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>
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(regs, 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,
168 if (!user_mode(regs))
171 current->thread.trap_nr = BUS_ADRERR;
172 #ifdef CONFIG_MEMORY_FAILURE
173 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
174 unsigned int lsb = 0; /* shutup gcc */
176 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
177 current->comm, current->pid, address);
179 if (fault & VM_FAULT_HWPOISON_LARGE)
180 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
181 if (fault & VM_FAULT_HWPOISON)
184 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb,
190 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, current);
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(bool is_exec, unsigned long error_code,
227 unsigned long address)
229 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
230 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
232 printk_ratelimited(KERN_CRIT "kernel tried to execute"
233 " exec-protected page (%lx) -"
234 "exploit attempt? (uid: %d)\n",
235 address, from_kuid(&init_user_ns,
238 return is_exec || (address >= TASK_SIZE);
241 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
242 struct vm_area_struct *vma, unsigned int flags,
246 * N.B. The POWER/Open ABI allows programs to access up to
247 * 288 bytes below the stack pointer.
248 * The kernel signal delivery code writes up to about 1.5kB
249 * below the stack pointer (r1) before decrementing it.
250 * The exec code can write slightly over 640kB to the stack
251 * before setting the user r1. Thus we allow the stack to
252 * expand to 1MB without further checks.
254 if (address + 0x100000 < vma->vm_end) {
255 unsigned int __user *nip = (unsigned int __user *)regs->nip;
256 /* get user regs even if this fault is in kernel mode */
257 struct pt_regs *uregs = current->thread.regs;
262 * A user-mode access to an address a long way below
263 * the stack pointer is only valid if the instruction
264 * is one which would update the stack pointer to the
265 * address accessed if the instruction completed,
266 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
267 * (or the byte, halfword, float or double forms).
269 * If we don't check this then any write to the area
270 * between the last mapped region and the stack will
271 * expand the stack rather than segfaulting.
273 if (address + 2048 >= uregs->gpr[1])
276 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
277 access_ok(VERIFY_READ, nip, sizeof(*nip))) {
282 res = __get_user_inatomic(inst, nip);
285 return !store_updates_sp(inst);
293 static bool access_error(bool is_write, bool is_exec,
294 struct vm_area_struct *vma)
297 * Allow execution from readable areas if the MMU does not
298 * provide separate controls over reading and executing.
300 * Note: That code used to not be enabled for 4xx/BookE.
301 * It is now as I/D cache coherency for these is done at
302 * set_pte_at() time and I see no reason why the test
303 * below wouldn't be valid on those processors. This -may-
304 * break programs compiled with a really old ABI though.
307 return !(vma->vm_flags & VM_EXEC) &&
308 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
309 !(vma->vm_flags & (VM_READ | VM_WRITE)));
313 if (unlikely(!(vma->vm_flags & VM_WRITE)))
318 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
321 * We should ideally do the vma pkey access check here. But in the
322 * fault path, handle_mm_fault() also does the same check. To avoid
323 * these multiple checks, we skip it here and handle access error due
329 #ifdef CONFIG_PPC_SMLPAR
330 static inline void cmo_account_page_fault(void)
332 if (firmware_has_feature(FW_FEATURE_CMO)) {
336 page_ins = be32_to_cpu(get_lppaca()->page_ins);
337 page_ins += 1 << PAGE_FACTOR;
338 get_lppaca()->page_ins = cpu_to_be32(page_ins);
343 static inline void cmo_account_page_fault(void) { }
344 #endif /* CONFIG_PPC_SMLPAR */
346 #ifdef CONFIG_PPC_BOOK3S
347 static void sanity_check_fault(bool is_write, bool is_user,
348 unsigned long error_code, unsigned long address)
351 * Userspace trying to access kernel address, we get PROTFAULT for that.
