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>
38 #include <asm/pgtable.h>
40 #include <asm/mmu_context.h>
41 #include <asm/siginfo.h>
42 #include <asm/debug.h>
45 static inline bool notify_page_fault(struct pt_regs *regs)
50 /* kprobe_running() needs smp_processor_id() */
51 if (!user_mode(regs)) {
53 if (kprobe_running() && kprobe_fault_handler(regs, 11))
57 #endif /* CONFIG_KPROBES */
59 if (unlikely(debugger_fault_handler(regs)))
66 * Check whether the instruction inst is a store using
67 * an update addressing form which will update r1.
69 static bool store_updates_sp(unsigned int inst)
71 /* check for 1 in the rA field */
72 if (((inst >> 16) & 0x1f) != 1)
74 /* check major opcode */
82 case OP_STD: /* std or stdu */
83 return (inst & 3) == 1;
85 /* check minor opcode */
86 switch ((inst >> 1) & 0x3ff) {
91 case OP_31_XOP_STFSUX:
92 case OP_31_XOP_STFDUX:
99 * do_page_fault error handling helpers
103 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
106 * If we are in kernel mode, bail out with a SEGV, this will
107 * be caught by the assembly which will restore the non-volatile
108 * registers before calling bad_page_fault()
110 if (!user_mode(regs))
113 _exception(SIGSEGV, regs, si_code, address);
118 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
120 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
123 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
125 struct mm_struct *mm = current->mm;
128 * Something tried to access memory that isn't in our memory map..
129 * Fix it, but check if it's kernel or user first..
131 up_read(&mm->mmap_sem);
133 return __bad_area_nosemaphore(regs, address, si_code);
136 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
138 return __bad_area(regs, address, SEGV_MAPERR);
141 static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address,
145 * If we are in kernel mode, bail out with a SEGV, this will
146 * be caught by the assembly which will restore the non-volatile
147 * registers before calling bad_page_fault()
149 if (!user_mode(regs))
152 _exception_pkey(regs, address, pkey);
157 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
159 return __bad_area(regs, address, SEGV_ACCERR);
162 static int do_sigbus(struct pt_regs *regs, unsigned long address,
165 if (!user_mode(regs))
168 current->thread.trap_nr = BUS_ADRERR;
169 #ifdef CONFIG_MEMORY_FAILURE
170 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
171 unsigned int lsb = 0; /* shutup gcc */
173 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
174 current->comm, current->pid, address);
176 if (fault & VM_FAULT_HWPOISON_LARGE)
177 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
178 if (fault & VM_FAULT_HWPOISON)
181 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb,
187 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, current);
191 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
195 * Kernel page fault interrupted by SIGKILL. We have no reason to
196 * continue processing.
198 if (fatal_signal_pending(current) && !user_mode(regs))
202 if (fault & VM_FAULT_OOM) {
204 * We ran out of memory, or some other thing happened to us that
205 * made us unable to handle the page fault gracefully.
207 if (!user_mode(regs))
209 pagefault_out_of_memory();
211 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
212 VM_FAULT_HWPOISON_LARGE))
213 return do_sigbus(regs, addr, fault);
214 else if (fault & VM_FAULT_SIGSEGV)
215 return bad_area_nosemaphore(regs, addr);
222 /* Is this a bad kernel fault ? */
223 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
224 unsigned long address, bool is_write)
226 int is_exec = TRAP(regs) == 0x400;
228 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
229 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
231 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
232 address >= TASK_SIZE ? "exec-protected" : "user",
234 from_kuid(&init_user_ns, current_uid()));
236 // Kernel exec fault is always bad
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 // Kernel fault on kernel address is bad
248 if (address >= TASK_SIZE)
251 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
252 if (!search_exception_tables(regs->nip))
255 // Read/write fault in a valid region (the exception table search passed
256 // above), but blocked by KUAP is bad, it can never succeed.
257 if (bad_kuap_fault(regs, is_write))
260 // What's left? Kernel fault on user in well defined regions (extable
261 // matched), and allowed by KUAP in the faulting context.
265 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
266 struct vm_area_struct *vma, unsigned int flags,
270 * N.B. The POWER/Open ABI allows programs to access up to
271 * 288 bytes below the stack pointer.
272 * The kernel signal delivery code writes up to about 1.5kB
273 * below the stack pointer (r1) before decrementing it.
274 * The exec code can write slightly over 640kB to the stack
275 * before setting the user r1. Thus we allow the stack to
276 * expand to 1MB without further checks.
278 if (address + 0x100000 < vma->vm_end) {
279 unsigned int __user *nip = (unsigned int __user *)regs->nip;
280 /* get user regs even if this fault is in kernel mode */
281 struct pt_regs *uregs = current->thread.regs;
286 * A user-mode access to an address a long way below
287 * the stack pointer is only valid if the instruction
288 * is one which would update the stack pointer to the
289 * address accessed if the instruction completed,
290 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
291 * (or the byte, halfword, float or double forms).
