pte update to relax the access will not result in a hash page table
361 * entry invalidate and hence can result in DSISR_PROTFAULT.
362 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
363 * the special !is_write in the below conditional.
365 * For platforms that doesn't supports coherent icache and do support
366 * per page noexec bit, we do setup things such that we do the
367 * sync between D/I cache via fault. But that is handled via low level
368 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
371 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
372 * check should handle those and hence we should fall to the bad_area
373 * handling correctly.
375 * For embedded with per page exec support that doesn't support coherent
376 * icache we do get PROTFAULT and we handle that D/I cache sync in
377 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
378 * is conditional for server MMU.
380 * For radix, we can get prot fault for autonuma case, because radix
381 * page table will have them marked noaccess for user.
383 if (!radix_enabled() && !is_write)
384 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
387 static void sanity_check_fault(bool is_write, unsigned long error_code) { }
388 #endif /* CONFIG_PPC_STD_MMU */
391 * Define the correct "is_write" bit in error_code based
392 * on the processor family
394 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
395 #define page_fault_is_write(__err) ((__err) & ESR_DST)
396 #define page_fault_is_bad(__err) (0)
398 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
399 #if defined(CONFIG_PPC_8xx)
400 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
401 #elif defined(CONFIG_PPC64)
402 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
404 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
409 * For 600- and 800-family processors, the error_code parameter is DSISR
410 * for a data fault, SRR1 for an instruction fault. For 400-family processors
411 * the error_code parameter is ESR for a data fault, 0 for an instruction
413 * For 64-bit processors, the error_code parameter is
414 * - DSISR for a non-SLB data access fault,
415 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
418 * The return value is 0 if the fault was handled, or the signal
419 * number if this is a kernel fault that can't be handled here.
421 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
422 unsigned long error_code)
424 struct vm_area_struct * vma;
425 struct mm_struct *mm = current->mm;
426 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
427 int is_exec = TRAP(regs) == 0x400;
428 int is_user = user_mode(regs);
429 int is_write = page_fault_is_write(error_code);
430 vm_fault_t fault, major = 0;
431 bool must_retry = false;
433 if (notify_page_fault(regs))
436 if (unlikely(page_fault_is_bad(error_code))) {
438 _exception(SIGBUS, regs, BUS_OBJERR, address);
444 /* Additional sanity check(s) */
445 sanity_check_fault(is_write, error_code);
448 * The kernel should never take an execute fault nor should it
449 * take a page fault to a kernel address.
451 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
455 * If we're in an interrupt, have no user context or are running
456 * in a region with pagefaults disabled then we must not take the fault
458 if (unlikely(faulthandler_disabled() || !mm)) {
460 printk_ratelimited(KERN_ERR "Page fault in user mode"
461 " with faulthandler_disabled()=%d"
463 faulthandler_disabled(), mm);
464 return bad_area_nosemaphore(regs, address);
467 /* We restore the interrupt state now */
468 if (!arch_irq_disabled_regs(regs))
471 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
473 if (error_code & DSISR_KEYFAULT)
474 return bad_key_fault_exception(regs, address,
475 get_mm_addr_key(mm, address));
478 * We want to do this outside mmap_sem, because reading code around nip
479 * can result in fault, which will cause a deadlock when called with
483 flags |= FAULT_FLAG_USER;
485 flags |= FAULT_FLAG_WRITE;
487 flags |= FAULT_FLAG_INSTRUCTION;
489 /* When running in the kernel we expect faults to occur only to
490 * addresses in user space. All other faults represent errors in the
491 * kernel and should generate an OOPS. Unfortunately, in the case of an
492 * erroneous fault occurring in a code path which already holds mmap_sem
493 * we will deadlock attempting to validate the fault against the
494 * address space. Luckily the kernel only validly references user
495 * space from well defined areas of code, which are listed in the
498 * As the vast majority of faults will be valid we will only perform
499 * the source reference check when there is a possibility of a deadlock.
