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,
191 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, current);
195 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
199 * Kernel page fault interrupted by SIGKILL. We have no reason to
200 * continue processing.
202 if (fatal_signal_pending(current) && !user_mode(regs))
206 if (fault & VM_FAULT_OOM) {
208 * We ran out of memory, or some other thing happened to us that
209 * made us unable to handle the page fault gracefully.
211 if (!user_mode(regs))
213 pagefault_out_of_memory();
215 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
216 VM_FAULT_HWPOISON_LARGE))
217 return do_sigbus(regs, addr, fault);
218 else if (fault & VM_FAULT_SIGSEGV)
219 return bad_area_nosemaphore(regs, addr);
226 /* Is this a bad kernel fault ? */
227 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
228 unsigned long address, bool is_write)
230 int is_exec = TRAP(regs) == 0x400;
232 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
233 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
235 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
236 address >= TASK_SIZE ? "exec-protected" : "user",
238 from_kuid(&init_user_ns, current_uid()));
240 // Kernel exec fault is always bad
244 if (!is_exec && address < TASK_SIZE && (error_code & DSISR_PROTFAULT) &&
245 !search_exception_tables(regs->nip)) {
246 pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
248 from_kuid(&init_user_ns, current_uid()));
251 // Kernel fault on kernel address is bad
252 if (address >= TASK_SIZE)
255 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
256 if (!search_exception_tables(regs->nip))
259 // Read/write fault in a valid region (the exception table search passed
260 // above), but blocked by KUAP is bad, it can never succeed.
261 if (bad_kuap_fault(regs, is_write))
264 // What's left? Kernel fault on user in well defined regions (extable
265 // matched), and allowed by KUAP in the faulting context.
269 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
270 struct vm_area_struct *vma, unsigned int flags,
274 * N.B. The POWER/Open ABI allows programs to access up to
275 * 288 bytes below the stack pointer.
276 * The kernel signal delivery code writes up to about 1.5kB
277 * below the stack pointer (r1) before decrementing it.
278 * The exec code can write slightly over 640kB to the stack
279 * before setting the user r1. Thus we allow the stack to
280 * expand to 1MB without further checks.
282 if (address + 0x100000 < vma->vm_end) {
283 unsigned int __user *nip = (unsigned int __user *)regs->nip;
284 /* get user regs even if this fault is in kernel mode */
285 struct pt_regs *uregs = current->thread.regs;
290 * A user-mode access to an address a long way below
291 * the stack pointer is only valid if the instruction
292 * is one which would update the stack pointer to the
293 * address accessed if the instruction completed,
294 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
295 * (or the byte, halfword, float or double forms).
297 * If we don't check this then any write to the area
298 * between the last mapped region and the stack will
299 * expand the stack rather than segfaulting.
301 if (address + 2048 >= uregs->gpr[1])
304 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
305 access_ok(nip, sizeof(*nip))) {
310 res = __get_user_inatomic(inst, nip);
313 return !store_updates_sp(inst);
321 static bool access_error(bool is_write, bool is_exec,
322 struct vm_area_struct *vma)
325 * Allow execution from readable areas if the MMU does not
326 * provide separate controls over reading and executing.
328 * Note: That code used to not be enabled for 4xx/BookE.
329 * It is now as I/D cache coherency for these is done at
330 * set_pte_at() time and I see no reason why the test
331 * below wouldn't be valid on those processors. This -may-
332 * break programs compiled with a really old ABI though.
335 return !(vma->vm_flags & VM_EXEC) &&
336 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
337 !(vma->vm_flags & (VM_READ | VM_WRITE)));
341 if (unlikely(!(vma->vm_flags & VM_WRITE)))
346 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
349 * We should ideally do the vma pkey access check here. But in the
350 * fault path, handle_mm_fault() also does the same check. To avoid
351 * these multiple checks, we skip it here and handle access error due
357 #ifdef CONFIG_PPC_SMLPAR
358 static inline void cmo_account_page_fault(void)
360 if (firmware_has_feature(FW_FEATURE_CMO)) {
364 page_ins = be32_to_cpu(get_lppaca()->page_ins);
365 page_ins += 1 << PAGE_FACTOR;
366 get_lppaca()->page_ins = cpu_to_be32(page_ins);
371 static inline void cmo_account_page_fault(void) { }
372 #endif /* CONFIG_PPC_SMLPAR */
374 #ifdef CONFIG_PPC_BOOK3S
375 static void sanity_check_fault(bool is_write, bool is_user,
376 unsigned long error_code, unsigned long address)
379 * Userspace trying to access kernel address, we get PROTFAULT for that.
