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/tlbflush.h>
46 #include <asm/siginfo.h>
47 #include <asm/debug.h>
49 static inline bool notify_page_fault(struct pt_regs *regs)
54 /* kprobe_running() needs smp_processor_id() */
55 if (!user_mode(regs)) {
57 if (kprobe_running() && kprobe_fault_handler(regs, 11))
61 #endif /* CONFIG_KPROBES */
63 if (unlikely(debugger_fault_handler(regs)))
70 * Check whether the instruction inst is a store using
71 * an update addressing form which will update r1.
73 static bool store_updates_sp(unsigned int inst)
75 /* check for 1 in the rA field */
76 if (((inst >> 16) & 0x1f) != 1)
78 /* check major opcode */
86 case 62: /* std or stdu */
87 return (inst & 3) == 1;
89 /* check minor opcode */
90 switch ((inst >> 1) & 0x3ff) {
95 case 695: /* stfsux */
96 case 759: /* stfdux */
103 * do_page_fault error handling helpers
107 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
110 * If we are in kernel mode, bail out with a SEGV, this will
111 * be caught by the assembly which will restore the non-volatile
112 * registers before calling bad_page_fault()
114 if (!user_mode(regs))
117 _exception(SIGSEGV, regs, si_code, address);
122 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
124 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
127 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
129 struct mm_struct *mm = current->mm;
132 * Something tried to access memory that isn't in our memory map..
133 * Fix it, but check if it's kernel or user first..
135 up_read(&mm->mmap_sem);
137 return __bad_area_nosemaphore(regs, address, si_code);
140 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
142 return __bad_area(regs, address, SEGV_MAPERR);
145 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
147 return __bad_area(regs, address, SEGV_ACCERR);
150 static int do_sigbus(struct pt_regs *regs, unsigned long address,
154 unsigned int lsb = 0;
156 if (!user_mode(regs))
159 current->thread.trap_nr = BUS_ADRERR;
160 info.si_signo = SIGBUS;
162 info.si_code = BUS_ADRERR;
163 info.si_addr = (void __user *)address;
164 #ifdef CONFIG_MEMORY_FAILURE
165 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
166 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
167 current->comm, current->pid, address);
168 info.si_code = BUS_MCEERR_AR;
171 if (fault & VM_FAULT_HWPOISON_LARGE)
172 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
173 if (fault & VM_FAULT_HWPOISON)
176 info.si_addr_lsb = lsb;
177 force_sig_info(SIGBUS, &info, current);
181 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
184 * Kernel page fault interrupted by SIGKILL. We have no reason to
185 * continue processing.
187 if (fatal_signal_pending(current) && !user_mode(regs))
191 if (fault & VM_FAULT_OOM) {
193 * We ran out of memory, or some other thing happened to us that
194 * made us unable to handle the page fault gracefully.
196 if (!user_mode(regs))
198 pagefault_out_of_memory();
200 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
201 VM_FAULT_HWPOISON_LARGE))
202 return do_sigbus(regs, addr, fault);
203 else if (fault & VM_FAULT_SIGSEGV)
204 return bad_area_nosemaphore(regs, addr);
211 /* Is this a bad kernel fault ? */
212 static bool bad_kernel_fault(bool is_exec, unsigned long error_code,
213 unsigned long address)
215 /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
216 if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
218 printk_ratelimited(KERN_CRIT "kernel tried to execute"
219 " exec-protected page (%lx) -"
220 "exploit attempt? (uid: %d)\n",
221 address, from_kuid(&init_user_ns,
224 return is_exec || (address >= TASK_SIZE);
227 // This comes from 64-bit struct rt_sigframe + __SIGNAL_FRAMESIZE
228 #define SIGFRAME_MAX_SIZE (4096 + 128)
230 static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
231 struct vm_area_struct *vma, unsigned int flags,
235 * N.B. The POWER/Open ABI allows programs to access up to
236 * 288 bytes below the stack pointer.
237 * The kernel signal delivery code writes a bit over 4KB
238 * below the stack pointer (r1) before decrementing it.
239 * The exec code can write slightly over 640kB to the stack
240 * before setting the user r1. Thus we allow the stack to
241 * expand to 1MB without further checks.
243 if (address + 0x100000 < vma->vm_end) {
244 unsigned int __user *nip = (unsigned int __user *)regs->nip;
245 /* get user regs even if this fault is in kernel mode */
246 struct pt_regs *uregs = current->thread.regs;
251 * A user-mode access to an address a long way below
252 * the stack pointer is only valid if the instruction
253 * is one which would update the stack pointer to the
254 * address accessed if the instruction completed,
255 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
256 * (or the byte, halfword, float or double forms).