293 * If we don't check this then any write to the area
294 * between the last mapped region and the stack will
295 * expand the stack rather than segfaulting.
297 if (address + 2048 >= uregs->gpr[1])
300 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
301 access_ok(nip, sizeof(*nip))) {
306 res = __get_user_inatomic(inst, nip);
309 return !store_updates_sp(inst);
317 static bool access_error(bool is_write, bool is_exec,
318 struct vm_area_struct *vma)
321 * Allow execution from readable areas if the MMU does not
322 * provide separate controls over reading and executing.
324 * Note: That code used to not be enabled for 4xx/BookE.
325 * It is now as I/D cache coherency for these is done at
326 * set_pte_at() time and I see no reason why the test
327 * below wouldn't be valid on those processors. This -may-
328 * break programs compiled with a really old ABI though.
331 return !(vma->vm_flags & VM_EXEC) &&
332 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
333 !(vma->vm_flags & (VM_READ | VM_WRITE)));
337 if (unlikely(!(vma->vm_flags & VM_WRITE)))
342 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
345 * We should ideally do the vma pkey access check here. But in the
346 * fault path, handle_mm_fault() also does the same check. To avoid
347 * these multiple checks, we skip it here and handle access error due
353 #ifdef CONFIG_PPC_SMLPAR
354 static inline void cmo_account_page_fault(void)
356 if (firmware_has_feature(FW_FEATURE_CMO)) {
360 page_ins = be32_to_cpu(get_lppaca()->page_ins);
361 page_ins += 1 << PAGE_FACTOR;
362 get_lppaca()->page_ins = cpu_to_be32(page_ins);
367 static inline void cmo_account_page_fault(void) { }
368 #endif /* CONFIG_PPC_SMLPAR */
370 #ifdef CONFIG_PPC_BOOK3S
371 static void sanity_check_fault(bool is_write, bool is_user,
372 unsigned long error_code, unsigned long address)
375 * Userspace trying to access kernel address, we get PROTFAULT for that.
377 if (is_user && address >= TASK_SIZE) {
378 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
379 current->comm, current->pid, address,
380 from_kuid(&init_user_ns, current_uid()));
385 * For hash translation mode, we should never get a
386 * PROTFAULT. Any update to pte to reduce access will result in us
387 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
388 * fault instead of DSISR_PROTFAULT.
390 * A pte update to relax the access will not result in a hash page table
391 * entry invalidate and hence can result in DSISR_PROTFAULT.
392 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
393 * the special !is_write in the below conditional.
395 * For platforms that doesn't supports coherent icache and do support
396 * per page noexec bit, we do setup things such that we do the
397 * sync between D/I cache via fault. But that is handled via low level
398 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
401 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
402 * check should handle those and hence we should fall to the bad_area
403 * handling correctly.
405 * For embedded with per page exec support that doesn't support coherent
406 * icache we do get PROTFAULT and we handle that D/I cache sync in
407 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
408 * is conditional for server MMU.
410 * For radix, we can get prot fault for autonuma case, because radix
411 * page table will have them marked noaccess for user.
413 if (radix_enabled() || is_write)
416 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
419 static void sanity_check_fault(bool is_write, bool is_user,
420 unsigned long error_code, unsigned long address) { }
421 #endif /* CONFIG_PPC_BOOK3S */
424 * Define the correct "is_write" bit in error_code based
425 * on the processor family
427 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
428 #define page_fault_is_write(__err) ((__err) & ESR_DST)
429 #define page_fault_is_bad(__err) (0)
431 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
432 #if defined(CONFIG_PPC_8xx)
433 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
434 #elif defined(CONFIG_PPC64)
435 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
437 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
442 * For 600- and 800-family processors, the error_code parameter is DSISR
443 * for a data fault, SRR1 for an instruction fault. For 400-family processors
444 * the error_code parameter is ESR for a data fault, 0 for an instruction
446 * For 64-bit processors, the error_code parameter is
447 * - DSISR for a non-SLB data access fault,
448 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
451 * The return value is 0 if the fault was handled, or the signal
452 * number if this is a kernel fault that can't be handled here.
