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/ptrace.h>
26 #include <linux/mman.h>
28 #include <linux/interrupt.h>
29 #include <linux/highmem.h>
30 #include <linux/extable.h>
31 #include <linux/kprobes.h>
32 #include <linux/kdebug.h>
33 #include <linux/perf_event.h>
34 #include <linux/ratelimit.h>
35 #include <linux/context_tracking.h>
36 #include <linux/hugetlb.h>
37 #include <linux/uaccess.h>
39 #include <asm/firmware.h>
41 #include <asm/pgtable.h>
43 #include <asm/mmu_context.h>
44 #include <asm/tlbflush.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 at regs->nip is a store using
70 * an update addressing form which will update r1.
72 static bool store_updates_sp(struct pt_regs *regs)
76 if (get_user(inst, (unsigned int __user *)regs->nip))
78 /* check for 1 in the rA field */
79 if (((inst >> 16) & 0x1f) != 1)
81 /* check major opcode */
89 case 62: /* std or stdu */
90 return (inst & 3) == 1;
92 /* check minor opcode */
93 switch ((inst >> 1) & 0x3ff) {
98 case 695: /* stfsux */
99 case 759: /* stfdux */
106 * do_page_fault error handling helpers
110 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code,
114 * If we are in kernel mode, bail out with a SEGV, this will
115 * be caught by the assembly which will restore the non-volatile
116 * registers before calling bad_page_fault()
118 if (!user_mode(regs))
121 _exception_pkey(SIGSEGV, regs, si_code, address, pkey);
126 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
128 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR, 0);
131 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code,
134 struct mm_struct *mm = current->mm;
137 * Something tried to access memory that isn't in our memory map..
138 * Fix it, but check if it's kernel or user first..
140 up_read(&mm->mmap_sem);
142 return __bad_area_nosemaphore(regs, address, si_code, pkey);
145 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
147 return __bad_area(regs, address, SEGV_MAPERR, 0);
150 static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address,
153 return __bad_area_nosemaphore(regs, address, SEGV_PKUERR, pkey);
156 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
158 return __bad_area(regs, address, SEGV_ACCERR, 0);
161 static int do_sigbus(struct pt_regs *regs, unsigned long address,
165 unsigned int lsb = 0;
167 if (!user_mode(regs))
170 current->thread.trap_nr = BUS_ADRERR;
171 info.si_signo = SIGBUS;
173 info.si_code = BUS_ADRERR;
174 info.si_addr = (void __user *)address;
175 #ifdef CONFIG_MEMORY_FAILURE
176 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
177 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
178 current->comm, current->pid, address);
179 info.si_code = BUS_MCEERR_AR;
182 if (fault & VM_FAULT_HWPOISON_LARGE)
183 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
184 if (fault & VM_FAULT_HWPOISON)
187 info.si_addr_lsb = lsb;
188 force_sig_info(SIGBUS, &info, current);
192 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
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(bool is_exec, unsigned long error_code,
224 unsigned long address)
226 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT))) {
227 printk_ratelimited(KERN_CRIT "kernel tried to execute"
228 " exec-protected page (%lx) -"
229 "exploit attempt? (uid: %d)\n",
230 address, from_kuid(&init_user_ns,
233 return is_exec || (address >= TASK_SIZE);
236 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
237 struct vm_area_struct *vma,
238 bool store_update_sp)
241 * N.B. The POWER/Open ABI allows programs to access up to
242 * 288 bytes below the stack pointer.
243 * The kernel signal delivery code writes up to about 1.5kB
244 * below the stack pointer (r1) before decrementing it.
245 * The exec code can write slightly over 640kB to the stack
246 * before setting the user r1. Thus we allow the stack to
247 * expand to 1MB without further checks.
249 if (address + 0x100000 < vma->vm_end) {
250 /* get user regs even if this fault is in kernel mode */
251 struct pt_regs *uregs = current->thread.regs;
256 * A user-mode access to an address a long way below
257 * the stack pointer is only valid if the instruction
258 * is one which would update the stack pointer to the
259 * address accessed if the instruction completed,
260 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
261 * (or the byte, halfword, float or double forms).
263 * If we don't check this then any write to the area
264 * between the last mapped region and the stack will
265 * expand the stack rather than segfaulting.
267 if (address + 2048 < uregs->gpr[1] && !store_update_sp)
273 static bool access_error(bool is_write, bool is_exec,
274 struct vm_area_struct *vma)
277 * Allow execution from readable areas if the MMU does not
278 * provide separate controls over reading and executing.
280 * Note: That code used to not be enabled for 4xx/BookE.