381 if (is_user && address >= TASK_SIZE) {
382 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
383 current->comm, current->pid, address,
384 from_kuid(&init_user_ns, current_uid()));
389 * For hash translation mode, we should never get a
390 * PROTFAULT. Any update to pte to reduce access will result in us
391 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
392 * fault instead of DSISR_PROTFAULT.
394 * A pte update to relax the access will not result in a hash page table
395 * entry invalidate and hence can result in DSISR_PROTFAULT.
396 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
397 * the special !is_write in the below conditional.
399 * For platforms that doesn't supports coherent icache and do support
400 * per page noexec bit, we do setup things such that we do the
401 * sync between D/I cache via fault. But that is handled via low level
402 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
405 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
406 * check should handle those and hence we should fall to the bad_area
407 * handling correctly.
409 * For embedded with per page exec support that doesn't support coherent
410 * icache we do get PROTFAULT and we handle that D/I cache sync in
411 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
412 * is conditional for server MMU.
414 * For radix, we can get prot fault for autonuma case, because radix
415 * page table will have them marked noaccess for user.
417 if (radix_enabled() || is_write)
420 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
423 static void sanity_check_fault(bool is_write, bool is_user,
424 unsigned long error_code, unsigned long address) { }
425 #endif /* CONFIG_PPC_BOOK3S */
428 * Define the correct "is_write" bit in error_code based
429 * on the processor family
431 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
432 #define page_fault_is_write(__err) ((__err) & ESR_DST)
433 #define page_fault_is_bad(__err) (0)
435 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
436 #if defined(CONFIG_PPC_8xx)
437 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
438 #elif defined(CONFIG_PPC64)
439 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
441 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
446 * For 600- and 800-family processors, the error_code parameter is DSISR
447 * for a data fault, SRR1 for an instruction fault. For 400-family processors
448 * the error_code parameter is ESR for a data fault, 0 for an instruction
450 * For 64-bit processors, the error_code parameter is
451 * - DSISR for a non-SLB data access fault,
452 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
455 * The return value is 0 if the fault was handled, or the signal
456 * number if this is a kernel fault that can't be handled here.
458 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
459 unsigned long error_code)
461 struct vm_area_struct * vma;
462 struct mm_struct *mm = current->mm;
463 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
464 int is_exec = TRAP(regs) == 0x400;
465 int is_user = user_mode(regs);
466 int is_write = page_fault_is_write(error_code);
467 vm_fault_t fault, major = 0;
468 bool must_retry = false;
470 if (notify_page_fault(regs))
473 if (unlikely(page_fault_is_bad(error_code))) {
475 _exception(SIGBUS, regs, BUS_OBJERR, address);
481 /* Additional sanity check(s) */
482 sanity_check_fault(is_write, is_user, error_code, address);
485 * The kernel should never take an execute fault nor should it
486 * take a page fault to a kernel address or a page fault to a user
487 * address outside of dedicated places
489 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write)))
493 * If we're in an interrupt, have no user context or are running
494 * in a region with pagefaults disabled then we must not take the fault
496 if (unlikely(faulthandler_disabled() || !mm)) {
498 printk_ratelimited(KERN_ERR "Page fault in user mode"
499 " with faulthandler_disabled()=%d"
501 faulthandler_disabled(), mm);
502 return bad_area_nosemaphore(regs, address);
505 /* We restore the interrupt state now */
506 if (!arch_irq_disabled_regs(regs))
509 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
511 if (error_code & DSISR_KEYFAULT)
512 return bad_key_fault_exception(regs, address,
513 get_mm_addr_key(mm, address));
516 * We want to do this outside mmap_sem, because reading code around nip
517 * can result in fault, which will cause a deadlock when called with
521 flags |= FAULT_FLAG_USER;
523 flags |= FAULT_FLAG_WRITE;
525 flags |= FAULT_FLAG_INSTRUCTION;
527 /* When running in the kernel we expect faults to occur only to
528 * addresses in user space. All other faults represent errors in the
529 * kernel and should generate an OOPS. Unfortunately, in the case of an
530 * erroneous fault occurring in a code path which already holds mmap_sem
531 * we will deadlock attempting to validate the fault against the
532 * address space. Luckily the kernel only validly references user
533 * space from well defined areas of code, which are listed in the
536 * As the vast majority of faults will be valid we will only perform
537 * the source reference check when there is a possibility of a deadlock.
