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>
46 * Check whether the instruction inst is a store using
47 * an update addressing form which will update r1.
49 static bool store_updates_sp(unsigned int inst)
51 /* check for 1 in the rA field */
52 if (((inst >> 16) & 0x1f) != 1)
54 /* check major opcode */
62 case OP_STD: /* std or stdu */
63 return (inst & 3) == 1;
65 /* check minor opcode */
66 switch ((inst >> 1) & 0x3ff) {
71 case OP_31_XOP_STFSUX:
72 case OP_31_XOP_STFDUX:
79 * do_page_fault error handling helpers
83 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
86 * If we are in kernel mode, bail out with a SEGV, this will
87 * be caught by the assembly which will restore the non-volatile
88 * registers before calling bad_page_fault()
93 _exception(SIGSEGV, regs, si_code, address);
98 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
100 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
103 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
105 struct mm_struct *mm = current->mm;
108 * Something tried to access memory that isn't in our memory map..
109 * Fix it, but check if it's kernel or user first..
111 up_read(&mm->mmap_sem);
113 return __bad_area_nosemaphore(regs, address, si_code);
116 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
118 return __bad_area(regs, address, SEGV_MAPERR);
121 static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address,
125 * If we are in kernel mode, bail out with a SEGV, this will
126 * be caught by the assembly which will restore the non-volatile
127 * registers before calling bad_page_fault()
129 if (!user_mode(regs))
132 _exception_pkey(regs, address, pkey);
137 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
139 return __bad_area(regs, address, SEGV_ACCERR);
142 static int do_sigbus(struct pt_regs *regs, unsigned long address,
145 if (!user_mode(regs))
148 current->thread.trap_nr = BUS_ADRERR;
149 #ifdef CONFIG_MEMORY_FAILURE
150 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
151 unsigned int lsb = 0; /* shutup gcc */
153 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
154 current->comm, current->pid, address);
156 if (fault & VM_FAULT_HWPOISON_LARGE)
157 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
158 if (fault & VM_FAULT_HWPOISON)
161 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
166 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
170 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
174 * Kernel page fault interrupted by SIGKILL. We have no reason to
175 * continue processing.
177 if (fatal_signal_pending(current) && !user_mode(regs))
181 if (fault & VM_FAULT_OOM) {
183 * We ran out of memory, or some other thing happened to us that
184 * made us unable to handle the page fault gracefully.
186 if (!user_mode(regs))
188 pagefault_out_of_memory();
190 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
191 VM_FAULT_HWPOISON_LARGE))
192 return do_sigbus(regs, addr, fault);
193 else if (fault & VM_FAULT_SIGSEGV)
194 return bad_area_nosemaphore(regs, addr);
201 /* Is this a bad kernel fault ? */
202 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
203 unsigned long address, bool is_write)
205 int is_exec = TRAP(regs) == 0x400;
207 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
208 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
210 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
211 address >= TASK_SIZE ? "exec-protected" : "user",
213 from_kuid(&init_user_ns, current_uid()));
215 // Kernel exec fault is always bad
219 if (!is_exec && address < TASK_SIZE && (error_code & DSISR_PROTFAULT) &&
220 !search_exception_tables(regs->nip)) {
221 pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
223 from_kuid(&init_user_ns, current_uid()));
226 // Kernel fault on kernel address is bad
227 if (address >= TASK_SIZE)
230 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
231 if (!search_exception_tables(regs->nip))
234 // Read/write fault in a valid region (the exception table search passed
235 // above), but blocked by KUAP is bad, it can never succeed.
236 if (bad_kuap_fault(regs, address, is_write))
239 // What's left? Kernel fault on user in well defined regions (extable
240 // matched), and allowed by KUAP in the faulting context.
244 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
245 struct vm_area_struct *vma, unsigned int flags,
249 * N.B. The POWER/Open ABI allows programs to access up to
250 * 288 bytes below the stack pointer.
251 * The kernel signal delivery code writes up to about 1.5kB
252 * below the stack pointer (r1) before decrementing it.
253 * The exec code can write slightly over 640kB to the stack
254 * before setting the user r1. Thus we allow the stack to
255 * expand to 1MB without further checks.
257 if (address + 0x100000 < vma->vm_end) {
258 unsigned int __user *nip = (unsigned int __user *)regs->nip;
259 /* get user regs even if this fault is in kernel mode */
260 struct pt_regs *uregs = current->thread.regs;
265 * A user-mode access to an address a long way below
266 * the stack pointer is only valid if the instruction
267 * is one which would update the stack pointer to the
268 * address accessed if the instruction completed,
269 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
270 * (or the byte, halfword, float or double forms).
272 * If we don't check this then any write to the area
273 * between the last mapped region and the stack will
274 * expand the stack rather than segfaulting.
276 if (address + 2048 >= uregs->gpr[1])
279 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
280 access_ok(nip, sizeof(*nip))) {
285 res = __get_user_inatomic(inst, nip);
288 return !store_updates_sp(inst);
296 static bool access_error(bool is_write, bool is_exec,
297 struct vm_area_struct *vma)
300 * Allow execution from readable areas if the MMU does not
301 * provide separate controls over reading and executing.