281 * It is now as I/D cache coherency for these is done at
282 * set_pte_at() time and I see no reason why the test
283 * below wouldn't be valid on those processors. This -may-
284 * break programs compiled with a really old ABI though.
287 return !(vma->vm_flags & VM_EXEC) &&
288 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
289 !(vma->vm_flags & (VM_READ | VM_WRITE)));
293 if (unlikely(!(vma->vm_flags & VM_WRITE)))
298 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
301 * We should ideally do the vma pkey access check here. But in the
302 * fault path, handle_mm_fault() also does the same check. To avoid
303 * these multiple checks, we skip it here and handle access error due
309 #ifdef CONFIG_PPC_SMLPAR
310 static inline void cmo_account_page_fault(void)
312 if (firmware_has_feature(FW_FEATURE_CMO)) {
316 page_ins = be32_to_cpu(get_lppaca()->page_ins);
317 page_ins += 1 << PAGE_FACTOR;
318 get_lppaca()->page_ins = cpu_to_be32(page_ins);
323 static inline void cmo_account_page_fault(void) { }
324 #endif /* CONFIG_PPC_SMLPAR */
326 #ifdef CONFIG_PPC_STD_MMU
327 static void sanity_check_fault(bool is_write, unsigned long error_code)
330 * For hash translation mode, we should never get a
331 * PROTFAULT. Any update to pte to reduce access will result in us
332 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
333 * fault instead of DSISR_PROTFAULT.
335 * A pte update to relax the access will not result in a hash page table
336 * entry invalidate and hence can result in DSISR_PROTFAULT.
337 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
338 * the special !is_write in the below conditional.
340 * For platforms that doesn't supports coherent icache and do support
341 * per page noexec bit, we do setup things such that we do the
342 * sync between D/I cache via fault. But that is handled via low level
343 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
346 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
347 * check should handle those and hence we should fall to the bad_area
348 * handling correctly.
350 * For embedded with per page exec support that doesn't support coherent
351 * icache we do get PROTFAULT and we handle that D/I cache sync in
352 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
353 * is conditional for server MMU.
355 * For radix, we can get prot fault for autonuma case, because radix
356 * page table will have them marked noaccess for user.
358 if (!radix_enabled() && !is_write)
359 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
362 static void sanity_check_fault(bool is_write, unsigned long error_code) { }
363 #endif /* CONFIG_PPC_STD_MMU */
366 * Define the correct "is_write" bit in error_code based
367 * on the processor family
369 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
370 #define page_fault_is_write(__err) ((__err) & ESR_DST)
371 #define page_fault_is_bad(__err) (0)
373 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
374 #if defined(CONFIG_PPC_8xx)
375 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
376 #elif defined(CONFIG_PPC64)
377 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
379 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
384 * For 600- and 800-family processors, the error_code parameter is DSISR
385 * for a data fault, SRR1 for an instruction fault. For 400-family processors
386 * the error_code parameter is ESR for a data fault, 0 for an instruction
388 * For 64-bit processors, the error_code parameter is
389 * - DSISR for a non-SLB data access fault,
390 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
393 * The return value is 0 if the fault was handled, or the signal
394 * number if this is a kernel fault that can't be handled here.
396 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
397 unsigned long error_code)
399 struct vm_area_struct * vma;
400 struct mm_struct *mm = current->mm;
401 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
402 int is_exec = TRAP(regs) == 0x400;
403 int is_user = user_mode(regs);
404 int is_write = page_fault_is_write(error_code);
405 int fault, major = 0;
406 bool store_update_sp = false;
408 if (notify_page_fault(regs))
411 if (unlikely(page_fault_is_bad(error_code))) {
413 _exception(SIGBUS, regs, BUS_OBJERR, address);
419 /* Additional sanity check(s) */
420 sanity_check_fault(is_write, error_code);
423 * The kernel should never take an execute fault nor should it
424 * take a page fault to a kernel address.
426 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
430 * If we're in an interrupt, have no user context or are running
431 * in a region with pagefaults disabled then we must not take the fault
433 if (unlikely(faulthandler_disabled() || !mm)) {
435 printk_ratelimited(KERN_ERR "Page fault in user mode"
436 " with faulthandler_disabled()=%d"
438 faulthandler_disabled(), mm);
439 return bad_area_nosemaphore(regs, address);
442 /* We restore the interrupt state now */
443 if (!arch_irq_disabled_regs(regs))
446 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
448 if (error_code & DSISR_KEYFAULT)
449 return bad_key_fault_exception(regs, address,
450 get_mm_addr_key(mm, address));
453 * We want to do this outside mmap_sem, because reading code around nip
454 * can result in fault, which will cause a deadlock when called with
457 if (is_write && is_user)
458 store_update_sp = store_updates_sp(regs);
461 flags |= FAULT_FLAG_USER;
463 flags |= FAULT_FLAG_WRITE;
465 flags |= FAULT_FLAG_INSTRUCTION;
467 /* When running in the kernel we expect faults to occur only to
468 * addresses in user space. All other faults represent errors in the
469 * kernel and should generate an OOPS. Unfortunately, in the case of an
470 * erroneous fault occurring in a code path which already holds mmap_sem
471 * we will deadlock attempting to validate the fault against the
472 * address space. Luckily the kernel only validly references user
473 * space from well defined areas of code, which are listed in the
476 * As the vast majority of faults will be valid we will only perform
477 * the source reference check when there is a possibility of a deadlock.
