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(struct pt_regs *regs, unsigned long error_code,
227 unsigned long address)
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()));
240 if (!is_exec && address < TASK_SIZE && (error_code & DSISR_PROTFAULT) &&
241 !search_exception_tables(regs->nip)) {
242 pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
244 from_kuid(&init_user_ns, current_uid()));
247 return is_exec || (address >= TASK_SIZE) || !search_exception_tables(regs->nip);
250 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
251 struct vm_area_struct *vma, unsigned int flags,
255 * N.B. The POWER/Open ABI allows programs to access up to
256 * 288 bytes below the stack pointer.
257 * The kernel signal delivery code writes up to about 1.5kB
258 * below the stack pointer (r1) before decrementing it.
259 * The exec code can write slightly over 640kB to the stack
260 * before setting the user r1. Thus we allow the stack to
261 * expand to 1MB without further checks.
263 if (address + 0x100000 < vma->vm_end) {
264 unsigned int __user *nip = (unsigned int __user *)regs->nip;
265 /* get user regs even if this fault is in kernel mode */
266 struct pt_regs *uregs = current->thread.regs;
271 * A user-mode access to an address a long way below
272 * the stack pointer is only valid if the instruction
273 * is one which would update the stack pointer to the
274 * address accessed if the instruction completed,
275 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
276 * (or the byte, halfword, float or double forms).
278 * If we don't check this then any write to the area
279 * between the last mapped region and the stack will
280 * expand the stack rather than segfaulting.
282 if (address + 2048 >= uregs->gpr[1])
285 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
286 access_ok(nip, sizeof(*nip))) {
291 res = __get_user_inatomic(inst, nip);
294 return !store_updates_sp(inst);
302 static bool access_error(bool is_write, bool is_exec,
303 struct vm_area_struct *vma)
306 * Allow execution from readable areas if the MMU does not
307 * provide separate controls over reading and executing.
309 * Note: That code used to not be enabled for 4xx/BookE.
310 * It is now as I/D cache coherency for these is done at
311 * set_pte_at() time and I see no reason why the test
312 * below wouldn't be valid on those processors. This -may-
313 * break programs compiled with a really old ABI though.
316 return !(vma->vm_flags & VM_EXEC) &&
317 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
318 !(vma->vm_flags & (VM_READ | VM_WRITE)));
322 if (unlikely(!(vma->vm_flags & VM_WRITE)))
327 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
330 * We should ideally do the vma pkey access check here. But in the
331 * fault path, handle_mm_fault() also does the same check. To avoid
332 * these multiple checks, we skip it here and handle access error due
338 #ifdef CONFIG_PPC_SMLPAR
339 static inline void cmo_account_page_fault(void)
341 if (firmware_has_feature(FW_FEATURE_CMO)) {
345 page_ins = be32_to_cpu(get_lppaca()->page_ins);
346 page_ins += 1 << PAGE_FACTOR;
347 get_lppaca()->page_ins = cpu_to_be32(page_ins);
352 static inline void cmo_account_page_fault(void) { }
353 #endif /* CONFIG_PPC_SMLPAR */
355 #ifdef CONFIG_PPC_BOOK3S
356 static void sanity_check_fault(bool is_write, bool is_user,
357 unsigned long error_code, unsigned long address)
360 * Userspace trying to access kernel address, we get PROTFAULT for that.
362 if (is_user && address >= TASK_SIZE) {
363 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
364 current->comm, current->pid, address,
365 from_kuid(&init_user_ns, current_uid()));
370 * For hash translation mode, we should never get a
371 * PROTFAULT. Any update to pte to reduce access will result in us
372 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
373 * fault instead of DSISR_PROTFAULT.
375 * A pte update to relax the access will not result in a hash page table
376 * entry invalidate and hence can result in DSISR_PROTFAULT.
377 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
378 * the special !is_write in the below conditional.
380 * For platforms that doesn't supports coherent icache and do support
381 * per page noexec bit, we do setup things such that we do the
382 * sync between D/I cache via fault. But that is handled via low level
383 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
386 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
387 * check should handle those and hence we should fall to the bad_area
388 * handling correctly.
390 * For embedded with per page exec support that doesn't support coherent
391 * icache we do get PROTFAULT and we handle that D/I cache sync in
392 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
393 * is conditional for server MMU.
395 * For radix, we can get prot fault for autonuma case, because radix
396 * page table will have them marked noaccess for user.
398 if (radix_enabled() || is_write)
401 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
404 static void sanity_check_fault(bool is_write, bool is_user,
405 unsigned long error_code, unsigned long address) { }
406 #endif /* CONFIG_PPC_BOOK3S */
409 * Define the correct "is_write" bit in error_code based
410 * on the processor family
412 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
413 #define page_fault_is_write(__err) ((__err) & ESR_DST)
414 #define page_fault_is_bad(__err) (0)
416 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
417 #if defined(CONFIG_PPC_8xx)
418 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
419 #elif defined(CONFIG_PPC64)
420 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
422 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
427 * For 600- and 800-family processors, the error_code parameter is DSISR
428 * for a data fault, SRR1 for an instruction fault. For 400-family processors
429 * the error_code parameter is ESR for a data fault, 0 for an instruction
431 * For 64-bit processors, the error_code parameter is
432 * - DSISR for a non-SLB data access fault,
433 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
436 * The return value is 0 if the fault was handled, or the signal
437 * number if this is a kernel fault that can't be handled here.