258 * If we don't check this then any write to the area
259 * between the last mapped region and the stack will
260 * expand the stack rather than segfaulting.
262 if (address + SIGFRAME_MAX_SIZE >= uregs->gpr[1])
265 if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
266 access_ok(VERIFY_READ, nip, sizeof(*nip))) {
271 res = __get_user_inatomic(inst, nip);
274 return !store_updates_sp(inst);
282 static bool access_error(bool is_write, bool is_exec,
283 struct vm_area_struct *vma)
286 * Allow execution from readable areas if the MMU does not
287 * provide separate controls over reading and executing.
289 * Note: That code used to not be enabled for 4xx/BookE.
290 * It is now as I/D cache coherency for these is done at
291 * set_pte_at() time and I see no reason why the test
292 * below wouldn't be valid on those processors. This -may-
293 * break programs compiled with a really old ABI though.
296 return !(vma->vm_flags & VM_EXEC) &&
297 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
298 !(vma->vm_flags & (VM_READ | VM_WRITE)));
302 if (unlikely(!(vma->vm_flags & VM_WRITE)))
307 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
313 #ifdef CONFIG_PPC_SMLPAR
314 static inline void cmo_account_page_fault(void)
316 if (firmware_has_feature(FW_FEATURE_CMO)) {
320 page_ins = be32_to_cpu(get_lppaca()->page_ins);
321 page_ins += 1 << PAGE_FACTOR;
322 get_lppaca()->page_ins = cpu_to_be32(page_ins);
327 static inline void cmo_account_page_fault(void) { }
328 #endif /* CONFIG_PPC_SMLPAR */
330 #ifdef CONFIG_PPC_STD_MMU
331 static void sanity_check_fault(bool is_write, unsigned long error_code)
334 * For hash translation mode, we should never get a
335 * PROTFAULT. Any update to pte to reduce access will result in us
336 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
337 * fault instead of DSISR_PROTFAULT.
339 * A pte update to relax the access will not result in a hash page table
340 * entry invalidate and hence can result in DSISR_PROTFAULT.
341 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
342 * the special !is_write in the below conditional.
344 * For platforms that doesn't supports coherent icache and do support
345 * per page noexec bit, we do setup things such that we do the
346 * sync between D/I cache via fault. But that is handled via low level
347 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
350 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
351 * check should handle those and hence we should fall to the bad_area
352 * handling correctly.
354 * For embedded with per page exec support that doesn't support coherent
355 * icache we do get PROTFAULT and we handle that D/I cache sync in
356 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
357 * is conditional for server MMU.
359 * For radix, we can get prot fault for autonuma case, because radix
360 * page table will have them marked noaccess for user.
362 if (!radix_enabled() && !is_write)
363 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
366 static void sanity_check_fault(bool is_write, unsigned long error_code) { }
367 #endif /* CONFIG_PPC_STD_MMU */
370 * Define the correct "is_write" bit in error_code based
371 * on the processor family
373 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
374 #define page_fault_is_write(__err) ((__err) & ESR_DST)
375 #define page_fault_is_bad(__err) (0)
377 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE)
378 #if defined(CONFIG_PPC_8xx)
379 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G)
380 #elif defined(CONFIG_PPC64)
381 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S)
383 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S)
388 * For 600- and 800-family processors, the error_code parameter is DSISR
389 * for a data fault, SRR1 for an instruction fault. For 400-family processors
390 * the error_code parameter is ESR for a data fault, 0 for an instruction
392 * For 64-bit processors, the error_code parameter is
393 * - DSISR for a non-SLB data access fault,
394 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
397 * The return value is 0 if the fault was handled, or the signal
398 * number if this is a kernel fault that can't be handled here.
400 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
401 unsigned long error_code)
403 struct vm_area_struct * vma;
404 struct mm_struct *mm = current->mm;
405 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
406 int is_exec = TRAP(regs) == 0x400;
407 int is_user = user_mode(regs);
408 int is_write = page_fault_is_write(error_code);
409 int fault, major = 0;
410 bool must_retry = false;
412 if (notify_page_fault(regs))
415 if (unlikely(page_fault_is_bad(error_code))) {
417 _exception(SIGBUS, regs, BUS_OBJERR, address);
423 /* Additional sanity check(s) */
424 sanity_check_fault(is_write, error_code);
427 * The kernel should never take an execute fault nor should it
428 * take a page fault to a kernel address.