538 * Attempt to lock the address space, if we cannot we then validate the
539 * source. If this is invalid we can skip the address space check,
540 * thus avoiding the deadlock.
542 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
543 if (!is_user && !search_exception_tables(regs->nip))
544 return bad_area_nosemaphore(regs, address);
547 down_read(&mm->mmap_sem);
550 * The above down_read_trylock() might have succeeded in
551 * which case we'll have missed the might_sleep() from
557 vma = find_vma(mm, address);
559 return bad_area(regs, address);
560 if (likely(vma->vm_start <= address))
562 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
563 return bad_area(regs, address);
565 /* The stack is being expanded, check if it's valid */
566 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
569 return bad_area(regs, address);
571 up_read(&mm->mmap_sem);
572 if (fault_in_pages_readable((const char __user *)regs->nip,
573 sizeof(unsigned int)))
574 return bad_area_nosemaphore(regs, address);
578 /* Try to expand it */
579 if (unlikely(expand_stack(vma, address)))
580 return bad_area(regs, address);
583 if (unlikely(access_error(is_write, is_exec, vma)))
584 return bad_access(regs, address);
587 * If for any reason at all we couldn't handle the fault,
588 * make sure we exit gracefully rather than endlessly redo
591 fault = handle_mm_fault(vma, address, flags);
593 #ifdef CONFIG_PPC_MEM_KEYS
595 * we skipped checking for access error due to key earlier.
596 * Check that using handle_mm_fault error return.
598 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
599 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
601 int pkey = vma_pkey(vma);
603 up_read(&mm->mmap_sem);
604 return bad_key_fault_exception(regs, address, pkey);
606 #endif /* CONFIG_PPC_MEM_KEYS */
608 major |= fault & VM_FAULT_MAJOR;
611 * Handle the retry right now, the mmap_sem has been released in that
614 if (unlikely(fault & VM_FAULT_RETRY)) {
615 /* We retry only once */
616 if (flags & FAULT_FLAG_ALLOW_RETRY) {
618 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
621 flags &= ~FAULT_FLAG_ALLOW_RETRY;
622 flags |= FAULT_FLAG_TRIED;
623 if (!fatal_signal_pending(current))
628 * User mode? Just return to handle the fatal exception otherwise
629 * return to bad_page_fault
631 return is_user ? 0 : SIGBUS;
634 up_read(¤t->mm->mmap_sem);
636 if (unlikely(fault & VM_FAULT_ERROR))
637 return mm_fault_error(regs, address, fault);
640 * Major/minor page fault accounting.
644 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
645 cmo_account_page_fault();
648 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
652 NOKPROBE_SYMBOL(__do_page_fault);
654 int do_page_fault(struct pt_regs *regs, unsigned long address,
655 unsigned long error_code)
657 enum ctx_state prev_state = exception_enter();
658 int rc = __do_page_fault(regs, address, error_code);
659 exception_exit(prev_state);
662 NOKPROBE_SYMBOL(do_page_fault);
665 * bad_page_fault is called when we have a bad access from the kernel.
666 * It is called from the DSI and ISI handlers in head.S and from some
667 * of the procedures in traps.c.
669 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
671 const struct exception_table_entry *entry;
673 /* Are we prepared to handle this fault? */
674 if ((entry = search_exception_tables(regs->nip)) != NULL) {
675 regs->nip = extable_fixup(entry);
679 /* kernel has accessed a bad area */
681 switch (TRAP(regs)) {
685 pr_alert("BUG: %s at 0x%08lx\n",
686 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
687 "Unable to handle kernel data access", regs->dar);
691 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
692 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
695 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
699 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
703 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
706 if (task_stack_end_corrupted(current))
707 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
709 die("Kernel access of bad area", regs, sig);