303 * Note: That code used to not be enabled for 4xx/BookE.
304 * It is now as I/D cache coherency for these is done at
305 * set_pte_at() time and I see no reason why the test
306 * below wouldn't be valid on those processors. This -may-
307 * break programs compiled with a really old ABI though.
310 return !(vma->vm_flags & VM_EXEC) &&
311 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
312 !(vma->vm_flags & (VM_READ | VM_WRITE)));
316 if (unlikely(!(vma->vm_flags & VM_WRITE)))
321 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
324 * We should ideally do the vma pkey access check here. But in the
325 * fault path, handle_mm_fault() also does the same check. To avoid
326 * these multiple checks, we skip it here and handle access error due
332 #ifdef CONFIG_PPC_SMLPAR
333 static inline void cmo_account_page_fault(void)
335 if (firmware_has_feature(FW_FEATURE_CMO)) {
339 page_ins = be32_to_cpu(get_lppaca()->page_ins);
340 page_ins += 1 << PAGE_FACTOR;
341 get_lppaca()->page_ins = cpu_to_be32(page_ins);
346 static inline void cmo_account_page_fault(void) { }
347 #endif /* CONFIG_PPC_SMLPAR */
349 #ifdef CONFIG_PPC_BOOK3S
350 static void sanity_check_fault(bool is_write, bool is_user,
351 unsigned long error_code, unsigned long address)
354 * Userspace trying to access kernel address, we get PROTFAULT for that.
356 if (is_user && address >= TASK_SIZE) {
357 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
358 current->comm, current->pid, address,
359 from_kuid(&init_user_ns, current_uid()));
364 * For hash translation mode, we should never get a
365 * PROTFAULT. Any update to pte to reduce access will result in us
366 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
367 * fault instead of DSISR_PROTFAULT.
369 * A pte update to relax the access will not result in a hash page table
370 * entry invalidate and hence can result in DSISR_PROTFAULT.
371 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
372 * the special !is_write in the below conditional.
374 * For platforms that doesn't supports coherent icache and do support
375 * per page noexec bit, we do setup things such that we do the
376 * sync between D/I cache via fault. But that is handled via low level
377 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
380 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
381 * check should handle those and hence we should fall to the bad_area
382 * handling correctly.
384 * For embedded with per page exec support that doesn't support coherent
385 * icache we do get PROTFAULT and we handle that D/I cache sync in
386 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
387 * is conditional for server MMU.
389 * For radix, we can get prot fault for autonuma case, because radix
390 * page table will have them marked noaccess for user.
392 if (radix_enabled() || is_write)
395 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
398 static void sanity_check_fault(bool is_write, bool is_user,
399 unsigned long error_code, unsigned long address) { }
400 #endif /* CONFIG_PPC_BOOK3S */
403 * Define the correct "is_write" bit in error_code based
404 * on the processor family
406 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
407 #define page_fault_is_write(__err) ((__err) & ESR_DST)
408 #define page_fault_is_bad(__err) (0)
410 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
411 #if defined(CONFIG_PPC_8xx)
412 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
413 #elif defined(CONFIG_PPC64)
414 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
416 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
421 * For 600- and 800-family processors, the error_code parameter is DSISR
422 * for a data fault, SRR1 for an instruction fault. For 400-family processors
423 * the error_code parameter is ESR for a data fault, 0 for an instruction
425 * For 64-bit processors, the error_code parameter is
426 * - DSISR for a non-SLB data access fault,
427 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
430 * The return value is 0 if the fault was handled, or the signal
431 * number if this is a kernel fault that can't be handled here.
433 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
434 unsigned long error_code)
436 struct vm_area_struct * vma;
437 struct mm_struct *mm = current->mm;
438 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
439 int is_exec = TRAP(regs) == 0x400;
440 int is_user = user_mode(regs);
441 int is_write = page_fault_is_write(error_code);
442 vm_fault_t fault, major = 0;
443 bool must_retry = false;
444 bool kprobe_fault = kprobe_page_fault(regs, 11);
446 if (unlikely(debugger_fault_handler(regs) || kprobe_fault))
449 if (unlikely(page_fault_is_bad(error_code))) {
451 _exception(SIGBUS, regs, BUS_OBJERR, address);
457 /* Additional sanity check(s) */
458 sanity_check_fault(is_write, is_user, error_code, address);
461 * The kernel should never take an execute fault nor should it
462 * take a page fault to a kernel address or a page fault to a user
463 * address outside of dedicated places
465 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write)))
469 * If we're in an interrupt, have no user context or are running
470 * in a region with pagefaults disabled then we must not take the fault
472 if (unlikely(faulthandler_disabled() || !mm)) {
474 printk_ratelimited(KERN_ERR "Page fault in user mode"
475 " with faulthandler_disabled()=%d"
477 faulthandler_disabled(), mm);
478 return bad_area_nosemaphore(regs, address);
481 /* We restore the interrupt state now */
482 if (!arch_irq_disabled_regs(regs))
485 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
487 if (error_code & DSISR_KEYFAULT)
488 return bad_key_fault_exception(regs, address,
489 get_mm_addr_key(mm, address));
492 * We want to do this outside mmap_sem, because reading code around nip
493 * can result in fault, which will cause a deadlock when called with
497 flags |= FAULT_FLAG_USER;
499 flags |= FAULT_FLAG_WRITE;
501 flags |= FAULT_FLAG_INSTRUCTION;
503 /* When running in the kernel we expect faults to occur only to
504 * addresses in user space. All other faults represent errors in the
505 * kernel and should generate an OOPS. Unfortunately, in the case of an
506 * erroneous fault occurring in a code path which already holds mmap_sem
507 * we will deadlock attempting to validate the fault against the
508 * address space. Luckily the kernel only validly references user
509 * space from well defined areas of code, which are listed in the
512 * As the vast majority of faults will be valid we will only perform
513 * the source reference check when there is a possibility of a deadlock.