430 if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address)))
434 * If we're in an interrupt, have no user context or are running
435 * in a region with pagefaults disabled then we must not take the fault
437 if (unlikely(faulthandler_disabled() || !mm)) {
439 printk_ratelimited(KERN_ERR "Page fault in user mode"
440 " with faulthandler_disabled()=%d"
442 faulthandler_disabled(), mm);
443 return bad_area_nosemaphore(regs, address);
446 /* We restore the interrupt state now */
447 if (!arch_irq_disabled_regs(regs))
450 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, 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
458 flags |= FAULT_FLAG_USER;
460 flags |= FAULT_FLAG_WRITE;
462 flags |= FAULT_FLAG_INSTRUCTION;
464 /* When running in the kernel we expect faults to occur only to
465 * addresses in user space. All other faults represent errors in the
466 * kernel and should generate an OOPS. Unfortunately, in the case of an
467 * erroneous fault occurring in a code path which already holds mmap_sem
468 * we will deadlock attempting to validate the fault against the
469 * address space. Luckily the kernel only validly references user
470 * space from well defined areas of code, which are listed in the
473 * As the vast majority of faults will be valid we will only perform
474 * the source reference check when there is a possibility of a deadlock.
475 * Attempt to lock the address space, if we cannot we then validate the
476 * source. If this is invalid we can skip the address space check,
477 * thus avoiding the deadlock.
479 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
480 if (!is_user && !search_exception_tables(regs->nip))
481 return bad_area_nosemaphore(regs, address);
484 down_read(&mm->mmap_sem);
487 * The above down_read_trylock() might have succeeded in
488 * which case we'll have missed the might_sleep() from
494 vma = find_vma(mm, address);
496 return bad_area(regs, address);
497 if (likely(vma->vm_start <= address))
499 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
500 return bad_area(regs, address);
502 /* The stack is being expanded, check if it's valid */
503 if (unlikely(bad_stack_expansion(regs, address, vma, flags,
506 return bad_area(regs, address);
508 up_read(&mm->mmap_sem);
509 if (fault_in_pages_readable((const char __user *)regs->nip,
510 sizeof(unsigned int)))
511 return bad_area_nosemaphore(regs, address);
515 /* Try to expand it */
516 if (unlikely(expand_stack(vma, address)))
517 return bad_area(regs, address);
520 if (unlikely(access_error(is_write, is_exec, vma)))
521 return bad_access(regs, address);
524 * If for any reason at all we couldn't handle the fault,
525 * make sure we exit gracefully rather than endlessly redo
528 fault = handle_mm_fault(vma, address, flags);
529 major |= fault & VM_FAULT_MAJOR;
532 * Handle the retry right now, the mmap_sem has been released in that
535 if (unlikely(fault & VM_FAULT_RETRY)) {
536 /* We retry only once */
537 if (flags & FAULT_FLAG_ALLOW_RETRY) {
539 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
542 flags &= ~FAULT_FLAG_ALLOW_RETRY;
543 flags |= FAULT_FLAG_TRIED;
544 if (!fatal_signal_pending(current))
549 * User mode? Just return to handle the fatal exception otherwise
550 * return to bad_page_fault
552 return is_user ? 0 : SIGBUS;
555 up_read(¤t->mm->mmap_sem);
557 if (unlikely(fault & VM_FAULT_ERROR))
558 return mm_fault_error(regs, address, fault);
561 * Major/minor page fault accounting.
565 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
566 cmo_account_page_fault();
569 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
573 NOKPROBE_SYMBOL(__do_page_fault);
575 int do_page_fault(struct pt_regs *regs, unsigned long address,
576 unsigned long error_code)
578 enum ctx_state prev_state = exception_enter();
579 int rc = __do_page_fault(regs, address, error_code);
580 exception_exit(prev_state);
583 NOKPROBE_SYMBOL(do_page_fault);
586 * bad_page_fault is called when we have a bad access from the kernel.
587 * It is called from the DSI and ISI handlers in head.S and from some
588 * of the procedures in traps.c.
590 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
592 const struct exception_table_entry *entry;
594 /* Are we prepared to handle this fault? */
595 if ((entry = search_exception_tables(regs->nip)) != NULL) {
596 regs->nip = extable_fixup(entry);
600 /* kernel has accessed a bad area */
602 switch (regs->trap) {
605 pr_alert("BUG: %s at 0x%08lx\n",
606 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
607 "Unable to handle kernel data access", regs->dar);
611 pr_alert("BUG: Unable to handle kernel instruction fetch%s",
612 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
615 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
619 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
623 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
626 if (task_stack_end_corrupted(current))
627 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
629 die("Kernel access of bad area", regs, sig);