353 if (is_user && address >= TASK_SIZE) {
354 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
355 current->comm, current->pid, address,
356 from_kuid(&init_user_ns, current_uid()));
361 * For hash translation mode, we should never get a
362 * PROTFAULT. Any update to pte to reduce access will result in us
363 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
364 * fault instead of DSISR_PROTFAULT.
366 * A pte update to relax the access will not result in a hash page table
367 * entry invalidate and hence can result in DSISR_PROTFAULT.
368 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
369 * the special !is_write in the below conditional.
371 * For platforms that doesn't supports coherent icache and do support
372 * per page noexec bit, we do setup things such that we do the
373 * sync between D/I cache via fault. But that is handled via low level
374 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
377 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
378 * check should handle those and hence we should fall to the bad_area
379 * handling correctly.
381 * For embedded with per page exec support that doesn't support coherent
382 * icache we do get PROTFAULT and we handle that D/I cache sync in
383 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
384 * is conditional for server MMU.
386 * For radix, we can get prot fault for autonuma case, because radix
387 * page table will have them marked noaccess for user.
389 if (radix_enabled() || is_write)
392 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
395 static void sanity_check_fault(bool is_write, bool is_user,
396 unsigned long error_code, unsigned long address) { }
397 #endif /* CONFIG_PPC_BOOK3S */
400 * Define the correct "is_write" bit in error_code based
401 * on the processor family
403 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
404 #define page_fault_is_write(__err) ((__err) & ESR_DST)
405 #define page_fault_is_bad(__err) (0)
407 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
408 #if defined(CONFIG_PPC_8xx)
409 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
410 #elif defined(CONFIG_PPC64)
411 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
413 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
418 * For 600- and 800-family processors, the error_code parameter is DSISR
419 * for a data fault, SRR1 for an instruction fault. For 400-family processors
420 * the error_code parameter is ESR for a data fault, 0 for an instruction
422 * For 64-bit processors, the error_code parameter is
423 * - DSISR for a non-SLB data access fault,
424 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
427 * The return value is 0 if the fault was handled, or the signal
428 * number if this is a kernel fault that can't be handled here.
430 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
431 unsigned long error_code)
433 struct vm_area_struct * vma;
434 struct mm_struct *mm = current->mm;
435 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
436 int is_exec = TRAP(regs) == 0x400;
437 int is_user = user_mode(regs);
438 int is_write = page_fault_is_write(error_code);
439 vm_fault_t fault, major = 0;
440 bool must_retry = false;
442 if (notify_page_fault(regs))
445 if (unlikely(page_fault_is_bad(error_code))) {
447 _exception(SIGBUS, regs, BUS_OBJERR, address);
453 /* Additional sanity check(s) */
454 sanity_check_fault(is_write, is_user, error_code, address);
457 * The kernel should never take an execute fault nor should it
458 * take a page fault to a kernel address.
460 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
464 * If we're in an interrupt, have no user context or are running
465 * in a region with pagefaults disabled then we must not take the fault
467 if (unlikely(faulthandler_disabled() || !mm)) {
469 printk_ratelimited(KERN_ERR "Page fault in user mode"
470 " with faulthandler_disabled()=%d"
472 faulthandler_disabled(), mm);
473 return bad_area_nosemaphore(regs, address);
476 /* We restore the interrupt state now */
477 if (!arch_irq_disabled_regs(regs))
480 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
482 if (error_code & DSISR_KEYFAULT)
483 return bad_key_fault_exception(regs, address,
484 get_mm_addr_key(mm, address));
487 * We want to do this outside mmap_sem, because reading code around nip
488 * can result in fault, which will cause a deadlock when called with
492 flags |= FAULT_FLAG_USER;
494 flags |= FAULT_FLAG_WRITE;
496 flags |= FAULT_FLAG_INSTRUCTION;
498 /* When running in the kernel we expect faults to occur only to
499 * addresses in user space. All other faults represent errors in the
500 * kernel and should generate an OOPS. Unfortunately, in the case of an
501 * erroneous fault occurring in a code path which already holds mmap_sem
502 * we will deadlock attempting to validate the fault against the
503 * address space. Luckily the kernel only validly references user
504 * space from well defined areas of code, which are listed in the
507 * As the vast majority of faults will be valid we will only perform
508 * the source reference check when there is a possibility of a deadlock.