439 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
440 unsigned long error_code)
442 struct vm_area_struct * vma;
443 struct mm_struct *mm = current->mm;
444 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
445 int is_exec = TRAP(regs) == 0x400;
446 int is_user = user_mode(regs);
447 int is_write = page_fault_is_write(error_code);
448 vm_fault_t fault, major = 0;
449 bool must_retry = false;
451 if (notify_page_fault(regs))
454 if (unlikely(page_fault_is_bad(error_code))) {
456 _exception(SIGBUS, regs, BUS_OBJERR, address);
462 /* Additional sanity check(s) */
463 sanity_check_fault(is_write, is_user, error_code, address);
466 * The kernel should never take an execute fault nor should it
467 * take a page fault to a kernel address or a page fault to a user
468 * address outside of dedicated places
470 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address)))
474 * If we're in an interrupt, have no user context or are running
475 * in a region with pagefaults disabled then we must not take the fault
477 if (unlikely(faulthandler_disabled() || !mm)) {
479 printk_ratelimited(KERN_ERR "Page fault in user mode"
480 " with faulthandler_disabled()=%d"
482 faulthandler_disabled(), mm);
483 return bad_area_nosemaphore(regs, address);
486 /* We restore the interrupt state now */
487 if (!arch_irq_disabled_regs(regs))
490 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
492 if (error_code & DSISR_KEYFAULT)
493 return bad_key_fault_exception(regs, address,
494 get_mm_addr_key(mm, address));
497 * We want to do this outside mmap_sem, because reading code around nip
498 * can result in fault, which will cause a deadlock when called with
502 flags |= FAULT_FLAG_USER;
504 flags |= FAULT_FLAG_WRITE;
506 flags |= FAULT_FLAG_INSTRUCTION;
508 /* When running in the kernel we expect faults to occur only to
509 * addresses in user space. All other faults represent errors in the
510 * kernel and should generate an OOPS. Unfortunately, in the case of an
511 * erroneous fault occurring in a code path which already holds mmap_sem
512 * we will deadlock attempting to validate the fault against the
513 * address space. Luckily the kernel only validly references user
514 * space from well defined areas of code, which are listed in the
517 * As the vast majority of faults will be valid we will only perform
518 * the source reference check when there is a possibility of a deadlock.
519 * Attempt to lock the address space, if we cannot we then validate the
520 * source. If this is invalid we can skip the address space check,
521 * thus avoiding the deadlock.
523 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
524 if (!is_user && !search_exception_tables(regs->nip))
525 return bad_area_nosemaphore(regs, address);
528 down_read(&mm->mmap_sem);
531 * The above down_read_trylock() might have succeeded in
532 * which case we'll have missed the might_sleep() from
538 vma = find_vma(mm, address);
540 return bad_area(regs, address);
541 if (likely(vma->vm_start <= address))
543 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
544 return bad_area(regs, address);
546 /* The stack is being expanded, check if it's valid */
547 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
550 return bad_area(regs, address);
552 up_read(&mm->mmap_sem);
553 if (fault_in_pages_readable((const char __user *)regs->nip,
554 sizeof(unsigned int)))
555 return bad_area_nosemaphore(regs, address);
559 /* Try to expand it */
560 if (unlikely(expand_stack(vma, address)))
561 return bad_area(regs, address);
564 if (unlikely(access_error(is_write, is_exec, vma)))
565 return bad_access(regs, address);
568 * If for any reason at all we couldn't handle the fault,
569 * make sure we exit gracefully rather than endlessly redo
572 fault = handle_mm_fault(vma, address, flags);
574 #ifdef CONFIG_PPC_MEM_KEYS
576 * we skipped checking for access error due to key earlier.
577 * Check that using handle_mm_fault error return.
579 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
580 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
582 int pkey = vma_pkey(vma);
584 up_read(&mm->mmap_sem);
585 return bad_key_fault_exception(regs, address, pkey);
587 #endif /* CONFIG_PPC_MEM_KEYS */
589 major |= fault & VM_FAULT_MAJOR;
592 * Handle the retry right now, the mmap_sem has been released in that
595 if (unlikely(fault & VM_FAULT_RETRY)) {
596 /* We retry only once */
597 if (flags & FAULT_FLAG_ALLOW_RETRY) {
599 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
602 flags &= ~FAULT_FLAG_ALLOW_RETRY;
603 flags |= FAULT_FLAG_TRIED;
604 if (!fatal_signal_pending(current))
609 * User mode? Just return to handle the fatal exception otherwise
610 * return to bad_page_fault
612 return is_user ? 0 : SIGBUS;
615 up_read(¤t->mm->mmap_sem);
617 if (unlikely(fault & VM_FAULT_ERROR))
618 return mm_fault_error(regs, address, fault);
621 * Major/minor page fault accounting.
625 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
626 cmo_account_page_fault();
629 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
633 NOKPROBE_SYMBOL(__do_page_fault);
635 int do_page_fault(struct pt_regs *regs, unsigned long address,
636 unsigned long error_code)
638 enum ctx_state prev_state = exception_enter();
639 int rc = __do_page_fault(regs, address, error_code);
640 exception_exit(prev_state);
643 NOKPROBE_SYMBOL(do_page_fault);
646 * bad_page_fault is called when we have a bad access from the kernel.
647 * It is called from the DSI and ISI handlers in head.S and from some
648 * of the procedures in traps.c.
650 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
652 const struct exception_table_entry *entry;
654 /* Are we prepared to handle this fault? */
655 if ((entry = search_exception_tables(regs->nip)) != NULL) {
656 regs->nip = extable_fixup(entry);
660 /* kernel has accessed a bad area */
662 switch (TRAP(regs)) {
666 pr_alert("BUG: %s at 0x%08lx\n",
667 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
668 "Unable to handle kernel data access", regs->dar);
672 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
673 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
676 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
680 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
684 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
687 if (task_stack_end_corrupted(current))
688 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
690 die("Kernel access of bad area", regs, sig);