514 * Attempt to lock the address space, if we cannot we then validate the
515 * source. If this is invalid we can skip the address space check,
516 * thus avoiding the deadlock.
518 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
519 if (!is_user && !search_exception_tables(regs->nip))
520 return bad_area_nosemaphore(regs, address);
523 down_read(&mm->mmap_sem);
526 * The above down_read_trylock() might have succeeded in
527 * which case we'll have missed the might_sleep() from
533 vma = find_vma(mm, address);
535 return bad_area(regs, address);
536 if (likely(vma->vm_start <= address))
538 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
539 return bad_area(regs, address);
541 /* The stack is being expanded, check if it's valid */
542 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
545 return bad_area(regs, address);
547 up_read(&mm->mmap_sem);
548 if (fault_in_pages_readable((const char __user *)regs->nip,
549 sizeof(unsigned int)))
550 return bad_area_nosemaphore(regs, address);
554 /* Try to expand it */
555 if (unlikely(expand_stack(vma, address)))
556 return bad_area(regs, address);
559 if (unlikely(access_error(is_write, is_exec, vma)))
560 return bad_access(regs, address);
563 * If for any reason at all we couldn't handle the fault,
564 * make sure we exit gracefully rather than endlessly redo
567 fault = handle_mm_fault(vma, address, flags);
569 #ifdef CONFIG_PPC_MEM_KEYS
571 * we skipped checking for access error due to key earlier.
572 * Check that using handle_mm_fault error return.
574 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
575 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
577 int pkey = vma_pkey(vma);
579 up_read(&mm->mmap_sem);
580 return bad_key_fault_exception(regs, address, pkey);
582 #endif /* CONFIG_PPC_MEM_KEYS */
584 major |= fault & VM_FAULT_MAJOR;
587 * Handle the retry right now, the mmap_sem has been released in that
590 if (unlikely(fault & VM_FAULT_RETRY)) {
591 /* We retry only once */
592 if (flags & FAULT_FLAG_ALLOW_RETRY) {
594 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
597 flags &= ~FAULT_FLAG_ALLOW_RETRY;
598 flags |= FAULT_FLAG_TRIED;
599 if (!fatal_signal_pending(current))
604 * User mode? Just return to handle the fatal exception otherwise
605 * return to bad_page_fault
607 return is_user ? 0 : SIGBUS;
610 up_read(¤t->mm->mmap_sem);
612 if (unlikely(fault & VM_FAULT_ERROR))
613 return mm_fault_error(regs, address, fault);
616 * Major/minor page fault accounting.
620 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
621 cmo_account_page_fault();
624 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
628 NOKPROBE_SYMBOL(__do_page_fault);
630 int do_page_fault(struct pt_regs *regs, unsigned long address,
631 unsigned long error_code)
633 enum ctx_state prev_state = exception_enter();
634 int rc = __do_page_fault(regs, address, error_code);
635 exception_exit(prev_state);
638 NOKPROBE_SYMBOL(do_page_fault);
641 * bad_page_fault is called when we have a bad access from the kernel.
642 * It is called from the DSI and ISI handlers in head.S and from some
643 * of the procedures in traps.c.
645 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
647 const struct exception_table_entry *entry;
648 int is_write = page_fault_is_write(regs->dsisr);
650 /* Are we prepared to handle this fault? */
651 if ((entry = search_exception_tables(regs->nip)) != NULL) {
652 regs->nip = extable_fixup(entry);
656 /* kernel has accessed a bad area */
658 switch (TRAP(regs)) {
662 pr_alert("BUG: %s on %s at 0x%08lx\n",
663 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
664 "Unable to handle kernel data access",
665 is_write ? "write" : "read", regs->dar);
669 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
670 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
673 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
677 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
681 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
684 if (task_stack_end_corrupted(current))
685 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
687 die("Kernel access of bad area", regs, sig);