509 * Attempt to lock the address space, if we cannot we then validate the
510 * source. If this is invalid we can skip the address space check,
511 * thus avoiding the deadlock.
513 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
514 if (!is_user && !search_exception_tables(regs->nip))
515 return bad_area_nosemaphore(regs, address);
518 down_read(&mm->mmap_sem);
521 * The above down_read_trylock() might have succeeded in
522 * which case we'll have missed the might_sleep() from
528 vma = find_vma(mm, address);
530 return bad_area(regs, address);
531 if (likely(vma->vm_start <= address))
533 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
534 return bad_area(regs, address);
536 /* The stack is being expanded, check if it's valid */
537 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
540 return bad_area(regs, address);
542 up_read(&mm->mmap_sem);
543 if (fault_in_pages_readable((const char __user *)regs->nip,
544 sizeof(unsigned int)))
545 return bad_area_nosemaphore(regs, address);
549 /* Try to expand it */
550 if (unlikely(expand_stack(vma, address)))
551 return bad_area(regs, address);
554 if (unlikely(access_error(is_write, is_exec, vma)))
555 return bad_access(regs, address);
558 * If for any reason at all we couldn't handle the fault,
559 * make sure we exit gracefully rather than endlessly redo
562 fault = handle_mm_fault(vma, address, flags);
564 #ifdef CONFIG_PPC_MEM_KEYS
566 * we skipped checking for access error due to key earlier.
567 * Check that using handle_mm_fault error return.
569 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
570 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
572 int pkey = vma_pkey(vma);
574 up_read(&mm->mmap_sem);
575 return bad_key_fault_exception(regs, address, pkey);
577 #endif /* CONFIG_PPC_MEM_KEYS */
579 major |= fault & VM_FAULT_MAJOR;
582 * Handle the retry right now, the mmap_sem has been released in that
585 if (unlikely(fault & VM_FAULT_RETRY)) {
586 /* We retry only once */
587 if (flags & FAULT_FLAG_ALLOW_RETRY) {
589 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
592 flags &= ~FAULT_FLAG_ALLOW_RETRY;
593 flags |= FAULT_FLAG_TRIED;
594 if (!fatal_signal_pending(current))
599 * User mode? Just return to handle the fatal exception otherwise
600 * return to bad_page_fault
602 return is_user ? 0 : SIGBUS;
605 up_read(¤t->mm->mmap_sem);
607 if (unlikely(fault & VM_FAULT_ERROR))
608 return mm_fault_error(regs, address, fault);
611 * Major/minor page fault accounting.
615 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
616 cmo_account_page_fault();
619 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
623 NOKPROBE_SYMBOL(__do_page_fault);
625 int do_page_fault(struct pt_regs *regs, unsigned long address,
626 unsigned long error_code)
628 enum ctx_state prev_state = exception_enter();
629 int rc = __do_page_fault(regs, address, error_code);
630 exception_exit(prev_state);
633 NOKPROBE_SYMBOL(do_page_fault);
636 * bad_page_fault is called when we have a bad access from the kernel.
637 * It is called from the DSI and ISI handlers in head.S and from some
638 * of the procedures in traps.c.
640 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
642 const struct exception_table_entry *entry;
644 /* Are we prepared to handle this fault? */
645 if ((entry = search_exception_tables(regs->nip)) != NULL) {
646 regs->nip = extable_fixup(entry);
650 /* kernel has accessed a bad area */
652 switch (TRAP(regs)) {
656 pr_alert("BUG: %s at 0x%08lx\n",
657 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
658 "Unable to handle kernel data access", regs->dar);
662 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
663 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
666 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
670 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
674 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
677 if (task_stack_end_corrupted(current))
678 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
680 die("Kernel access of bad area", regs, sig);