478 * Attempt to lock the address space, if we cannot we then validate the
479 * source. If this is invalid we can skip the address space check,
480 * thus avoiding the deadlock.
482 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
483 if (!is_user && !search_exception_tables(regs->nip))
484 return bad_area_nosemaphore(regs, address);
487 down_read(&mm->mmap_sem);
490 * The above down_read_trylock() might have succeeded in
491 * which case we'll have missed the might_sleep() from
497 vma = find_vma(mm, address);
499 return bad_area(regs, address);
500 if (likely(vma->vm_start <= address))
502 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
503 return bad_area(regs, address);
505 /* The stack is being expanded, check if it's valid */
506 if (unlikely(bad_stack_expansion(regs, address, vma, store_update_sp)))
507 return bad_area(regs, address);
509 /* Try to expand it */
510 if (unlikely(expand_stack(vma, address)))
511 return bad_area(regs, address);
514 if (unlikely(access_error(is_write, is_exec, vma)))
515 return bad_access(regs, address);
518 * If for any reason at all we couldn't handle the fault,
519 * make sure we exit gracefully rather than endlessly redo
522 fault = handle_mm_fault(vma, address, flags);
524 #ifdef CONFIG_PPC_MEM_KEYS
526 * we skipped checking for access error due to key earlier.
527 * Check that using handle_mm_fault error return.
529 if (unlikely(fault & VM_FAULT_SIGSEGV) &&
530 !arch_vma_access_permitted(vma, is_write, is_exec, 0)) {
532 int pkey = vma_pkey(vma);
534 up_read(&mm->mmap_sem);
535 return bad_key_fault_exception(regs, address, pkey);
537 #endif /* CONFIG_PPC_MEM_KEYS */
539 major |= fault & VM_FAULT_MAJOR;
542 * Handle the retry right now, the mmap_sem has been released in that
545 if (unlikely(fault & VM_FAULT_RETRY)) {
546 /* We retry only once */
547 if (flags & FAULT_FLAG_ALLOW_RETRY) {
549 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
552 flags &= ~FAULT_FLAG_ALLOW_RETRY;
553 flags |= FAULT_FLAG_TRIED;
554 if (!fatal_signal_pending(current))
559 * User mode? Just return to handle the fatal exception otherwise
560 * return to bad_page_fault
562 return is_user ? 0 : SIGBUS;
565 up_read(¤t->mm->mmap_sem);
567 if (unlikely(fault & VM_FAULT_ERROR))
568 return mm_fault_error(regs, address, fault);
571 * Major/minor page fault accounting.
575 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
576 cmo_account_page_fault();
579 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
583 NOKPROBE_SYMBOL(__do_page_fault);
585 int do_page_fault(struct pt_regs *regs, unsigned long address,
586 unsigned long error_code)
588 enum ctx_state prev_state = exception_enter();
589 int rc = __do_page_fault(regs, address, error_code);
590 exception_exit(prev_state);
593 NOKPROBE_SYMBOL(do_page_fault);
596 * bad_page_fault is called when we have a bad access from the kernel.
597 * It is called from the DSI and ISI handlers in head.S and from some
598 * of the procedures in traps.c.
600 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
602 const struct exception_table_entry *entry;
604 /* Are we prepared to handle this fault? */
605 if ((entry = search_exception_tables(regs->nip)) != NULL) {
606 regs->nip = extable_fixup(entry);
610 /* kernel has accessed a bad area */
612 switch (TRAP(regs)) {
615 printk(KERN_ALERT "Unable to handle kernel paging request for "
616 "data at address 0x%08lx\n", regs->dar);
620 printk(KERN_ALERT "Unable to handle kernel paging request for "
621 "instruction fetch\n");
624 printk(KERN_ALERT "Unable to handle kernel paging request for "
625 "unaligned access at address 0x%08lx\n", regs->dar);
628 printk(KERN_ALERT "Unable to handle kernel paging request for "
632 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
635 if (task_stack_end_corrupted(current))
636 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
638 die("Kernel access of bad area", regs, sig);