500 * Attempt to lock the address space, if we cannot we then validate the
501 * source. If this is invalid we can skip the address space check,
502 * thus avoiding the deadlock.
504 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
505 if (!is_user && !search_exception_tables(regs->nip))
506 return bad_area_nosemaphore(regs, address);
509 down_read(&mm->mmap_sem);
512 * The above down_read_trylock() might have succeeded in
513 * which case we'll have missed the might_sleep() from
519 vma = find_vma(mm, address);
521 return bad_area(regs, address);
522 if (likely(vma->vm_start <= address))
524 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
525 return bad_area(regs, address);
527 /* The stack is being expanded, check if it's valid */
528 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
531 return bad_area(regs, address);
533 up_read(&mm->mmap_sem);
534 if (fault_in_pages_readable((const char __user *)regs->nip,
535 sizeof(unsigned int)))
536 return bad_area_nosemaphore(regs, address);
540 /* Try to expand it */
541 if (unlikely(expand_stack(vma, address)))
542 return bad_area(regs, address);
545 if (unlikely(access_error(is_write, is_exec, vma)))
546 return bad_access(regs, address);
549 * If for any reason at all we couldn't handle the fault,
550 * make sure we exit gracefully rather than endlessly redo
553 fault = handle_mm_fault(vma, address, flags);
555 #ifdef CONFIG_PPC_MEM_KEYS
557 * we skipped checking for access error due to key earlier.
558 * Check that using handle_mm_fault error return.
560 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
561 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
563 int pkey = vma_pkey(vma);
565 up_read(&mm->mmap_sem);
566 return bad_key_fault_exception(regs, address, pkey);
568 #endif /* CONFIG_PPC_MEM_KEYS */
570 major |= fault & VM_FAULT_MAJOR;
573 * Handle the retry right now, the mmap_sem has been released in that
576 if (unlikely(fault & VM_FAULT_RETRY)) {
577 /* We retry only once */
578 if (flags & FAULT_FLAG_ALLOW_RETRY) {
580 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
583 flags &= ~FAULT_FLAG_ALLOW_RETRY;
584 flags |= FAULT_FLAG_TRIED;
585 if (!fatal_signal_pending(current))
590 * User mode? Just return to handle the fatal exception otherwise
591 * return to bad_page_fault
593 return is_user ? 0 : SIGBUS;
596 up_read(¤t->mm->mmap_sem);
598 if (unlikely(fault & VM_FAULT_ERROR))
599 return mm_fault_error(regs, address, fault);
602 * Major/minor page fault accounting.
606 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
607 cmo_account_page_fault();
610 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
614 NOKPROBE_SYMBOL(__do_page_fault);
616 int do_page_fault(struct pt_regs *regs, unsigned long address,
617 unsigned long error_code)
619 enum ctx_state prev_state = exception_enter();
620 int rc = __do_page_fault(regs, address, error_code);
621 exception_exit(prev_state);
624 NOKPROBE_SYMBOL(do_page_fault);
627 * bad_page_fault is called when we have a bad access from the kernel.
628 * It is called from the DSI and ISI handlers in head.S and from some
629 * of the procedures in traps.c.
631 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
633 const struct exception_table_entry *entry;
635 /* Are we prepared to handle this fault? */
636 if ((entry = search_exception_tables(regs->nip)) != NULL) {
637 regs->nip = extable_fixup(entry);
641 /* kernel has accessed a bad area */
643 switch (TRAP(regs)) {
646 printk(KERN_ALERT "Unable to handle kernel paging request for "
647 "data at address 0x%08lx\n", regs->dar);
651 printk(KERN_ALERT "Unable to handle kernel paging request for "
652 "instruction fetch\n");
655 printk(KERN_ALERT "Unable to handle kernel paging request for "
656 "unaligned access at address 0x%08lx\n", regs->dar);
659 printk(KERN_ALERT "Unable to handle kernel paging request for "
663 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
666 if (task_stack_end_corrupted(current))
667 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
669 die("Kernel access of bad area", regs, sig);