454 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
455 unsigned long error_code)
457 struct vm_area_struct * vma;
458 struct mm_struct *mm = current->mm;
459 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
460 int is_exec = TRAP(regs) == 0x400;
461 int is_user = user_mode(regs);
462 int is_write = page_fault_is_write(error_code);
463 vm_fault_t fault, major = 0;
464 bool must_retry = false;
466 if (notify_page_fault(regs))
469 if (unlikely(page_fault_is_bad(error_code))) {
471 _exception(SIGBUS, regs, BUS_OBJERR, address);
477 /* Additional sanity check(s) */
478 sanity_check_fault(is_write, is_user, error_code, address);
481 * The kernel should never take an execute fault nor should it
482 * take a page fault to a kernel address or a page fault to a user
483 * address outside of dedicated places
485 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write)))
489 * If we're in an interrupt, have no user context or are running
490 * in a region with pagefaults disabled then we must not take the fault
492 if (unlikely(faulthandler_disabled() || !mm)) {
494 printk_ratelimited(KERN_ERR "Page fault in user mode"
495 " with faulthandler_disabled()=%d"
497 faulthandler_disabled(), mm);
498 return bad_area_nosemaphore(regs, address);
501 /* We restore the interrupt state now */
502 if (!arch_irq_disabled_regs(regs))
505 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
507 if (error_code & DSISR_KEYFAULT)
508 return bad_key_fault_exception(regs, address,
509 get_mm_addr_key(mm, address));
512 * We want to do this outside mmap_sem, because reading code around nip
513 * can result in fault, which will cause a deadlock when called with
517 flags |= FAULT_FLAG_USER;
519 flags |= FAULT_FLAG_WRITE;
521 flags |= FAULT_FLAG_INSTRUCTION;
523 /* When running in the kernel we expect faults to occur only to
524 * addresses in user space. All other faults represent errors in the
525 * kernel and should generate an OOPS. Unfortunately, in the case of an
526 * erroneous fault occurring in a code path which already holds mmap_sem
527 * we will deadlock attempting to validate the fault against the
528 * address space. Luckily the kernel only validly references user
529 * space from well defined areas of code, which are listed in the
532 * As the vast majority of faults will be valid we will only perform
533 * the source reference check when there is a possibility of a deadlock.
534 * Attempt to lock the address space, if we cannot we then validate the
535 * source. If this is invalid we can skip the address space check,
536 * thus avoiding the deadlock.
538 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
539 if (!is_user && !search_exception_tables(regs->nip))
540 return bad_area_nosemaphore(regs, address);
543 down_read(&mm->mmap_sem);
546 * The above down_read_trylock() might have succeeded in
547 * which case we'll have missed the might_sleep() from
553 vma = find_vma(mm, address);
555 return bad_area(regs, address);
556 if (likely(vma->vm_start <= address))
558 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
559 return bad_area(regs, address);
561 /* The stack is being expanded, check if it's valid */
562 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
565 return bad_area(regs, address);
567 up_read(&mm->mmap_sem);
568 if (fault_in_pages_readable((const char __user *)regs->nip,
569 sizeof(unsigned int)))
570 return bad_area_nosemaphore(regs, address);
574 /* Try to expand it */
575 if (unlikely(expand_stack(vma, address)))
576 return bad_area(regs, address);
579 if (unlikely(access_error(is_write, is_exec, vma)))
580 return bad_access(regs, address);
583 * If for any reason at all we couldn't handle the fault,
584 * make sure we exit gracefully rather than endlessly redo
587 fault = handle_mm_fault(vma, address, flags);
589 #ifdef CONFIG_PPC_MEM_KEYS
591 * we skipped checking for access error due to key earlier.
592 * Check that using handle_mm_fault error return.
594 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
595 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
597 int pkey = vma_pkey(vma);
599 up_read(&mm->mmap_sem);
600 return bad_key_fault_exception(regs, address, pkey);
602 #endif /* CONFIG_PPC_MEM_KEYS */
604 major |= fault & VM_FAULT_MAJOR;
607 * Handle the retry right now, the mmap_sem has been released in that
610 if (unlikely(fault & VM_FAULT_RETRY)) {
611 /* We retry only once */
612 if (flags & FAULT_FLAG_ALLOW_RETRY) {
614 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
617 flags &= ~FAULT_FLAG_ALLOW_RETRY;
618 flags |= FAULT_FLAG_TRIED;
619 if (!fatal_signal_pending(current))
624 * User mode? Just return to handle the fatal exception otherwise
625 * return to bad_page_fault
627 return is_user ? 0 : SIGBUS;
630 up_read(¤t->mm->mmap_sem);
632 if (unlikely(fault & VM_FAULT_ERROR))
633 return mm_fault_error(regs, address, fault);
636 * Major/minor page fault accounting.
640 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
641 cmo_account_page_fault();
644 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
648 NOKPROBE_SYMBOL(__do_page_fault);
650 int do_page_fault(struct pt_regs *regs, unsigned long address,
651 unsigned long error_code)
653 enum ctx_state prev_state = exception_enter();
654 int rc = __do_page_fault(regs, address, error_code);
655 exception_exit(prev_state);
658 NOKPROBE_SYMBOL(do_page_fault);
661 * bad_page_fault is called when we have a bad access from the kernel.
662 * It is called from the DSI and ISI handlers in head.S and from some
663 * of the procedures in traps.c.
665 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
667 const struct exception_table_entry *entry;
669 /* Are we prepared to handle this fault? */
670 if ((entry = search_exception_tables(regs->nip)) != NULL) {
671 regs->nip = extable_fixup(entry);
675 /* kernel has accessed a bad area */
677 switch (TRAP(regs)) {
681 pr_alert("BUG: %s at 0x%08lx\n",
682 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
683 "Unable to handle kernel data access", 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);